drm/amd/display: roll back quality EASF and ISHARP and dc dependency changes
authorSamson Tam <Samson.Tam@amd.com>
Sun, 14 Jul 2024 20:31:05 +0000 (16:31 -0400)
committerAlex Deucher <alexander.deucher@amd.com>
Sat, 27 Jul 2024 21:30:28 +0000 (17:30 -0400)
[Why]
Seeing several regressions related to quality EASF and ISHARP changes
and removing dc dependency changes.

[How]
Roll back SPL changes

Signed-off-by: Samson Tam <Samson.Tam@amd.com>
Reviewed-by: Martin Leung <martin.leung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Rodrigo Siqueira <rodrigo.siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
21 files changed:
drivers/gpu/drm/amd/display/dc/core/dc_resource.c
drivers/gpu/drm/amd/display/dc/dc_spl_translate.c
drivers/gpu/drm/amd/display/dc/dc_spl_translate.h
drivers/gpu/drm/amd/display/dc/dml2/dml21/dml21_translation_helper.c
drivers/gpu/drm/amd/display/dc/dpp/dcn401/dcn401_dpp_dscl.c
drivers/gpu/drm/amd/display/dc/resource/dcn401/dcn401_resource.c
drivers/gpu/drm/amd/display/dc/spl/Makefile
drivers/gpu/drm/amd/display/dc/spl/dc_spl.c
drivers/gpu/drm/amd/display/dc/spl/dc_spl_filters.c [deleted file]
drivers/gpu/drm/amd/display/dc/spl/dc_spl_filters.h [deleted file]
drivers/gpu/drm/amd/display/dc/spl/dc_spl_isharp_filters.c
drivers/gpu/drm/amd/display/dc/spl/dc_spl_isharp_filters.h
drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_easf_filters.c [deleted file]
drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_easf_filters.h [deleted file]
drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_filters.c
drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_filters.h
drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_filters_old.c [new file with mode: 0644]
drivers/gpu/drm/amd/display/dc/spl/dc_spl_types.h
drivers/gpu/drm/amd/display/dc/spl/spl_debug.h [deleted file]
drivers/gpu/drm/amd/display/dc/spl/spl_fixpt31_32.c [deleted file]
drivers/gpu/drm/amd/display/dc/spl/spl_fixpt31_32.h [deleted file]

index 5c9091f2a8b2d9ba480b3b03acc0d6b96b2ba23c..4f5b23520365db5bc34c5fefc5c56ea18c0d4860 100644 (file)
@@ -1511,6 +1511,8 @@ bool resource_build_scaling_params(struct pipe_ctx *pipe_ctx)
                        pipe_ctx->plane_res.scl_data.lb_params.depth = LB_PIXEL_DEPTH_30BPP;
 
                pipe_ctx->plane_res.scl_data.lb_params.alpha_en = plane_state->per_pixel_alpha;
+               spl_out->scl_data.h_active = pipe_ctx->plane_res.scl_data.h_active;
+               spl_out->scl_data.v_active = pipe_ctx->plane_res.scl_data.v_active;
 
                // Convert pipe_ctx to respective input params for SPL
                translate_SPL_in_params_from_pipe_ctx(pipe_ctx, spl_in);
index 339d092e711c3bfa7a9656e6ee2c4c2ef9efbec7..bcc596724a4f74360a3c19baddfd528bd866a46c 100644 (file)
@@ -42,26 +42,26 @@ static void populate_spltaps_from_taps(struct spl_taps *spl_scaling_quality,
 static void populate_taps_from_spltaps(struct scaling_taps *scaling_quality,
                const struct spl_taps *spl_scaling_quality)
 {
-       scaling_quality->h_taps_c = spl_scaling_quality->h_taps_c + 1;
-       scaling_quality->h_taps = spl_scaling_quality->h_taps + 1;
-       scaling_quality->v_taps_c = spl_scaling_quality->v_taps_c + 1;
-       scaling_quality->v_taps = spl_scaling_quality->v_taps + 1;
+       scaling_quality->h_taps_c = spl_scaling_quality->h_taps_c;
+       scaling_quality->h_taps = spl_scaling_quality->h_taps;
+       scaling_quality->v_taps_c = spl_scaling_quality->v_taps_c;
+       scaling_quality->v_taps = spl_scaling_quality->v_taps;
 }
 static void populate_ratios_from_splratios(struct scaling_ratios *ratios,
-               const struct ratio *spl_ratios)
+               const struct spl_ratios *spl_ratios)
 {
-       ratios->horz = dc_fixpt_from_ux_dy(spl_ratios->h_scale_ratio >> 5, 3, 19);
-       ratios->vert = dc_fixpt_from_ux_dy(spl_ratios->v_scale_ratio >> 5, 3, 19);
-       ratios->horz_c = dc_fixpt_from_ux_dy(spl_ratios->h_scale_ratio_c >> 5, 3, 19);
-       ratios->vert_c = dc_fixpt_from_ux_dy(spl_ratios->v_scale_ratio_c >> 5, 3, 19);
+       ratios->horz = spl_ratios->horz;
+       ratios->vert = spl_ratios->vert;
+       ratios->horz_c = spl_ratios->horz_c;
+       ratios->vert_c = spl_ratios->vert_c;
 }
 static void populate_inits_from_splinits(struct scl_inits *inits,
-               const struct init *spl_inits)
+               const struct spl_inits *spl_inits)
 {
-       inits->h = dc_fixpt_from_int_dy(spl_inits->h_filter_init_int, spl_inits->h_filter_init_frac >> 5, 0, 19);
-       inits->v = dc_fixpt_from_int_dy(spl_inits->v_filter_init_int, spl_inits->v_filter_init_frac >> 5, 0, 19);
-       inits->h_c = dc_fixpt_from_int_dy(spl_inits->h_filter_init_int_c, spl_inits->h_filter_init_frac_c >> 5, 0, 19);
-       inits->v_c = dc_fixpt_from_int_dy(spl_inits->v_filter_init_int_c, spl_inits->v_filter_init_frac_c >> 5, 0, 19);
+       inits->h = spl_inits->h;
+       inits->v = spl_inits->v;
+       inits->h_c = spl_inits->h_c;
+       inits->v_c = spl_inits->v_c;
 }
 /// @brief Translate SPL input parameters from pipe context
 /// @param pipe_ctx
@@ -170,15 +170,6 @@ void translate_SPL_in_params_from_pipe_ctx(struct pipe_ctx *pipe_ctx, struct spl
        /* Translate transfer function */
        spl_in->basic_in.tf_type = (enum spl_transfer_func_type) plane_state->in_transfer_func.type;
        spl_in->basic_in.tf_predefined_type = (enum spl_transfer_func_predefined) plane_state->in_transfer_func.tf;
-
-       spl_in->h_active = pipe_ctx->plane_res.scl_data.h_active;
-       spl_in->v_active = pipe_ctx->plane_res.scl_data.v_active;
-       /* Check if it is stream is in fullscreen and if its HDR.
-        * Use this to determine sharpness levels
-        */
-       spl_in->is_fullscreen = dm_helpers_is_fullscreen(pipe_ctx->stream->ctx, pipe_ctx->stream);
-       spl_in->is_hdr_on = dm_helpers_is_hdr_on(pipe_ctx->stream->ctx, pipe_ctx->stream);
-
 }
 
 /// @brief Translate SPL output parameters to pipe context
@@ -187,15 +178,15 @@ void translate_SPL_in_params_from_pipe_ctx(struct pipe_ctx *pipe_ctx, struct spl
 void translate_SPL_out_params_to_pipe_ctx(struct pipe_ctx *pipe_ctx, struct spl_out *spl_out)
 {
        // Make scaler data recout point to spl output field recout
-       populate_rect_from_splrect(&pipe_ctx->plane_res.scl_data.recout, &spl_out->dscl_prog_data->recout);
+       populate_rect_from_splrect(&pipe_ctx->plane_res.scl_data.recout, &spl_out->scl_data.recout);
        // Make scaler data ratios point to spl output field ratios
-       populate_ratios_from_splratios(&pipe_ctx->plane_res.scl_data.ratios, &spl_out->dscl_prog_data->ratios);
+       populate_ratios_from_splratios(&pipe_ctx->plane_res.scl_data.ratios, &spl_out->scl_data.ratios);
        // Make scaler data viewport point to spl output field viewport
-       populate_rect_from_splrect(&pipe_ctx->plane_res.scl_data.viewport, &spl_out->dscl_prog_data->viewport);
+       populate_rect_from_splrect(&pipe_ctx->plane_res.scl_data.viewport, &spl_out->scl_data.viewport);
        // Make scaler data viewport_c point to spl output field viewport_c
-       populate_rect_from_splrect(&pipe_ctx->plane_res.scl_data.viewport_c, &spl_out->dscl_prog_data->viewport_c);
+       populate_rect_from_splrect(&pipe_ctx->plane_res.scl_data.viewport_c, &spl_out->scl_data.viewport_c);
        // Make scaler data taps point to spl output field scaling taps
-       populate_taps_from_spltaps(&pipe_ctx->plane_res.scl_data.taps, &spl_out->dscl_prog_data->taps);
+       populate_taps_from_spltaps(&pipe_ctx->plane_res.scl_data.taps, &spl_out->scl_data.taps);
        // Make scaler data init point to spl output field init
-       populate_inits_from_splinits(&pipe_ctx->plane_res.scl_data.inits, &spl_out->dscl_prog_data->init);
+       populate_inits_from_splinits(&pipe_ctx->plane_res.scl_data.inits, &spl_out->scl_data.inits);
 }
index eaa5c5373b2849c8842ba0a389d7572b4e7e3ab8..c73d640c3632f9bfdbbad5fe16940cc7106a2f51 100644 (file)
@@ -6,7 +6,6 @@
 #define __DC_SPL_TRANSLATE_H__
 #include "dc.h"
 #include "resource.h"
-#include "dm_helpers.h"
 
 /* Map SPL input parameters to pipe context
  * @pipe_ctx: pipe context
index 4164cda90b2ab519f82a0f316f8eaf53bb57bac4..0f34688e40587465aa8b908246b67b98741e0af3 100644 (file)
@@ -788,14 +788,6 @@ static void populate_dml21_plane_config_from_plane_state(struct dml2_context *dm
                 * certain cases. Hence do corrective active and disable scaling.
                 */
                plane->composition.scaler_info.enabled = false;
-       } else if ((plane_state->ctx->dc->config.use_spl == true) &&
-               (plane->composition.scaler_info.enabled == false)) {
-               /* To enable sharpener for 1:1, scaler must be enabled.  If use_spl is set, then
-                *  allow case where ratio is 1 but taps > 1
-                */
-               if ((scaler_data->taps.h_taps > 1) || (scaler_data->taps.v_taps > 1) ||
-                       (scaler_data->taps.h_taps_c > 1) || (scaler_data->taps.v_taps_c > 1))
-                       plane->composition.scaler_info.enabled = true;
        }
 
        /* always_scale is only used for debug purposes not used in production but has to be
index 3a3745597f0c4aaaa8f7e276114dba569f61aa69..88d24e36fe00709ff0ca6129b23c736bd87dfb02 100644 (file)
@@ -280,8 +280,7 @@ static void dpp401_dscl_set_scaler_filter(
 static void dpp401_dscl_set_scl_filter(
                struct dcn401_dpp *dpp,
                const struct scaler_data *scl_data,
-               bool chroma_coef_mode,
-               bool force_coeffs_update)
+               bool chroma_coef_mode)
 {
        bool h_2tap_hardcode_coef_en = false;
        bool v_2tap_hardcode_coef_en = false;
@@ -344,7 +343,7 @@ static void dpp401_dscl_set_scl_filter(
                                                        || (filter_v_c && (filter_v_c != dpp->filter_v_c));
                }
 
-               if ((filter_updated) || (force_coeffs_update)) {
+               if (filter_updated) {
                        uint32_t scl_mode = REG_READ(SCL_MODE);
 
                        if (!h_2tap_hardcode_coef_en && filter_h) {
@@ -657,252 +656,274 @@ static void dpp401_dscl_set_recout(struct dcn401_dpp *dpp,
                  RECOUT_HEIGHT, recout->height);
 }
 /**
- * dpp401_dscl_program_easf_v - Program EASF_V
+ * dpp401_dscl_program_easf - Program EASF
  *
  * @dpp_base: High level DPP struct
  * @scl_data: scalaer_data info
  *
- * This is the primary function to program vertical EASF registers
+ * This is the primary function to program EASF
  *
  */
-static void dpp401_dscl_program_easf_v(struct dpp *dpp_base, const struct scaler_data *scl_data)
+static void dpp401_dscl_program_easf(struct dpp *dpp_base, const struct scaler_data *scl_data)
 {
        struct dcn401_dpp *dpp = TO_DCN401_DPP(dpp_base);
 
        PERF_TRACE();
+       REG_UPDATE(DSCL_SC_MODE,
+                       SCL_SC_MATRIX_MODE, scl_data->dscl_prog_data.easf_matrix_mode);
+       REG_UPDATE(DSCL_SC_MODE,
+                       SCL_SC_LTONL_EN, scl_data->dscl_prog_data.easf_ltonl_en);
        /* DSCL_EASF_V_MODE */
-       REG_SET_3(DSCL_EASF_V_MODE, 0,
-                       SCL_EASF_V_EN, scl_data->dscl_prog_data.easf_v_en,
-                       SCL_EASF_V_2TAP_SHARP_FACTOR, scl_data->dscl_prog_data.easf_v_sharp_factor,
+       REG_UPDATE(DSCL_EASF_V_MODE,
+                       SCL_EASF_V_EN, scl_data->dscl_prog_data.easf_v_en);
+       REG_UPDATE(DSCL_EASF_V_MODE,
+                       SCL_EASF_V_2TAP_SHARP_FACTOR, scl_data->dscl_prog_data.easf_v_sharp_factor);
+       REG_UPDATE(DSCL_EASF_V_MODE,
                        SCL_EASF_V_RINGEST_FORCE_EN, scl_data->dscl_prog_data.easf_v_ring);
-
-       if (!scl_data->dscl_prog_data.easf_v_en) {
-               PERF_TRACE();
-               return;
-       }
-
-       /* DSCL_EASF_V_BF_CNTL */
-       REG_SET_6(DSCL_EASF_V_BF_CNTL, 0,
-                       SCL_EASF_V_BF1_EN, scl_data->dscl_prog_data.easf_v_bf1_en,
-                       SCL_EASF_V_BF2_MODE, scl_data->dscl_prog_data.easf_v_bf2_mode,
-                       SCL_EASF_V_BF3_MODE, scl_data->dscl_prog_data.easf_v_bf3_mode,
-                       SCL_EASF_V_BF2_FLAT1_GAIN, scl_data->dscl_prog_data.easf_v_bf2_flat1_gain,
-                       SCL_EASF_V_BF2_FLAT2_GAIN, scl_data->dscl_prog_data.easf_v_bf2_flat2_gain,
+       REG_UPDATE(DSCL_EASF_V_BF_CNTL,
+                       SCL_EASF_V_BF1_EN, scl_data->dscl_prog_data.easf_v_bf1_en);
+       REG_UPDATE(DSCL_EASF_V_BF_CNTL,
+                       SCL_EASF_V_BF2_MODE, scl_data->dscl_prog_data.easf_v_bf2_mode);
+       REG_UPDATE(DSCL_EASF_V_BF_CNTL,
+                       SCL_EASF_V_BF3_MODE, scl_data->dscl_prog_data.easf_v_bf3_mode);
+       REG_UPDATE(DSCL_EASF_V_BF_CNTL,
+                       SCL_EASF_V_BF2_FLAT1_GAIN, scl_data->dscl_prog_data.easf_v_bf2_flat1_gain);
+       REG_UPDATE(DSCL_EASF_V_BF_CNTL,
+                       SCL_EASF_V_BF2_FLAT2_GAIN, scl_data->dscl_prog_data.easf_v_bf2_flat2_gain);
+       REG_UPDATE(DSCL_EASF_V_BF_CNTL,
                        SCL_EASF_V_BF2_ROC_GAIN, scl_data->dscl_prog_data.easf_v_bf2_roc_gain);
-       /* DSCL_EASF_V_RINGEST_3TAP_CNTLn */
-       REG_SET_2(DSCL_EASF_V_RINGEST_3TAP_CNTL1, 0,
-               SCL_EASF_V_RINGEST_3TAP_DNTILT_UPTILT, scl_data->dscl_prog_data.easf_v_ringest_3tap_dntilt_uptilt,
+       REG_UPDATE(DSCL_EASF_V_RINGEST_3TAP_CNTL1,
+               SCL_EASF_V_RINGEST_3TAP_DNTILT_UPTILT, scl_data->dscl_prog_data.easf_v_ringest_3tap_dntilt_uptilt);
+       REG_UPDATE(DSCL_EASF_V_RINGEST_3TAP_CNTL1,
                SCL_EASF_V_RINGEST_3TAP_UPTILT_MAXVAL, scl_data->dscl_prog_data.easf_v_ringest_3tap_uptilt_max);
-       REG_SET_2(DSCL_EASF_V_RINGEST_3TAP_CNTL2, 0,
-               SCL_EASF_V_RINGEST_3TAP_DNTILT_SLOPE, scl_data->dscl_prog_data.easf_v_ringest_3tap_dntilt_slope,
+       REG_UPDATE(DSCL_EASF_V_RINGEST_3TAP_CNTL2,
+               SCL_EASF_V_RINGEST_3TAP_DNTILT_SLOPE, scl_data->dscl_prog_data.easf_v_ringest_3tap_dntilt_slope);
+       REG_UPDATE(DSCL_EASF_V_RINGEST_3TAP_CNTL2,
                SCL_EASF_V_RINGEST_3TAP_UPTILT1_SLOPE, scl_data->dscl_prog_data.easf_v_ringest_3tap_uptilt1_slope);
-       REG_SET_2(DSCL_EASF_V_RINGEST_3TAP_CNTL3, 0,
-               SCL_EASF_V_RINGEST_3TAP_UPTILT2_SLOPE, scl_data->dscl_prog_data.easf_v_ringest_3tap_uptilt2_slope,
+       REG_UPDATE(DSCL_EASF_V_RINGEST_3TAP_CNTL3,
+               SCL_EASF_V_RINGEST_3TAP_UPTILT2_SLOPE, scl_data->dscl_prog_data.easf_v_ringest_3tap_uptilt2_slope);
+       REG_UPDATE(DSCL_EASF_V_RINGEST_3TAP_CNTL3,
                SCL_EASF_V_RINGEST_3TAP_UPTILT2_OFFSET, scl_data->dscl_prog_data.easf_v_ringest_3tap_uptilt2_offset);
-       /* DSCL_EASF_V_RINGEST_EVENTAP_REDUCE */
-       REG_SET_2(DSCL_EASF_V_RINGEST_EVENTAP_REDUCE, 0,
-               SCL_EASF_V_RINGEST_EVENTAP_REDUCEG1, scl_data->dscl_prog_data.easf_v_ringest_eventap_reduceg1,
+       REG_UPDATE(DSCL_EASF_V_RINGEST_EVENTAP_REDUCE,
+               SCL_EASF_V_RINGEST_EVENTAP_REDUCEG1, scl_data->dscl_prog_data.easf_v_ringest_eventap_reduceg1);
+       REG_UPDATE(DSCL_EASF_V_RINGEST_EVENTAP_REDUCE,
                SCL_EASF_V_RINGEST_EVENTAP_REDUCEG2, scl_data->dscl_prog_data.easf_v_ringest_eventap_reduceg2);
-       /* DSCL_EASF_V_RINGEST_EVENTAP_GAIN */
-       REG_SET_2(DSCL_EASF_V_RINGEST_EVENTAP_GAIN, 0,
-               SCL_EASF_V_RINGEST_EVENTAP_GAIN1, scl_data->dscl_prog_data.easf_v_ringest_eventap_gain1,
+       REG_UPDATE(DSCL_EASF_V_RINGEST_EVENTAP_GAIN,
+               SCL_EASF_V_RINGEST_EVENTAP_GAIN1, scl_data->dscl_prog_data.easf_v_ringest_eventap_gain1);
+       REG_UPDATE(DSCL_EASF_V_RINGEST_EVENTAP_GAIN,
                SCL_EASF_V_RINGEST_EVENTAP_GAIN2, scl_data->dscl_prog_data.easf_v_ringest_eventap_gain2);
-       /* DSCL_EASF_V_BF_FINAL_MAX_MIN */
-       REG_SET_4(DSCL_EASF_V_BF_FINAL_MAX_MIN, 0,
-                       SCL_EASF_V_BF_MAXA, scl_data->dscl_prog_data.easf_v_bf_maxa,
-                       SCL_EASF_V_BF_MAXB, scl_data->dscl_prog_data.easf_v_bf_maxb,
-                       SCL_EASF_V_BF_MINA, scl_data->dscl_prog_data.easf_v_bf_mina,
+       REG_UPDATE(DSCL_EASF_V_BF_FINAL_MAX_MIN,
+                       SCL_EASF_V_BF_MAXA, scl_data->dscl_prog_data.easf_v_bf_maxa);
+       REG_UPDATE(DSCL_EASF_V_BF_FINAL_MAX_MIN,
+                       SCL_EASF_V_BF_MAXB, scl_data->dscl_prog_data.easf_v_bf_maxb);
+       REG_UPDATE(DSCL_EASF_V_BF_FINAL_MAX_MIN,
+                       SCL_EASF_V_BF_MINA, scl_data->dscl_prog_data.easf_v_bf_mina);
+       REG_UPDATE(DSCL_EASF_V_BF_FINAL_MAX_MIN,
                        SCL_EASF_V_BF_MINB, scl_data->dscl_prog_data.easf_v_bf_minb);
-       /* DSCL_EASF_V_BF1_PWL_SEGn */
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG0, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg0,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg0,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG0,
+                       SCL_EASF_V_BF1_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg0);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG0,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg0);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG0,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG0, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg0);
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG1, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg1,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg1,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG1,
+                       SCL_EASF_V_BF1_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg1);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG1,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg1);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG1,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG1, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg1);
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG2, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg2,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg2,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG2,
+                       SCL_EASF_V_BF1_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg2);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG2,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg2);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG2,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG2, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg2);
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG3, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg3,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg3,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG3,
+                       SCL_EASF_V_BF1_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg3);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG3,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg3);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG3,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG3, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg3);
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG4, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg4,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg4,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG4,
+                       SCL_EASF_V_BF1_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg4);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG4,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg4);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG4,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG4, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg4);
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG5, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg5,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG5, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg5,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG5,
+                       SCL_EASF_V_BF1_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg5);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG5,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG5, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg5);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG5,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG5, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg5);
-       REG_SET_3(DSCL_EASF_V_BF1_PWL_SEG6, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG6, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg6,
-                       SCL_EASF_V_BF1_PWL_BASE_SEG6, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg6,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG6,
+                       SCL_EASF_V_BF1_PWL_IN_SEG6, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg6);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG6,
+                       SCL_EASF_V_BF1_PWL_BASE_SEG6, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg6);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG6,
                        SCL_EASF_V_BF1_PWL_SLOPE_SEG6, scl_data->dscl_prog_data.easf_v_bf1_pwl_slope_seg6);
-       REG_SET_2(DSCL_EASF_V_BF1_PWL_SEG7, 0,
-                       SCL_EASF_V_BF1_PWL_IN_SEG7, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg7,
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG7,
+                       SCL_EASF_V_BF1_PWL_IN_SEG7, scl_data->dscl_prog_data.easf_v_bf1_pwl_in_seg7);
+       REG_UPDATE(DSCL_EASF_V_BF1_PWL_SEG7,
                        SCL_EASF_V_BF1_PWL_BASE_SEG7, scl_data->dscl_prog_data.easf_v_bf1_pwl_base_seg7);
-       /* DSCL_EASF_V_BF3_PWL_SEGn */
-       REG_SET_3(DSCL_EASF_V_BF3_PWL_SEG0, 0,
-                       SCL_EASF_V_BF3_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set0,
-                       SCL_EASF_V_BF3_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set0,
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG0,
+                       SCL_EASF_V_BF3_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set0);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG0,
+                       SCL_EASF_V_BF3_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set0);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG0,
                        SCL_EASF_V_BF3_PWL_SLOPE_SEG0, scl_data->dscl_prog_data.easf_v_bf3_pwl_slope_set0);
-       REG_SET_3(DSCL_EASF_V_BF3_PWL_SEG1, 0,
-                       SCL_EASF_V_BF3_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set1,
-                       SCL_EASF_V_BF3_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set1,
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG1,
+                       SCL_EASF_V_BF3_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set1);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG1,
+                       SCL_EASF_V_BF3_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set1);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG1,
                        SCL_EASF_V_BF3_PWL_SLOPE_SEG1, scl_data->dscl_prog_data.easf_v_bf3_pwl_slope_set1);
-       REG_SET_3(DSCL_EASF_V_BF3_PWL_SEG2, 0,
-                       SCL_EASF_V_BF3_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set2,
-                       SCL_EASF_V_BF3_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set2,
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG2,
+                       SCL_EASF_V_BF3_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set2);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG2,
+                       SCL_EASF_V_BF3_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set2);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG2,
                        SCL_EASF_V_BF3_PWL_SLOPE_SEG2, scl_data->dscl_prog_data.easf_v_bf3_pwl_slope_set2);
-       REG_SET_3(DSCL_EASF_V_BF3_PWL_SEG3, 0,
-                       SCL_EASF_V_BF3_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set3,
-                       SCL_EASF_V_BF3_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set3,
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG3,
+                       SCL_EASF_V_BF3_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set3);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG3,
+                       SCL_EASF_V_BF3_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set3);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG3,
                        SCL_EASF_V_BF3_PWL_SLOPE_SEG3, scl_data->dscl_prog_data.easf_v_bf3_pwl_slope_set3);
-       REG_SET_3(DSCL_EASF_V_BF3_PWL_SEG4, 0,
-                       SCL_EASF_V_BF3_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set4,
-                       SCL_EASF_V_BF3_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set4,
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG4,
+                       SCL_EASF_V_BF3_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set4);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG4,
+                       SCL_EASF_V_BF3_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set4);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG4,
                        SCL_EASF_V_BF3_PWL_SLOPE_SEG4, scl_data->dscl_prog_data.easf_v_bf3_pwl_slope_set4);
-       REG_SET_2(DSCL_EASF_V_BF3_PWL_SEG5, 0,
-                       SCL_EASF_V_BF3_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set5,
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG5,
+                       SCL_EASF_V_BF3_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_v_bf3_pwl_in_set5);
+       REG_UPDATE(DSCL_EASF_V_BF3_PWL_SEG5,
                        SCL_EASF_V_BF3_PWL_BASE_SEG5, scl_data->dscl_prog_data.easf_v_bf3_pwl_base_set5);
-       PERF_TRACE();
-}
-/**
- * dpp401_dscl_program_easf_h - Program EASF_H
- *
- * @dpp_base: High level DPP struct
- * @scl_data: scalaer_data info
- *
- * This is the primary function to program horizontal EASF registers
- *
- */
-static void dpp401_dscl_program_easf_h(struct dpp *dpp_base, const struct scaler_data *scl_data)
-{
-       struct dcn401_dpp *dpp = TO_DCN401_DPP(dpp_base);
-
-       PERF_TRACE();
        /* DSCL_EASF_H_MODE */
-       REG_SET_3(DSCL_EASF_H_MODE, 0,
-                       SCL_EASF_H_EN, scl_data->dscl_prog_data.easf_h_en,
-                       SCL_EASF_H_2TAP_SHARP_FACTOR, scl_data->dscl_prog_data.easf_h_sharp_factor,
+       REG_UPDATE(DSCL_EASF_H_MODE,
+                       SCL_EASF_H_EN, scl_data->dscl_prog_data.easf_h_en);
+       REG_UPDATE(DSCL_EASF_H_MODE,
+                       SCL_EASF_H_2TAP_SHARP_FACTOR, scl_data->dscl_prog_data.easf_h_sharp_factor);
+       REG_UPDATE(DSCL_EASF_H_MODE,
                        SCL_EASF_H_RINGEST_FORCE_EN, scl_data->dscl_prog_data.easf_h_ring);
-
-       if (!scl_data->dscl_prog_data.easf_h_en) {
-               PERF_TRACE();
-               return;
-       }
-
-       /* DSCL_EASF_H_BF_CNTL */
-       REG_SET_6(DSCL_EASF_H_BF_CNTL, 0,
-                       SCL_EASF_H_BF1_EN, scl_data->dscl_prog_data.easf_h_bf1_en,
-                       SCL_EASF_H_BF2_MODE, scl_data->dscl_prog_data.easf_h_bf2_mode,
-                       SCL_EASF_H_BF3_MODE, scl_data->dscl_prog_data.easf_h_bf3_mode,
-                       SCL_EASF_H_BF2_FLAT1_GAIN, scl_data->dscl_prog_data.easf_h_bf2_flat1_gain,
-                       SCL_EASF_H_BF2_FLAT2_GAIN, scl_data->dscl_prog_data.easf_h_bf2_flat2_gain,
+       REG_UPDATE(DSCL_EASF_H_BF_CNTL,
+                       SCL_EASF_H_BF1_EN, scl_data->dscl_prog_data.easf_h_bf1_en);
+       REG_UPDATE(DSCL_EASF_H_BF_CNTL,
+                       SCL_EASF_H_BF2_MODE, scl_data->dscl_prog_data.easf_h_bf2_mode);
+       REG_UPDATE(DSCL_EASF_H_BF_CNTL,
+                       SCL_EASF_H_BF3_MODE, scl_data->dscl_prog_data.easf_h_bf3_mode);
+       REG_UPDATE(DSCL_EASF_H_BF_CNTL,
+                       SCL_EASF_H_BF2_FLAT1_GAIN, scl_data->dscl_prog_data.easf_h_bf2_flat1_gain);
+       REG_UPDATE(DSCL_EASF_H_BF_CNTL,
+                       SCL_EASF_H_BF2_FLAT2_GAIN, scl_data->dscl_prog_data.easf_h_bf2_flat2_gain);
+       REG_UPDATE(DSCL_EASF_H_BF_CNTL,
                        SCL_EASF_H_BF2_ROC_GAIN, scl_data->dscl_prog_data.easf_h_bf2_roc_gain);
-       /* DSCL_EASF_H_RINGEST_EVENTAP_REDUCE */
-       REG_SET_2(DSCL_EASF_H_RINGEST_EVENTAP_REDUCE, 0,
-                       SCL_EASF_H_RINGEST_EVENTAP_REDUCEG1, scl_data->dscl_prog_data.easf_h_ringest_eventap_reduceg1,
+       REG_UPDATE(DSCL_EASF_H_RINGEST_EVENTAP_REDUCE,
+                       SCL_EASF_H_RINGEST_EVENTAP_REDUCEG1, scl_data->dscl_prog_data.easf_h_ringest_eventap_reduceg1);
+       REG_UPDATE(DSCL_EASF_H_RINGEST_EVENTAP_REDUCE,
                        SCL_EASF_H_RINGEST_EVENTAP_REDUCEG2, scl_data->dscl_prog_data.easf_h_ringest_eventap_reduceg2);
-       /* DSCL_EASF_H_RINGEST_EVENTAP_GAIN */
-       REG_SET_2(DSCL_EASF_H_RINGEST_EVENTAP_GAIN, 0,
-                       SCL_EASF_H_RINGEST_EVENTAP_GAIN1, scl_data->dscl_prog_data.easf_h_ringest_eventap_gain1,
+       REG_UPDATE(DSCL_EASF_H_RINGEST_EVENTAP_GAIN,
+                       SCL_EASF_H_RINGEST_EVENTAP_GAIN1, scl_data->dscl_prog_data.easf_h_ringest_eventap_gain1);
+       REG_UPDATE(DSCL_EASF_H_RINGEST_EVENTAP_GAIN,
                        SCL_EASF_H_RINGEST_EVENTAP_GAIN2, scl_data->dscl_prog_data.easf_h_ringest_eventap_gain2);
-       /* DSCL_EASF_H_BF_FINAL_MAX_MIN */
-       REG_SET_4(DSCL_EASF_H_BF_FINAL_MAX_MIN, 0,
-                       SCL_EASF_H_BF_MAXA, scl_data->dscl_prog_data.easf_h_bf_maxa,
-                       SCL_EASF_H_BF_MAXB, scl_data->dscl_prog_data.easf_h_bf_maxb,
-                       SCL_EASF_H_BF_MINA, scl_data->dscl_prog_data.easf_h_bf_mina,
+       REG_UPDATE(DSCL_EASF_H_BF_FINAL_MAX_MIN,
+                       SCL_EASF_H_BF_MAXA, scl_data->dscl_prog_data.easf_h_bf_maxa);
+       REG_UPDATE(DSCL_EASF_H_BF_FINAL_MAX_MIN,
+                       SCL_EASF_H_BF_MAXB, scl_data->dscl_prog_data.easf_h_bf_maxb);
+       REG_UPDATE(DSCL_EASF_H_BF_FINAL_MAX_MIN,
+                       SCL_EASF_H_BF_MINA, scl_data->dscl_prog_data.easf_h_bf_mina);
+       REG_UPDATE(DSCL_EASF_H_BF_FINAL_MAX_MIN,
                        SCL_EASF_H_BF_MINB, scl_data->dscl_prog_data.easf_h_bf_minb);
-       /* DSCL_EASF_H_BF1_PWL_SEGn */
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG0, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg0,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg0,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG0,
+                       SCL_EASF_H_BF1_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg0);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG0,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg0);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG0,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG0, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg0);
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG1, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg1,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg1,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG1,
+                       SCL_EASF_H_BF1_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg1);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG1,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg1);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG1,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG1, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg1);
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG2, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg2,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg2,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG2,
+                       SCL_EASF_H_BF1_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg2);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG2,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg2);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG2,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG2, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg2);
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG3, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg3,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg3,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG3,
+                       SCL_EASF_H_BF1_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg3);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG3,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg3);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG3,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG3, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg3);
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG4, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg4,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg4,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG4,
+                       SCL_EASF_H_BF1_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg4);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG4,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg4);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG4,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG4, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg4);
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG5, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg5,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG5, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg5,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG5,
+                       SCL_EASF_H_BF1_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg5);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG5,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG5, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg5);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG5,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG5, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg5);
-       REG_SET_3(DSCL_EASF_H_BF1_PWL_SEG6, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG6, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg6,
-                       SCL_EASF_H_BF1_PWL_BASE_SEG6, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg6,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG6,
+                       SCL_EASF_H_BF1_PWL_IN_SEG6, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg6);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG6,
+                       SCL_EASF_H_BF1_PWL_BASE_SEG6, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg6);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG6,
                        SCL_EASF_H_BF1_PWL_SLOPE_SEG6, scl_data->dscl_prog_data.easf_h_bf1_pwl_slope_seg6);
-       REG_SET_2(DSCL_EASF_H_BF1_PWL_SEG7, 0,
-                       SCL_EASF_H_BF1_PWL_IN_SEG7, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg7,
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG7,
+                       SCL_EASF_H_BF1_PWL_IN_SEG7, scl_data->dscl_prog_data.easf_h_bf1_pwl_in_seg7);
+       REG_UPDATE(DSCL_EASF_H_BF1_PWL_SEG7,
                        SCL_EASF_H_BF1_PWL_BASE_SEG7, scl_data->dscl_prog_data.easf_h_bf1_pwl_base_seg7);
-       /* DSCL_EASF_H_BF3_PWL_SEGn */
-       REG_SET_3(DSCL_EASF_H_BF3_PWL_SEG0, 0,
-                       SCL_EASF_H_BF3_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set0,
-                       SCL_EASF_H_BF3_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set0,
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG0,
+                       SCL_EASF_H_BF3_PWL_IN_SEG0, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set0);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG0,
+                       SCL_EASF_H_BF3_PWL_BASE_SEG0, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set0);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG0,
                        SCL_EASF_H_BF3_PWL_SLOPE_SEG0, scl_data->dscl_prog_data.easf_h_bf3_pwl_slope_set0);
-       REG_SET_3(DSCL_EASF_H_BF3_PWL_SEG1, 0,
-                       SCL_EASF_H_BF3_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set1,
-                       SCL_EASF_H_BF3_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set1,
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG1,
+                       SCL_EASF_H_BF3_PWL_IN_SEG1, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set1);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG1,
+                       SCL_EASF_H_BF3_PWL_BASE_SEG1, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set1);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG1,
                        SCL_EASF_H_BF3_PWL_SLOPE_SEG1, scl_data->dscl_prog_data.easf_h_bf3_pwl_slope_set1);
-       REG_SET_3(DSCL_EASF_H_BF3_PWL_SEG2, 0,
-                       SCL_EASF_H_BF3_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set2,
-                       SCL_EASF_H_BF3_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set2,
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG2,
+                       SCL_EASF_H_BF3_PWL_IN_SEG2, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set2);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG2,
+                       SCL_EASF_H_BF3_PWL_BASE_SEG2, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set2);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG2,
                        SCL_EASF_H_BF3_PWL_SLOPE_SEG2, scl_data->dscl_prog_data.easf_h_bf3_pwl_slope_set2);
-       REG_SET_3(DSCL_EASF_H_BF3_PWL_SEG3, 0,
-                       SCL_EASF_H_BF3_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set3,
-                       SCL_EASF_H_BF3_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set3,
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG3,
+                       SCL_EASF_H_BF3_PWL_IN_SEG3, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set3);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG3,
+                       SCL_EASF_H_BF3_PWL_BASE_SEG3, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set3);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG3,
                        SCL_EASF_H_BF3_PWL_SLOPE_SEG3, scl_data->dscl_prog_data.easf_h_bf3_pwl_slope_set3);
-       REG_SET_3(DSCL_EASF_H_BF3_PWL_SEG4, 0,
-                       SCL_EASF_H_BF3_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set4,
-                       SCL_EASF_H_BF3_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set4,
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG4,
+                       SCL_EASF_H_BF3_PWL_IN_SEG4, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set4);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG4,
+                       SCL_EASF_H_BF3_PWL_BASE_SEG4, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set4);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG4,
                        SCL_EASF_H_BF3_PWL_SLOPE_SEG4, scl_data->dscl_prog_data.easf_h_bf3_pwl_slope_set4);
-       REG_SET_2(DSCL_EASF_H_BF3_PWL_SEG5, 0,
-                       SCL_EASF_H_BF3_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set5,
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG5,
+                       SCL_EASF_H_BF3_PWL_IN_SEG5, scl_data->dscl_prog_data.easf_h_bf3_pwl_in_set5);
+       REG_UPDATE(DSCL_EASF_H_BF3_PWL_SEG5,
                        SCL_EASF_H_BF3_PWL_BASE_SEG5, scl_data->dscl_prog_data.easf_h_bf3_pwl_base_set5);
-       PERF_TRACE();
-}
-/**
- * dpp401_dscl_program_easf - Program EASF
- *
- * @dpp_base: High level DPP struct
- * @scl_data: scalaer_data info
- *
- * This is the primary function to program EASF
- *
- */
-static void dpp401_dscl_program_easf(struct dpp *dpp_base, const struct scaler_data *scl_data)
-{
-       struct dcn401_dpp *dpp = TO_DCN401_DPP(dpp_base);
-
-       PERF_TRACE();
-       /* DSCL_SC_MODE */
-       REG_SET_2(DSCL_SC_MODE, 0,
-                       SCL_SC_MATRIX_MODE, scl_data->dscl_prog_data.easf_matrix_mode,
-                       SCL_SC_LTONL_EN, scl_data->dscl_prog_data.easf_ltonl_en);
        /* DSCL_EASF_SC_MATRIX_C0C1, DSCL_EASF_SC_MATRIX_C2C3 */
-       REG_SET_2(DSCL_SC_MATRIX_C0C1, 0,
-                       SCL_SC_MATRIX_C0, scl_data->dscl_prog_data.easf_matrix_c0,
+       REG_UPDATE(DSCL_SC_MATRIX_C0C1,
+                       SCL_SC_MATRIX_C0, scl_data->dscl_prog_data.easf_matrix_c0);
+       REG_UPDATE(DSCL_SC_MATRIX_C0C1,
                        SCL_SC_MATRIX_C1, scl_data->dscl_prog_data.easf_matrix_c1);
-       REG_SET_2(DSCL_SC_MATRIX_C2C3, 0,
-                       SCL_SC_MATRIX_C2, scl_data->dscl_prog_data.easf_matrix_c2,
+       REG_UPDATE(DSCL_SC_MATRIX_C2C3,
+                       SCL_SC_MATRIX_C2, scl_data->dscl_prog_data.easf_matrix_c2);
+       REG_UPDATE(DSCL_SC_MATRIX_C2C3,
                        SCL_SC_MATRIX_C3, scl_data->dscl_prog_data.easf_matrix_c3);
-       dpp401_dscl_program_easf_v(dpp_base, scl_data);
-       dpp401_dscl_program_easf_h(dpp_base, scl_data);
        PERF_TRACE();
 }
 /**
@@ -937,11 +958,10 @@ static void dpp401_dscl_set_isharp_filter(
 
        REG_UPDATE(ISHARP_DELTA_CTRL,
                ISHARP_DELTA_LUT_HOST_SELECT, 0);
-       /* LUT data write is auto-indexed.  Write index once */
-       REG_SET(ISHARP_DELTA_INDEX, 0,
-                       ISHARP_DELTA_INDEX, 0);
        for (level = 0; level < NUM_LEVELS; level++)    {
                filter_data = filter[level];
+               REG_SET(ISHARP_DELTA_INDEX, 0,
+                               ISHARP_DELTA_INDEX, level);
                REG_SET(ISHARP_DELTA_DATA, 0,
                                ISHARP_DELTA_DATA, filter_data);
        }
@@ -957,74 +977,107 @@ static void dpp401_dscl_set_isharp_filter(
  *
  */
 static void dpp401_dscl_program_isharp(struct dpp *dpp_base,
-               const struct scaler_data *scl_data,
-               bool *bs_coeffs_updated)
+               const struct scaler_data *scl_data)
 {
        struct dcn401_dpp *dpp = TO_DCN401_DPP(dpp_base);
-       *bs_coeffs_updated = false;
 
        PERF_TRACE();
-       /* ISHARP_MODE */
-       REG_SET_6(ISHARP_MODE, 0,
-               ISHARP_EN, scl_data->dscl_prog_data.isharp_en,
-               ISHARP_NOISEDET_EN, scl_data->dscl_prog_data.isharp_noise_det.enable,
-               ISHARP_NOISEDET_MODE, scl_data->dscl_prog_data.isharp_noise_det.mode,
-               ISHARP_LBA_MODE, scl_data->dscl_prog_data.isharp_lba.mode,
-               ISHARP_FMT_MODE, scl_data->dscl_prog_data.isharp_fmt.mode,
-               ISHARP_FMT_NORM, scl_data->dscl_prog_data.isharp_fmt.norm);
-
-       /* Skip remaining register programming if ISHARP is disabled */
-       if (!scl_data->dscl_prog_data.isharp_en) {
-               PERF_TRACE();
-               return;
-       }
-
-       /* ISHARP_NOISEDET_THRESHOLD */
-       REG_SET_2(ISHARP_NOISEDET_THRESHOLD, 0,
-               ISHARP_NOISEDET_UTHRE, scl_data->dscl_prog_data.isharp_noise_det.uthreshold,
+       /* ISHARP_EN */
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_EN, scl_data->dscl_prog_data.isharp_en);
+       /* ISHARP_NOISEDET_EN */
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_NOISEDET_EN, scl_data->dscl_prog_data.isharp_noise_det.enable);
+       /* ISHARP_NOISEDET_MODE */
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_NOISEDET_MODE, scl_data->dscl_prog_data.isharp_noise_det.mode);
+       /* ISHARP_NOISEDET_UTHRE */
+       REG_UPDATE(ISHARP_NOISEDET_THRESHOLD,
+               ISHARP_NOISEDET_UTHRE, scl_data->dscl_prog_data.isharp_noise_det.uthreshold);
+       /* ISHARP_NOISEDET_DTHRE */
+       REG_UPDATE(ISHARP_NOISEDET_THRESHOLD,
                ISHARP_NOISEDET_DTHRE, scl_data->dscl_prog_data.isharp_noise_det.dthreshold);
-
-       /* ISHARP_NOISE_GAIN_PWL */
-       REG_SET_3(ISHARP_NOISE_GAIN_PWL, 0,
-               ISHARP_NOISEDET_PWL_START_IN, scl_data->dscl_prog_data.isharp_noise_det.pwl_start_in,
-               ISHARP_NOISEDET_PWL_END_IN, scl_data->dscl_prog_data.isharp_noise_det.pwl_end_in,
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_NOISEDET_MODE, scl_data->dscl_prog_data.isharp_noise_det.mode);
+       /* ISHARP_NOISEDET_UTHRE */
+       REG_UPDATE(ISHARP_NOISEDET_THRESHOLD,
+               ISHARP_NOISEDET_UTHRE, scl_data->dscl_prog_data.isharp_noise_det.uthreshold);
+       /* ISHARP_NOISEDET_DTHRE */
+       REG_UPDATE(ISHARP_NOISEDET_THRESHOLD,
+               ISHARP_NOISEDET_DTHRE, scl_data->dscl_prog_data.isharp_noise_det.dthreshold);
+       /* ISHARP_NOISEDET_PWL_START_IN */
+       REG_UPDATE(ISHARP_NOISE_GAIN_PWL,
+               ISHARP_NOISEDET_PWL_START_IN, scl_data->dscl_prog_data.isharp_noise_det.pwl_start_in);
+       /* ISHARP_NOISEDET_PWL_END_IN */
+       REG_UPDATE(ISHARP_NOISE_GAIN_PWL,
+               ISHARP_NOISEDET_PWL_END_IN, scl_data->dscl_prog_data.isharp_noise_det.pwl_end_in);
+       /* ISHARP_NOISEDET_PWL_SLOPE */
+       REG_UPDATE(ISHARP_NOISE_GAIN_PWL,
                ISHARP_NOISEDET_PWL_SLOPE, scl_data->dscl_prog_data.isharp_noise_det.pwl_slope);
-
+       /* ISHARP_LBA_MODE */
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_LBA_MODE, scl_data->dscl_prog_data.isharp_lba.mode);
        /* ISHARP_LBA: IN_SEG, BASE_SEG, SLOPE_SEG */
-       REG_SET_3(ISHARP_LBA_PWL_SEG0, 0,
-               ISHARP_LBA_PWL_IN_SEG0, scl_data->dscl_prog_data.isharp_lba.in_seg[0],
-               ISHARP_LBA_PWL_BASE_SEG0, scl_data->dscl_prog_data.isharp_lba.base_seg[0],
+       REG_UPDATE(ISHARP_LBA_PWL_SEG0,
+               ISHARP_LBA_PWL_IN_SEG0, scl_data->dscl_prog_data.isharp_lba.in_seg[0]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG0,
+               ISHARP_LBA_PWL_BASE_SEG0, scl_data->dscl_prog_data.isharp_lba.base_seg[0]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG0,
                ISHARP_LBA_PWL_SLOPE_SEG0, scl_data->dscl_prog_data.isharp_lba.slope_seg[0]);
-       REG_SET_3(ISHARP_LBA_PWL_SEG1, 0,
-               ISHARP_LBA_PWL_IN_SEG1, scl_data->dscl_prog_data.isharp_lba.in_seg[1],
-               ISHARP_LBA_PWL_BASE_SEG1, scl_data->dscl_prog_data.isharp_lba.base_seg[1],
+       REG_UPDATE(ISHARP_LBA_PWL_SEG1,
+               ISHARP_LBA_PWL_IN_SEG1, scl_data->dscl_prog_data.isharp_lba.in_seg[1]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG1,
+               ISHARP_LBA_PWL_BASE_SEG1, scl_data->dscl_prog_data.isharp_lba.base_seg[1]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG1,
                ISHARP_LBA_PWL_SLOPE_SEG1, scl_data->dscl_prog_data.isharp_lba.slope_seg[1]);
-       REG_SET_3(ISHARP_LBA_PWL_SEG2, 0,
-               ISHARP_LBA_PWL_IN_SEG2, scl_data->dscl_prog_data.isharp_lba.in_seg[2],
-               ISHARP_LBA_PWL_BASE_SEG2, scl_data->dscl_prog_data.isharp_lba.base_seg[2],
+       REG_UPDATE(ISHARP_LBA_PWL_SEG2,
+               ISHARP_LBA_PWL_IN_SEG2, scl_data->dscl_prog_data.isharp_lba.in_seg[2]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG2,
+               ISHARP_LBA_PWL_BASE_SEG2, scl_data->dscl_prog_data.isharp_lba.base_seg[2]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG2,
                ISHARP_LBA_PWL_SLOPE_SEG2, scl_data->dscl_prog_data.isharp_lba.slope_seg[2]);
-       REG_SET_3(ISHARP_LBA_PWL_SEG3, 0,
-               ISHARP_LBA_PWL_IN_SEG3, scl_data->dscl_prog_data.isharp_lba.in_seg[3],
-               ISHARP_LBA_PWL_BASE_SEG3, scl_data->dscl_prog_data.isharp_lba.base_seg[3],
+       REG_UPDATE(ISHARP_LBA_PWL_SEG3,
+               ISHARP_LBA_PWL_IN_SEG3, scl_data->dscl_prog_data.isharp_lba.in_seg[3]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG3,
+               ISHARP_LBA_PWL_BASE_SEG3, scl_data->dscl_prog_data.isharp_lba.base_seg[3]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG3,
                ISHARP_LBA_PWL_SLOPE_SEG3, scl_data->dscl_prog_data.isharp_lba.slope_seg[3]);
-       REG_SET_3(ISHARP_LBA_PWL_SEG4, 0,
-               ISHARP_LBA_PWL_IN_SEG4, scl_data->dscl_prog_data.isharp_lba.in_seg[4],
-               ISHARP_LBA_PWL_BASE_SEG4, scl_data->dscl_prog_data.isharp_lba.base_seg[4],
+       REG_UPDATE(ISHARP_LBA_PWL_SEG4,
+               ISHARP_LBA_PWL_IN_SEG4, scl_data->dscl_prog_data.isharp_lba.in_seg[4]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG4,
+               ISHARP_LBA_PWL_BASE_SEG4, scl_data->dscl_prog_data.isharp_lba.base_seg[4]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG4,
                ISHARP_LBA_PWL_SLOPE_SEG4, scl_data->dscl_prog_data.isharp_lba.slope_seg[4]);
-       REG_SET_2(ISHARP_LBA_PWL_SEG5, 0,
-               ISHARP_LBA_PWL_IN_SEG5, scl_data->dscl_prog_data.isharp_lba.in_seg[5],
+       REG_UPDATE(ISHARP_LBA_PWL_SEG5,
+               ISHARP_LBA_PWL_IN_SEG5, scl_data->dscl_prog_data.isharp_lba.in_seg[5]);
+       REG_UPDATE(ISHARP_LBA_PWL_SEG5,
                ISHARP_LBA_PWL_BASE_SEG5, scl_data->dscl_prog_data.isharp_lba.base_seg[5]);
 
+       /* ISHARP_FMT_MODE */
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_FMT_MODE, scl_data->dscl_prog_data.isharp_fmt.mode);
+       /* ISHARP_FMT_NORM */
+       REG_UPDATE(ISHARP_MODE,
+               ISHARP_FMT_NORM, scl_data->dscl_prog_data.isharp_fmt.norm);
        /* ISHARP_DELTA_LUT */
        dpp401_dscl_set_isharp_filter(dpp, scl_data->dscl_prog_data.isharp_delta);
-
-       /* ISHARP_NLDELTA_SOFT_CLIP */
-       REG_SET_6(ISHARP_NLDELTA_SOFT_CLIP, 0,
-               ISHARP_NLDELTA_SCLIP_EN_P, scl_data->dscl_prog_data.isharp_nldelta_sclip.enable_p,
-               ISHARP_NLDELTA_SCLIP_PIVOT_P, scl_data->dscl_prog_data.isharp_nldelta_sclip.pivot_p,
-               ISHARP_NLDELTA_SCLIP_SLOPE_P, scl_data->dscl_prog_data.isharp_nldelta_sclip.slope_p,
-               ISHARP_NLDELTA_SCLIP_EN_N, scl_data->dscl_prog_data.isharp_nldelta_sclip.enable_n,
-               ISHARP_NLDELTA_SCLIP_PIVOT_N, scl_data->dscl_prog_data.isharp_nldelta_sclip.pivot_n,
+       /* ISHARP_NLDELTA_SCLIP_EN_P */
+       REG_UPDATE(ISHARP_NLDELTA_SOFT_CLIP,
+               ISHARP_NLDELTA_SCLIP_EN_P, scl_data->dscl_prog_data.isharp_nldelta_sclip.enable_p);
+       /* ISHARP_NLDELTA_SCLIP_PIVOT_P */
+       REG_UPDATE(ISHARP_NLDELTA_SOFT_CLIP,
+               ISHARP_NLDELTA_SCLIP_PIVOT_P, scl_data->dscl_prog_data.isharp_nldelta_sclip.pivot_p);
+       /* ISHARP_NLDELTA_SCLIP_SLOPE_P */
+       REG_UPDATE(ISHARP_NLDELTA_SOFT_CLIP,
+               ISHARP_NLDELTA_SCLIP_SLOPE_P, scl_data->dscl_prog_data.isharp_nldelta_sclip.slope_p);
+       /* ISHARP_NLDELTA_SCLIP_EN_N */
+       REG_UPDATE(ISHARP_NLDELTA_SOFT_CLIP,
+               ISHARP_NLDELTA_SCLIP_EN_N, scl_data->dscl_prog_data.isharp_nldelta_sclip.enable_n);
+       /* ISHARP_NLDELTA_SCLIP_PIVOT_N */
+       REG_UPDATE(ISHARP_NLDELTA_SOFT_CLIP,
+               ISHARP_NLDELTA_SCLIP_PIVOT_N, scl_data->dscl_prog_data.isharp_nldelta_sclip.pivot_n);
+       /* ISHARP_NLDELTA_SCLIP_SLOPE_N */
+       REG_UPDATE(ISHARP_NLDELTA_SOFT_CLIP,
                ISHARP_NLDELTA_SCLIP_SLOPE_N, scl_data->dscl_prog_data.isharp_nldelta_sclip.slope_n);
 
        /* Blur and Scale Coefficients - SCL_COEF_RAM_TAP_SELECT */
@@ -1034,14 +1087,12 @@ static void dpp401_dscl_program_isharp(struct dpp *dpp_base,
                                dpp, scl_data->taps.v_taps,
                                SCL_COEF_VERTICAL_BLUR_SCALE,
                                scl_data->dscl_prog_data.filter_blur_scale_v);
-                       *bs_coeffs_updated = true;
                }
                if (scl_data->dscl_prog_data.filter_blur_scale_h) {
                        dpp401_dscl_set_scaler_filter(
                                dpp, scl_data->taps.h_taps,
                                SCL_COEF_HORIZONTAL_BLUR_SCALE,
                                scl_data->dscl_prog_data.filter_blur_scale_h);
-                       *bs_coeffs_updated = true;
                }
        }
        PERF_TRACE();
@@ -1072,7 +1123,6 @@ void dpp401_dscl_set_scaler_manual_scale(struct dpp *dpp_base,
                        dpp_base, scl_data, dpp_base->ctx->dc->debug.always_scale);
        bool ycbcr = scl_data->format >= PIXEL_FORMAT_VIDEO_BEGIN
                                && scl_data->format <= PIXEL_FORMAT_VIDEO_END;
-       bool bs_coeffs_updated = false;
 
        if (memcmp(&dpp->scl_data, scl_data, sizeof(*scl_data)) == 0)
                return;
@@ -1132,7 +1182,7 @@ void dpp401_dscl_set_scaler_manual_scale(struct dpp *dpp_base,
        if (dscl_mode == DSCL_MODE_SCALING_444_BYPASS) {
                if (dpp->base.ctx->dc->config.prefer_easf)
                        dpp401_dscl_disable_easf(dpp_base, scl_data);
-               dpp401_dscl_program_isharp(dpp_base, scl_data, &bs_coeffs_updated);
+               dpp401_dscl_program_isharp(dpp_base, scl_data);
                return;
        }
 
@@ -1159,18 +1209,12 @@ void dpp401_dscl_set_scaler_manual_scale(struct dpp *dpp_base,
                SCL_V_NUM_TAPS_C, v_num_taps_c,
                SCL_H_NUM_TAPS_C, h_num_taps_c);
 
-       /* ISharp configuration
-        * - B&S coeffs are written to same coeff RAM as WB scaler coeffs
-        * - coeff RAM toggle is in EASF programming
-        * - if we are only programming B&S coeffs, then need to reprogram
-        *   WB scaler coeffs and toggle coeff RAM together
-        */
-       //if (dpp->base.ctx->dc->config.prefer_easf)
-       dpp401_dscl_program_isharp(dpp_base, scl_data, &bs_coeffs_updated);
-
-       dpp401_dscl_set_scl_filter(dpp, scl_data, ycbcr, bs_coeffs_updated);
+       dpp401_dscl_set_scl_filter(dpp, scl_data, ycbcr);
        /* Edge adaptive scaler function configuration */
        if (dpp->base.ctx->dc->config.prefer_easf)
                dpp401_dscl_program_easf(dpp_base, scl_data);
+       /* isharp configuration */
+       //if (dpp->base.ctx->dc->config.prefer_easf)
+       dpp401_dscl_program_isharp(dpp_base, scl_data);
        PERF_TRACE();
 }
index 5ee20753572e07ad44c1c0181964ef9c9753f469..9fcdf06d6aa4cd92ff44c2b285a7b6fcbe56ec63 100644 (file)
@@ -76,9 +76,6 @@
 
 #include "dml2/dml2_wrapper.h"
 
-#include "spl/dc_spl_scl_easf_filters.h"
-#include "spl/dc_spl_isharp_filters.h"
-
 #define DC_LOGGER_INIT(logger)
 
 enum dcn401_clk_src_array_id {
@@ -2126,10 +2123,6 @@ static bool dcn401_resource_construct(
        dc->dml2_options.max_segments_per_hubp = 20;
        dc->dml2_options.det_segment_size = DCN4_01_CRB_SEGMENT_SIZE_KB;
 
-       /* SPL */
-       spl_init_easf_filter_coeffs();
-       spl_init_blur_scale_coeffs();
-
        return true;
 
 create_fail:
index 05764d4d4604bd4355dc208383112b83b09cf04e..f8df85ea4d327329822a72cac3ccac42d530c0b2 100644 (file)
@@ -23,7 +23,7 @@
 # Makefile for the 'spl' sub-component of DAL.
 # It provides the scaling library interface.
 
-SPL = dc_spl.o dc_spl_scl_filters.o dc_spl_scl_easf_filters.o dc_spl_isharp_filters.o dc_spl_filters.o spl_fixpt31_32.o
+SPL = dc_spl.o dc_spl_scl_filters.o dc_spl_scl_filters_old.o dc_spl_isharp_filters.o
 
 AMD_DAL_SPL = $(addprefix $(AMDDALPATH)/dc/spl/,$(SPL))
 
index b8858ea7c776d47dec93475759393531e3eb626a..9eccdb38bed4882c213a388bf3799ad764df536e 100644 (file)
@@ -4,11 +4,9 @@
 
 #include "dc_spl.h"
 #include "dc_spl_scl_filters.h"
-#include "dc_spl_scl_easf_filters.h"
 #include "dc_spl_isharp_filters.h"
-#include "spl_debug.h"
 
-#define IDENTITY_RATIO(ratio) (spl_fixpt_u2d19(ratio) == (1 << 19))
+#define IDENTITY_RATIO(ratio) (dc_fixpt_u2d19(ratio) == (1 << 19))
 #define MIN_VIEWPORT_SIZE 12
 
 static struct spl_rect intersect_rec(const struct spl_rect *r0, const struct spl_rect *r1)
@@ -109,26 +107,26 @@ static struct spl_rect calculate_plane_rec_in_timing_active(
        const struct spl_rect *stream_src = &spl_in->basic_out.src_rect;
        const struct spl_rect *stream_dst = &spl_in->basic_out.dst_rect;
        struct spl_rect rec_out = {0};
-       struct spl_fixed31_32 temp;
+       struct fixed31_32 temp;
 
 
-       temp = spl_fixpt_from_fraction(rec_in->x * (long long)stream_dst->width,
+       temp = dc_fixpt_from_fraction(rec_in->x * (long long)stream_dst->width,
                        stream_src->width);
-       rec_out.x = stream_dst->x + spl_fixpt_round(temp);
+       rec_out.x = stream_dst->x + dc_fixpt_round(temp);
 
-       temp = spl_fixpt_from_fraction(
+       temp = dc_fixpt_from_fraction(
                        (rec_in->x + rec_in->width) * (long long)stream_dst->width,
                        stream_src->width);
-       rec_out.width = stream_dst->x + spl_fixpt_round(temp) - rec_out.x;
+       rec_out.width = stream_dst->x + dc_fixpt_round(temp) - rec_out.x;
 
-       temp = spl_fixpt_from_fraction(rec_in->y * (long long)stream_dst->height,
+       temp = dc_fixpt_from_fraction(rec_in->y * (long long)stream_dst->height,
                        stream_src->height);
-       rec_out.y = stream_dst->y + spl_fixpt_round(temp);
+       rec_out.y = stream_dst->y + dc_fixpt_round(temp);
 
-       temp = spl_fixpt_from_fraction(
+       temp = dc_fixpt_from_fraction(
                        (rec_in->y + rec_in->height) * (long long)stream_dst->height,
                        stream_src->height);
-       rec_out.height = stream_dst->y + spl_fixpt_round(temp) - rec_out.y;
+       rec_out.height = stream_dst->y + dc_fixpt_round(temp) - rec_out.y;
 
        return rec_out;
 }
@@ -146,7 +144,7 @@ static struct spl_rect calculate_mpc_slice_in_timing_active(
        mpc_rec.x = plane_clip_rec->x + mpc_rec.width * mpc_slice_idx;
        mpc_rec.height = plane_clip_rec->height;
        mpc_rec.y = plane_clip_rec->y;
-       SPL_ASSERT(mpc_slice_count == 1 ||
+       ASSERT(mpc_slice_count == 1 ||
                        spl_in->basic_out.view_format != SPL_VIEW_3D_SIDE_BY_SIDE ||
                        mpc_rec.width % 2 == 0);
 
@@ -159,7 +157,7 @@ static struct spl_rect calculate_mpc_slice_in_timing_active(
        }
 
        if (spl_in->basic_out.view_format == SPL_VIEW_3D_TOP_AND_BOTTOM) {
-               SPL_ASSERT(mpc_rec.height % 2 == 0);
+               ASSERT(mpc_rec.height % 2 == 0);
                mpc_rec.height /= 2;
        }
        return mpc_rec;
@@ -199,7 +197,7 @@ static struct spl_rect calculate_odm_slice_in_timing_active(struct spl_in *spl_i
        return spl_in->basic_out.odm_slice_rect;
 }
 
-static void spl_calculate_recout(struct spl_in *spl_in, struct spl_scratch *spl_scratch, struct spl_out *spl_out)
+static void spl_calculate_recout(struct spl_in *spl_in, struct spl_out *spl_out)
 {
        /*
         * A plane clip represents the desired plane size and position in Stream
@@ -342,23 +340,20 @@ static void spl_calculate_recout(struct spl_in *spl_in, struct spl_scratch *spl_
                /* shift the overlapping area so it is with respect to current
                 * ODM slice's position
                 */
-               spl_scratch->scl_data.recout = shift_rec(
+               spl_out->scl_data.recout = shift_rec(
                                &overlapping_area,
                                -odm_slice.x, -odm_slice.y);
-               spl_scratch->scl_data.recout.height -=
+               spl_out->scl_data.recout.height -=
                        spl_in->debug.visual_confirm_base_offset;
-               spl_scratch->scl_data.recout.height -=
+               spl_out->scl_data.recout.height -=
                        spl_in->debug.visual_confirm_dpp_offset;
        } else
                /* if there is no overlap, zero recout */
-               memset(&spl_scratch->scl_data.recout, 0,
+               memset(&spl_out->scl_data.recout, 0,
                                sizeof(struct spl_rect));
 }
-
 /* Calculate scaling ratios */
-static void spl_calculate_scaling_ratios(struct spl_in *spl_in,
-               struct spl_scratch *spl_scratch,
-               struct spl_out *spl_out)
+static void spl_calculate_scaling_ratios(struct spl_in *spl_in, struct spl_out *spl_out)
 {
        const int in_w = spl_in->basic_out.src_rect.width;
        const int in_h = spl_in->basic_out.src_rect.height;
@@ -369,75 +364,59 @@ static void spl_calculate_scaling_ratios(struct spl_in *spl_in,
        /*Swap surf_src height and width since scaling ratios are in recout rotation*/
        if (spl_in->basic_in.rotation == SPL_ROTATION_ANGLE_90 ||
                spl_in->basic_in.rotation == SPL_ROTATION_ANGLE_270)
-               spl_swap(surf_src.height, surf_src.width);
+               swap(surf_src.height, surf_src.width);
 
-       spl_scratch->scl_data.ratios.horz = spl_fixpt_from_fraction(
+       spl_out->scl_data.ratios.horz = dc_fixpt_from_fraction(
                                        surf_src.width,
                                        spl_in->basic_in.dst_rect.width);
-       spl_scratch->scl_data.ratios.vert = spl_fixpt_from_fraction(
+       spl_out->scl_data.ratios.vert = dc_fixpt_from_fraction(
                                        surf_src.height,
                                        spl_in->basic_in.dst_rect.height);
 
        if (spl_in->basic_out.view_format == SPL_VIEW_3D_SIDE_BY_SIDE)
-               spl_scratch->scl_data.ratios.horz.value *= 2;
+               spl_out->scl_data.ratios.horz.value *= 2;
        else if (spl_in->basic_out.view_format == SPL_VIEW_3D_TOP_AND_BOTTOM)
-               spl_scratch->scl_data.ratios.vert.value *= 2;
+               spl_out->scl_data.ratios.vert.value *= 2;
 
-       spl_scratch->scl_data.ratios.vert.value = spl_div64_s64(
-               spl_scratch->scl_data.ratios.vert.value * in_h, out_h);
-       spl_scratch->scl_data.ratios.horz.value = spl_div64_s64(
-               spl_scratch->scl_data.ratios.horz.value * in_w, out_w);
+       spl_out->scl_data.ratios.vert.value = div64_s64(
+               spl_out->scl_data.ratios.vert.value * in_h, out_h);
+       spl_out->scl_data.ratios.horz.value = div64_s64(
+               spl_out->scl_data.ratios.horz.value * in_w, out_w);
 
-       spl_scratch->scl_data.ratios.horz_c = spl_scratch->scl_data.ratios.horz;
-       spl_scratch->scl_data.ratios.vert_c = spl_scratch->scl_data.ratios.vert;
+       spl_out->scl_data.ratios.horz_c = spl_out->scl_data.ratios.horz;
+       spl_out->scl_data.ratios.vert_c = spl_out->scl_data.ratios.vert;
 
        if (spl_in->basic_in.format == SPL_PIXEL_FORMAT_420BPP8
                        || spl_in->basic_in.format == SPL_PIXEL_FORMAT_420BPP10) {
-               spl_scratch->scl_data.ratios.horz_c.value /= 2;
-               spl_scratch->scl_data.ratios.vert_c.value /= 2;
+               spl_out->scl_data.ratios.horz_c.value /= 2;
+               spl_out->scl_data.ratios.vert_c.value /= 2;
        }
-       spl_scratch->scl_data.ratios.horz = spl_fixpt_truncate(
-                       spl_scratch->scl_data.ratios.horz, 19);
-       spl_scratch->scl_data.ratios.vert = spl_fixpt_truncate(
-                       spl_scratch->scl_data.ratios.vert, 19);
-       spl_scratch->scl_data.ratios.horz_c = spl_fixpt_truncate(
-                       spl_scratch->scl_data.ratios.horz_c, 19);
-       spl_scratch->scl_data.ratios.vert_c = spl_fixpt_truncate(
-                       spl_scratch->scl_data.ratios.vert_c, 19);
-
-       /*
-        * Coefficient table and some registers are different based on ratio
-        * that is output/input.  Currently we calculate input/output
-        * Store 1/ratio in recip_ratio for those lookups
-        */
-       spl_scratch->scl_data.recip_ratios.horz = spl_fixpt_recip(
-                       spl_scratch->scl_data.ratios.horz);
-       spl_scratch->scl_data.recip_ratios.vert = spl_fixpt_recip(
-                       spl_scratch->scl_data.ratios.vert);
-       spl_scratch->scl_data.recip_ratios.horz_c = spl_fixpt_recip(
-                       spl_scratch->scl_data.ratios.horz_c);
-       spl_scratch->scl_data.recip_ratios.vert_c = spl_fixpt_recip(
-                       spl_scratch->scl_data.ratios.vert_c);
+       spl_out->scl_data.ratios.horz = dc_fixpt_truncate(
+                       spl_out->scl_data.ratios.horz, 19);
+       spl_out->scl_data.ratios.vert = dc_fixpt_truncate(
+                       spl_out->scl_data.ratios.vert, 19);
+       spl_out->scl_data.ratios.horz_c = dc_fixpt_truncate(
+                       spl_out->scl_data.ratios.horz_c, 19);
+       spl_out->scl_data.ratios.vert_c = dc_fixpt_truncate(
+                       spl_out->scl_data.ratios.vert_c, 19);
 }
-
 /* Calculate Viewport size */
-static void spl_calculate_viewport_size(struct spl_in *spl_in, struct spl_scratch *spl_scratch)
+static void spl_calculate_viewport_size(struct spl_in *spl_in, struct spl_out *spl_out)
 {
-       spl_scratch->scl_data.viewport.width = spl_fixpt_ceil(spl_fixpt_mul_int(spl_scratch->scl_data.ratios.horz,
-                                                       spl_scratch->scl_data.recout.width));
-       spl_scratch->scl_data.viewport.height = spl_fixpt_ceil(spl_fixpt_mul_int(spl_scratch->scl_data.ratios.vert,
-                                                       spl_scratch->scl_data.recout.height));
-       spl_scratch->scl_data.viewport_c.width = spl_fixpt_ceil(spl_fixpt_mul_int(spl_scratch->scl_data.ratios.horz_c,
-                                               spl_scratch->scl_data.recout.width));
-       spl_scratch->scl_data.viewport_c.height = spl_fixpt_ceil(spl_fixpt_mul_int(spl_scratch->scl_data.ratios.vert_c,
-                                               spl_scratch->scl_data.recout.height));
+       spl_out->scl_data.viewport.width = dc_fixpt_ceil(dc_fixpt_mul_int(spl_out->scl_data.ratios.horz,
+                                                       spl_out->scl_data.recout.width));
+       spl_out->scl_data.viewport.height = dc_fixpt_ceil(dc_fixpt_mul_int(spl_out->scl_data.ratios.vert,
+                                                       spl_out->scl_data.recout.height));
+       spl_out->scl_data.viewport_c.width = dc_fixpt_ceil(dc_fixpt_mul_int(spl_out->scl_data.ratios.horz_c,
+                                               spl_out->scl_data.recout.width));
+       spl_out->scl_data.viewport_c.height = dc_fixpt_ceil(dc_fixpt_mul_int(spl_out->scl_data.ratios.vert_c,
+                                               spl_out->scl_data.recout.height));
        if (spl_in->basic_in.rotation == SPL_ROTATION_ANGLE_90 ||
                        spl_in->basic_in.rotation == SPL_ROTATION_ANGLE_270) {
-               spl_swap(spl_scratch->scl_data.viewport.width, spl_scratch->scl_data.viewport.height);
-               spl_swap(spl_scratch->scl_data.viewport_c.width, spl_scratch->scl_data.viewport_c.height);
+               swap(spl_out->scl_data.viewport.width, spl_out->scl_data.viewport.height);
+               swap(spl_out->scl_data.viewport_c.width, spl_out->scl_data.viewport_c.height);
        }
 }
-
 static void spl_get_vp_scan_direction(enum spl_rotation_angle rotation,
                           bool horizontal_mirror,
                           bool *orthogonal_rotation,
@@ -461,7 +440,6 @@ static void spl_get_vp_scan_direction(enum spl_rotation_angle rotation,
        if (horizontal_mirror)
                *flip_horz_scan_dir = !*flip_horz_scan_dir;
 }
-
 /*
  * We completely calculate vp offset, size and inits here based entirely on scaling
  * ratios and recout for pixel perfect pipe combine.
@@ -471,13 +449,13 @@ static void spl_calculate_init_and_vp(bool flip_scan_dir,
                                int recout_size,
                                int src_size,
                                int taps,
-                               struct spl_fixed31_32 ratio,
-                               struct spl_fixed31_32 init_adj,
-                               struct spl_fixed31_32 *init,
+                               struct fixed31_32 ratio,
+                               struct fixed31_32 init_adj,
+                               struct fixed31_32 *init,
                                int *vp_offset,
                                int *vp_size)
 {
-       struct spl_fixed31_32 temp;
+       struct fixed31_32 temp;
        int int_part;
 
        /*
@@ -490,33 +468,33 @@ static void spl_calculate_init_and_vp(bool flip_scan_dir,
         * init_bot = init + scaling_ratio
         * to get pixel perfect combine add the fraction from calculating vp offset
         */
-       temp = spl_fixpt_mul_int(ratio, recout_offset_within_recout_full);
-       *vp_offset = spl_fixpt_floor(temp);
+       temp = dc_fixpt_mul_int(ratio, recout_offset_within_recout_full);
+       *vp_offset = dc_fixpt_floor(temp);
        temp.value &= 0xffffffff;
-       *init = spl_fixpt_add(spl_fixpt_div_int(spl_fixpt_add_int(ratio, taps + 1), 2), temp);
-       *init = spl_fixpt_add(*init, init_adj);
-       *init = spl_fixpt_truncate(*init, 19);
+       *init = dc_fixpt_add(dc_fixpt_div_int(dc_fixpt_add_int(ratio, taps + 1), 2), temp);
+       *init = dc_fixpt_add(*init, init_adj);
+       *init = dc_fixpt_truncate(*init, 19);
 
        /*
         * If viewport has non 0 offset and there are more taps than covered by init then
         * we should decrease the offset and increase init so we are never sampling
         * outside of viewport.
         */
-       int_part = spl_fixpt_floor(*init);
+       int_part = dc_fixpt_floor(*init);
        if (int_part < taps) {
                int_part = taps - int_part;
                if (int_part > *vp_offset)
                        int_part = *vp_offset;
                *vp_offset -= int_part;
-               *init = spl_fixpt_add_int(*init, int_part);
+               *init = dc_fixpt_add_int(*init, int_part);
        }
        /*
         * If taps are sampling outside of viewport at end of recout and there are more pixels
         * available in the surface we should increase the viewport size, regardless set vp to
         * only what is used.
         */
-       temp = spl_fixpt_add(*init, spl_fixpt_mul_int(ratio, recout_size - 1));
-       *vp_size = spl_fixpt_floor(temp);
+       temp = dc_fixpt_add(*init, dc_fixpt_mul_int(ratio, recout_size - 1));
+       *vp_size = dc_fixpt_floor(temp);
        if (*vp_size + *vp_offset > src_size)
                *vp_size = src_size - *vp_offset;
 
@@ -531,16 +509,15 @@ static void spl_calculate_init_and_vp(bool flip_scan_dir,
 
 static bool spl_is_yuv420(enum spl_pixel_format format)
 {
-       if ((format >= SPL_PIXEL_FORMAT_420BPP8) &&
-               (format <= SPL_PIXEL_FORMAT_420BPP10))
+       if ((format >= SPL_PIXEL_FORMAT_VIDEO_BEGIN) &&
+               (format <= SPL_PIXEL_FORMAT_VIDEO_END))
                return true;
 
        return false;
 }
 
 /*Calculate inits and viewport */
-static void spl_calculate_inits_and_viewports(struct spl_in *spl_in,
-               struct spl_scratch *spl_scratch)
+static void spl_calculate_inits_and_viewports(struct spl_in *spl_in, struct spl_out *spl_out)
 {
        struct spl_rect src = spl_in->basic_in.src_rect;
        struct spl_rect recout_dst_in_active_timing;
@@ -551,11 +528,11 @@ static void spl_calculate_inits_and_viewports(struct spl_in *spl_in,
        int vpc_div = (spl_in->basic_in.format == SPL_PIXEL_FORMAT_420BPP8
                        || spl_in->basic_in.format == SPL_PIXEL_FORMAT_420BPP10) ? 2 : 1;
        bool orthogonal_rotation, flip_vert_scan_dir, flip_horz_scan_dir;
-       struct spl_fixed31_32 init_adj_h = spl_fixpt_zero;
-       struct spl_fixed31_32 init_adj_v = spl_fixpt_zero;
+       struct fixed31_32 init_adj_h = dc_fixpt_zero;
+       struct fixed31_32 init_adj_v = dc_fixpt_zero;
 
        recout_clip_in_active_timing = shift_rec(
-                       &spl_scratch->scl_data.recout, odm_slice.x, odm_slice.y);
+                       &spl_out->scl_data.recout, odm_slice.x, odm_slice.y);
        recout_dst_in_active_timing = calculate_plane_rec_in_timing_active(
                        spl_in, &spl_in->basic_in.dst_rect);
        overlap_in_active_timing = intersect_rec(&recout_clip_in_active_timing,
@@ -578,8 +555,8 @@ static void spl_calculate_inits_and_viewports(struct spl_in *spl_in,
                        &flip_horz_scan_dir);
 
        if (orthogonal_rotation) {
-               spl_swap(src.width, src.height);
-               spl_swap(flip_vert_scan_dir, flip_horz_scan_dir);
+               swap(src.width, src.height);
+               swap(flip_vert_scan_dir, flip_horz_scan_dir);
        }
 
        if (spl_is_yuv420(spl_in->basic_in.format)) {
@@ -591,17 +568,17 @@ static void spl_calculate_inits_and_viewports(struct spl_in *spl_in,
                switch (spl_in->basic_in.cositing) {
 
                case CHROMA_COSITING_LEFT:
-                       init_adj_h = spl_fixpt_zero;
-                       init_adj_v = spl_fixpt_from_fraction(sign, 4);
+                       init_adj_h = dc_fixpt_zero;
+                       init_adj_v = dc_fixpt_from_fraction(sign, 2);
                        break;
                case CHROMA_COSITING_NONE:
-                       init_adj_h = spl_fixpt_from_fraction(sign, 4);
-                       init_adj_v = spl_fixpt_from_fraction(sign, 4);
+                       init_adj_h = dc_fixpt_from_fraction(sign, 2);
+                       init_adj_v = dc_fixpt_from_fraction(sign, 2);
                        break;
                case CHROMA_COSITING_TOPLEFT:
                default:
-                       init_adj_h = spl_fixpt_zero;
-                       init_adj_v = spl_fixpt_zero;
+                       init_adj_h = dc_fixpt_zero;
+                       init_adj_v = dc_fixpt_zero;
                        break;
                }
        }
@@ -609,60 +586,59 @@ static void spl_calculate_inits_and_viewports(struct spl_in *spl_in,
        spl_calculate_init_and_vp(
                        flip_horz_scan_dir,
                        recout_clip_in_recout_dst.x,
-                       spl_scratch->scl_data.recout.width,
+                       spl_out->scl_data.recout.width,
                        src.width,
-                       spl_scratch->scl_data.taps.h_taps,
-                       spl_scratch->scl_data.ratios.horz,
-                       spl_fixpt_zero,
-                       &spl_scratch->scl_data.inits.h,
-                       &spl_scratch->scl_data.viewport.x,
-                       &spl_scratch->scl_data.viewport.width);
+                       spl_out->scl_data.taps.h_taps,
+                       spl_out->scl_data.ratios.horz,
+                       dc_fixpt_zero,
+                       &spl_out->scl_data.inits.h,
+                       &spl_out->scl_data.viewport.x,
+                       &spl_out->scl_data.viewport.width);
        spl_calculate_init_and_vp(
                        flip_horz_scan_dir,
                        recout_clip_in_recout_dst.x,
-                       spl_scratch->scl_data.recout.width,
+                       spl_out->scl_data.recout.width,
                        src.width / vpc_div,
-                       spl_scratch->scl_data.taps.h_taps_c,
-                       spl_scratch->scl_data.ratios.horz_c,
+                       spl_out->scl_data.taps.h_taps_c,
+                       spl_out->scl_data.ratios.horz_c,
                        init_adj_h,
-                       &spl_scratch->scl_data.inits.h_c,
-                       &spl_scratch->scl_data.viewport_c.x,
-                       &spl_scratch->scl_data.viewport_c.width);
+                       &spl_out->scl_data.inits.h_c,
+                       &spl_out->scl_data.viewport_c.x,
+                       &spl_out->scl_data.viewport_c.width);
        spl_calculate_init_and_vp(
                        flip_vert_scan_dir,
                        recout_clip_in_recout_dst.y,
-                       spl_scratch->scl_data.recout.height,
+                       spl_out->scl_data.recout.height,
                        src.height,
-                       spl_scratch->scl_data.taps.v_taps,
-                       spl_scratch->scl_data.ratios.vert,
-                       spl_fixpt_zero,
-                       &spl_scratch->scl_data.inits.v,
-                       &spl_scratch->scl_data.viewport.y,
-                       &spl_scratch->scl_data.viewport.height);
+                       spl_out->scl_data.taps.v_taps,
+                       spl_out->scl_data.ratios.vert,
+                       dc_fixpt_zero,
+                       &spl_out->scl_data.inits.v,
+                       &spl_out->scl_data.viewport.y,
+                       &spl_out->scl_data.viewport.height);
        spl_calculate_init_and_vp(
                        flip_vert_scan_dir,
                        recout_clip_in_recout_dst.y,
-                       spl_scratch->scl_data.recout.height,
+                       spl_out->scl_data.recout.height,
                        src.height / vpc_div,
-                       spl_scratch->scl_data.taps.v_taps_c,
-                       spl_scratch->scl_data.ratios.vert_c,
+                       spl_out->scl_data.taps.v_taps_c,
+                       spl_out->scl_data.ratios.vert_c,
                        init_adj_v,
-                       &spl_scratch->scl_data.inits.v_c,
-                       &spl_scratch->scl_data.viewport_c.y,
-                       &spl_scratch->scl_data.viewport_c.height);
+                       &spl_out->scl_data.inits.v_c,
+                       &spl_out->scl_data.viewport_c.y,
+                       &spl_out->scl_data.viewport_c.height);
        if (orthogonal_rotation) {
-               spl_swap(spl_scratch->scl_data.viewport.x, spl_scratch->scl_data.viewport.y);
-               spl_swap(spl_scratch->scl_data.viewport.width, spl_scratch->scl_data.viewport.height);
-               spl_swap(spl_scratch->scl_data.viewport_c.x, spl_scratch->scl_data.viewport_c.y);
-               spl_swap(spl_scratch->scl_data.viewport_c.width, spl_scratch->scl_data.viewport_c.height);
+               swap(spl_out->scl_data.viewport.x, spl_out->scl_data.viewport.y);
+               swap(spl_out->scl_data.viewport.width, spl_out->scl_data.viewport.height);
+               swap(spl_out->scl_data.viewport_c.x, spl_out->scl_data.viewport_c.y);
+               swap(spl_out->scl_data.viewport_c.width, spl_out->scl_data.viewport_c.height);
        }
-       spl_scratch->scl_data.viewport.x += src.x;
-       spl_scratch->scl_data.viewport.y += src.y;
-       SPL_ASSERT(src.x % vpc_div == 0 && src.y % vpc_div == 0);
-       spl_scratch->scl_data.viewport_c.x += src.x / vpc_div;
-       spl_scratch->scl_data.viewport_c.y += src.y / vpc_div;
+       spl_out->scl_data.viewport.x += src.x;
+       spl_out->scl_data.viewport.y += src.y;
+       ASSERT(src.x % vpc_div == 0 && src.y % vpc_div == 0);
+       spl_out->scl_data.viewport_c.x += src.x / vpc_div;
+       spl_out->scl_data.viewport_c.y += src.y / vpc_div;
 }
-
 static void spl_handle_3d_recout(struct spl_in *spl_in, struct spl_rect *recout)
 {
        /*
@@ -671,7 +647,7 @@ static void spl_handle_3d_recout(struct spl_in *spl_in, struct spl_rect *recout)
         * This may break with rotation, good thing we aren't mixing hw rotation and 3d
         */
        if (spl_in->basic_in.mpc_combine_v) {
-               SPL_ASSERT(spl_in->basic_in.rotation == SPL_ROTATION_ANGLE_0 ||
+               ASSERT(spl_in->basic_in.rotation == SPL_ROTATION_ANGLE_0 ||
                        (spl_in->basic_out.view_format != SPL_VIEW_3D_TOP_AND_BOTTOM &&
                                        spl_in->basic_out.view_format != SPL_VIEW_3D_SIDE_BY_SIDE));
                if (spl_in->basic_out.view_format == SPL_VIEW_3D_TOP_AND_BOTTOM)
@@ -689,7 +665,6 @@ static void spl_clamp_viewport(struct spl_rect *viewport)
        if (viewport->width < MIN_VIEWPORT_SIZE)
                viewport->width = MIN_VIEWPORT_SIZE;
 }
-
 static bool spl_dscl_is_420_format(enum spl_pixel_format format)
 {
        if (format == SPL_PIXEL_FORMAT_420BPP8 ||
@@ -698,7 +673,6 @@ static bool spl_dscl_is_420_format(enum spl_pixel_format format)
        else
                return false;
 }
-
 static bool spl_dscl_is_video_format(enum spl_pixel_format format)
 {
        if (format >= SPL_PIXEL_FORMAT_VIDEO_BEGIN
@@ -707,21 +681,17 @@ static bool spl_dscl_is_video_format(enum spl_pixel_format format)
        else
                return false;
 }
-
 static enum scl_mode spl_get_dscl_mode(const struct spl_in *spl_in,
-                               const struct spl_scaler_data *data,
-                               bool enable_isharp, bool enable_easf)
+                               const struct spl_scaler_data *data)
 {
-       const long long one = spl_fixpt_one.value;
+       const long long one = dc_fixpt_one.value;
        enum spl_pixel_format pixel_format = spl_in->basic_in.format;
 
-       /* Bypass if ratio is 1:1 with no ISHARP or force scale on */
        if (data->ratios.horz.value == one
                        && data->ratios.vert.value == one
                        && data->ratios.horz_c.value == one
                        && data->ratios.vert_c.value == one
-                       && !spl_in->basic_out.always_scale
-                       && !enable_isharp)
+                       && !spl_in->basic_out.always_scale)
                return SCL_MODE_SCALING_444_BYPASS;
 
        if (!spl_dscl_is_420_format(pixel_format)) {
@@ -730,204 +700,69 @@ static enum scl_mode spl_get_dscl_mode(const struct spl_in *spl_in,
                else
                        return SCL_MODE_SCALING_444_RGB_ENABLE;
        }
-
-       /* Bypass YUV if at 1:1 with no ISHARP or if doing 2:1 YUV
-        *  downscale without EASF
-        */
-       if ((!enable_isharp) && (!enable_easf)) {
-               if (data->ratios.horz.value == one && data->ratios.vert.value == one)
-                       return SCL_MODE_SCALING_420_LUMA_BYPASS;
-               if (data->ratios.horz_c.value == one && data->ratios.vert_c.value == one)
-                       return SCL_MODE_SCALING_420_CHROMA_BYPASS;
-       }
+       if (data->ratios.horz.value == one && data->ratios.vert.value == one)
+               return SCL_MODE_SCALING_420_LUMA_BYPASS;
+       if (data->ratios.horz_c.value == one && data->ratios.vert_c.value == one)
+               return SCL_MODE_SCALING_420_CHROMA_BYPASS;
 
        return SCL_MODE_SCALING_420_YCBCR_ENABLE;
 }
-
-static bool spl_choose_lls_policy(enum spl_pixel_format format,
-       enum spl_transfer_func_type tf_type,
-       enum spl_transfer_func_predefined tf_predefined_type,
-       enum linear_light_scaling *lls_pref)
-{
-       if (spl_is_yuv420(format)) {
-               *lls_pref = LLS_PREF_NO;
-               if ((tf_type == SPL_TF_TYPE_PREDEFINED) ||
-                       (tf_type == SPL_TF_TYPE_DISTRIBUTED_POINTS))
-                       return true;
-       } else { /* RGB or YUV444 */
-               if ((tf_type == SPL_TF_TYPE_PREDEFINED) ||
-                       (tf_type == SPL_TF_TYPE_BYPASS)) {
-                       *lls_pref = LLS_PREF_YES;
-                       return true;
-               }
-       }
-       *lls_pref = LLS_PREF_NO;
-       return false;
-}
-
-/* Enable EASF ?*/
-static bool enable_easf(struct spl_in *spl_in, struct spl_scratch *spl_scratch)
-{
-       int vratio = 0;
-       int hratio = 0;
-       bool skip_easf = false;
-       bool lls_enable_easf = true;
-
-       /*
-        * If lls_pref is LLS_PREF_DONT_CARE, then use pixel format and transfer
-        *  function to determine whether to use LINEAR or NONLINEAR scaling
-        */
-       if (spl_in->lls_pref == LLS_PREF_DONT_CARE)
-               lls_enable_easf = spl_choose_lls_policy(spl_in->basic_in.format,
-                       spl_in->basic_in.tf_type, spl_in->basic_in.tf_predefined_type,
-                       &spl_in->lls_pref);
-
-       vratio = spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert);
-       hratio = spl_fixpt_ceil(spl_scratch->scl_data.ratios.horz);
-
-       if (!lls_enable_easf || spl_in->disable_easf)
-               skip_easf = true;
-
-       /*
-        * No EASF support for downscaling > 2:1
-        * EASF support for upscaling or downscaling up to 2:1
-        */
-       if ((vratio > 2) || (hratio > 2))
-               skip_easf = true;
-
-       /* Check for linear scaling or EASF preferred */
-       if (spl_in->lls_pref != LLS_PREF_YES && !spl_in->prefer_easf)
-               skip_easf = true;
-
-       return skip_easf;
-}
-
-/* Check if video is in fullscreen mode */
-static bool spl_is_video_fullscreen(struct spl_in *spl_in)
-{
-       if (spl_is_yuv420(spl_in->basic_in.format) && spl_in->is_fullscreen)
-               return true;
-       return false;
-}
-
-static bool spl_get_isharp_en(struct spl_in *spl_in,
-       struct spl_scratch *spl_scratch)
-{
-       bool enable_isharp = false;
-       int vratio = 0;
-       int hratio = 0;
-       struct spl_taps taps = spl_scratch->scl_data.taps;
-       bool fullscreen = spl_is_video_fullscreen(spl_in);
-
-       vratio = spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert);
-       hratio = spl_fixpt_ceil(spl_scratch->scl_data.ratios.horz);
-
-       /* Return if adaptive sharpness is disabled */
-       if (spl_in->adaptive_sharpness.enable == false)
-               return enable_isharp;
-
-       /* No iSHARP support for downscaling */
-       if (vratio > 1 || hratio > 1)
-               return enable_isharp;
-
-       // Scaling is up to 1:1 (no scaling) or upscaling
-
-       /*
-        * Apply sharpness to all RGB surfaces and to
-        *  NV12/P010 surfaces if in fullscreen
-        */
-       if (spl_is_yuv420(spl_in->basic_in.format) && !fullscreen)
-               return enable_isharp;
-
-       /*
-        * Apply sharpness if supports horizontal taps 4,6 AND
-        *  vertical taps 3, 4, 6
-        */
-       if ((taps.h_taps == 4 || taps.h_taps == 6) &&
-               (taps.v_taps == 3 || taps.v_taps == 4 || taps.v_taps == 6))
-               enable_isharp = true;
-
-       return enable_isharp;
-}
-
 /* Calculate optimal number of taps */
 static bool spl_get_optimal_number_of_taps(
-         int max_downscale_src_width, struct spl_in *spl_in, struct spl_scratch *spl_scratch,
-         const struct spl_taps *in_taps, bool *enable_easf_v, bool *enable_easf_h,
-         bool *enable_isharp)
+         int max_downscale_src_width, struct spl_in *spl_in, struct spl_out *spl_out,
+         const struct spl_taps *in_taps)
 {
        int num_part_y, num_part_c;
        int max_taps_y, max_taps_c;
        int min_taps_y, min_taps_c;
        enum lb_memory_config lb_config;
-       bool skip_easf = false;
 
-       if (spl_scratch->scl_data.viewport.width > spl_scratch->scl_data.h_active &&
+       if (spl_out->scl_data.viewport.width > spl_out->scl_data.h_active &&
                max_downscale_src_width != 0 &&
-               spl_scratch->scl_data.viewport.width > max_downscale_src_width)
+               spl_out->scl_data.viewport.width > max_downscale_src_width)
                return false;
-
-       /* Check if we are using EASF or not */
-       skip_easf = enable_easf(spl_in, spl_scratch);
-
        /*
         * Set default taps if none are provided
         * From programming guide: taps = min{ ceil(2*H_RATIO,1), 8} for downscaling
         * taps = 4 for upscaling
         */
-       if (skip_easf) {
-               if (in_taps->h_taps == 0) {
-                       if (spl_fixpt_ceil(spl_scratch->scl_data.ratios.horz) > 1)
-                               spl_scratch->scl_data.taps.h_taps = spl_min(2 * spl_fixpt_ceil(
-                                       spl_scratch->scl_data.ratios.horz), 8);
-                       else
-                               spl_scratch->scl_data.taps.h_taps = 4;
-               } else
-                       spl_scratch->scl_data.taps.h_taps = in_taps->h_taps;
-               if (in_taps->v_taps == 0) {
-                       if (spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert) > 1)
-                               spl_scratch->scl_data.taps.v_taps = spl_min(spl_fixpt_ceil(spl_fixpt_mul_int(
-                                       spl_scratch->scl_data.ratios.vert, 2)), 8);
-                       else
-                               spl_scratch->scl_data.taps.v_taps = 4;
-               } else
-                       spl_scratch->scl_data.taps.v_taps = in_taps->v_taps;
-               if (in_taps->v_taps_c == 0) {
-                       if (spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert_c) > 1)
-                               spl_scratch->scl_data.taps.v_taps_c = spl_min(spl_fixpt_ceil(spl_fixpt_mul_int(
-                                       spl_scratch->scl_data.ratios.vert_c, 2)), 8);
-                       else
-                               spl_scratch->scl_data.taps.v_taps_c = 4;
-               } else
-                       spl_scratch->scl_data.taps.v_taps_c = in_taps->v_taps_c;
-               if (in_taps->h_taps_c == 0) {
-                       if (spl_fixpt_ceil(spl_scratch->scl_data.ratios.horz_c) > 1)
-                               spl_scratch->scl_data.taps.h_taps_c = spl_min(2 * spl_fixpt_ceil(
-                                       spl_scratch->scl_data.ratios.horz_c), 8);
-                       else
-                               spl_scratch->scl_data.taps.h_taps_c = 4;
-               } else if ((in_taps->h_taps_c % 2) != 0 && in_taps->h_taps_c != 1)
-                       /* Only 1 and even h_taps_c are supported by hw */
-                       spl_scratch->scl_data.taps.h_taps_c = in_taps->h_taps_c - 1;
+       if (in_taps->h_taps == 0) {
+               if (dc_fixpt_ceil(spl_out->scl_data.ratios.horz) > 1)
+                       spl_out->scl_data.taps.h_taps = min(2 * dc_fixpt_ceil(spl_out->scl_data.ratios.horz), 8);
                else
-                       spl_scratch->scl_data.taps.h_taps_c = in_taps->h_taps_c;
-       } else {
-               if (spl_is_yuv420(spl_in->basic_in.format)) {
-                       spl_scratch->scl_data.taps.h_taps = 6;
-                       spl_scratch->scl_data.taps.v_taps = 6;
-                       spl_scratch->scl_data.taps.h_taps_c = 4;
-                       spl_scratch->scl_data.taps.v_taps_c = 4;
-               } else { /* RGB */
-                       spl_scratch->scl_data.taps.h_taps = 6;
-                       spl_scratch->scl_data.taps.v_taps = 6;
-                       spl_scratch->scl_data.taps.h_taps_c = 6;
-                       spl_scratch->scl_data.taps.v_taps_c = 6;
-               }
-       }
+                       spl_out->scl_data.taps.h_taps = 4;
+       } else
+               spl_out->scl_data.taps.h_taps = in_taps->h_taps;
+       if (in_taps->v_taps == 0) {
+               if (dc_fixpt_ceil(spl_out->scl_data.ratios.vert) > 1)
+                       spl_out->scl_data.taps.v_taps = min(dc_fixpt_ceil(dc_fixpt_mul_int(
+                                                       spl_out->scl_data.ratios.vert, 2)), 8);
+               else
+                       spl_out->scl_data.taps.v_taps = 4;
+       } else
+               spl_out->scl_data.taps.v_taps = in_taps->v_taps;
+       if (in_taps->v_taps_c == 0) {
+               if (dc_fixpt_ceil(spl_out->scl_data.ratios.vert_c) > 1)
+                       spl_out->scl_data.taps.v_taps_c = min(dc_fixpt_ceil(dc_fixpt_mul_int(
+                                                       spl_out->scl_data.ratios.vert_c, 2)), 8);
+               else
+                       spl_out->scl_data.taps.v_taps_c = 4;
+       } else
+               spl_out->scl_data.taps.v_taps_c = in_taps->v_taps_c;
+       if (in_taps->h_taps_c == 0) {
+               if (dc_fixpt_ceil(spl_out->scl_data.ratios.horz_c) > 1)
+                       spl_out->scl_data.taps.h_taps_c = min(2 * dc_fixpt_ceil(spl_out->scl_data.ratios.horz_c), 8);
+               else
+                       spl_out->scl_data.taps.h_taps_c = 4;
+       } else if ((in_taps->h_taps_c % 2) != 0 && in_taps->h_taps_c != 1)
+               /* Only 1 and even h_taps_c are supported by hw */
+               spl_out->scl_data.taps.h_taps_c = in_taps->h_taps_c - 1;
+       else
+               spl_out->scl_data.taps.h_taps_c = in_taps->h_taps_c;
 
        /*Ensure we can support the requested number of vtaps*/
-       min_taps_y = spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert);
-       min_taps_c = spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert_c);
+       min_taps_y = dc_fixpt_ceil(spl_out->scl_data.ratios.vert);
+       min_taps_c = dc_fixpt_ceil(spl_out->scl_data.ratios.vert_c);
 
        /* Use LB_MEMORY_CONFIG_3 for 4:2:0 */
        if ((spl_in->basic_in.format == SPL_PIXEL_FORMAT_420BPP8)
@@ -936,16 +771,16 @@ static bool spl_get_optimal_number_of_taps(
        else
                lb_config = LB_MEMORY_CONFIG_0;
        // Determine max vtap support by calculating how much line buffer can fit
-       spl_in->funcs->spl_calc_lb_num_partitions(spl_in->basic_out.alpha_en, &spl_scratch->scl_data,
+       spl_in->funcs->spl_calc_lb_num_partitions(spl_in->basic_out.alpha_en, &spl_out->scl_data,
                        lb_config, &num_part_y, &num_part_c);
        /* MAX_V_TAPS = MIN (NUM_LINES - MAX(CEILING(V_RATIO,1)-2, 0), 8) */
-       if (spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert) > 2)
-               max_taps_y = num_part_y - (spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert) - 2);
+       if (dc_fixpt_ceil(spl_out->scl_data.ratios.vert) > 2)
+               max_taps_y = num_part_y - (dc_fixpt_ceil(spl_out->scl_data.ratios.vert) - 2);
        else
                max_taps_y = num_part_y;
 
-       if (spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert_c) > 2)
-               max_taps_c = num_part_c - (spl_fixpt_ceil(spl_scratch->scl_data.ratios.vert_c) - 2);
+       if (dc_fixpt_ceil(spl_out->scl_data.ratios.vert_c) > 2)
+               max_taps_c = num_part_c - (dc_fixpt_ceil(spl_out->scl_data.ratios.vert_c) - 2);
        else
                max_taps_c = num_part_c;
 
@@ -954,108 +789,48 @@ static bool spl_get_optimal_number_of_taps(
        else if (max_taps_c < min_taps_c)
                return false;
 
-       if (spl_scratch->scl_data.taps.v_taps > max_taps_y)
-               spl_scratch->scl_data.taps.v_taps = max_taps_y;
-
-       if (spl_scratch->scl_data.taps.v_taps_c > max_taps_c)
-               spl_scratch->scl_data.taps.v_taps_c = max_taps_c;
+       if (spl_out->scl_data.taps.v_taps > max_taps_y)
+               spl_out->scl_data.taps.v_taps = max_taps_y;
+
+       if (spl_out->scl_data.taps.v_taps_c > max_taps_c)
+               spl_out->scl_data.taps.v_taps_c = max_taps_c;
+       if (spl_in->prefer_easf)        {
+               // EASF can be enabled only for taps 3,4,6
+               // If optimal no of taps is 5, then set it to 4
+               // If optimal no of taps is 7 or 8, then set it to 6
+               if (spl_out->scl_data.taps.v_taps == 5)
+                       spl_out->scl_data.taps.v_taps = 4;
+               if (spl_out->scl_data.taps.v_taps == 7 || spl_out->scl_data.taps.v_taps == 8)
+                       spl_out->scl_data.taps.v_taps = 6;
+
+               if (spl_out->scl_data.taps.v_taps_c == 5)
+                       spl_out->scl_data.taps.v_taps_c = 4;
+               if (spl_out->scl_data.taps.v_taps_c == 7 || spl_out->scl_data.taps.v_taps_c == 8)
+                       spl_out->scl_data.taps.v_taps_c = 6;
+
+               if (spl_out->scl_data.taps.h_taps == 5)
+                       spl_out->scl_data.taps.h_taps = 4;
+               if (spl_out->scl_data.taps.h_taps == 7 || spl_out->scl_data.taps.h_taps == 8)
+                       spl_out->scl_data.taps.h_taps = 6;
+
+               if (spl_out->scl_data.taps.h_taps_c == 5)
+                       spl_out->scl_data.taps.h_taps_c = 4;
+               if (spl_out->scl_data.taps.h_taps_c == 7 || spl_out->scl_data.taps.h_taps_c == 8)
+                       spl_out->scl_data.taps.h_taps_c = 6;
 
-       if (!skip_easf) {
-               /*
-                * RGB ( L + NL ) and Linear HDR support 6x6, 6x4, 6x3, 4x4, 4x3
-                * NL YUV420 only supports 6x6, 6x4 for Y and 4x4 for UV
-                *
-                * If LB does not support 3, 4, or 6 taps, then disable EASF_V
-                *  and only enable EASF_H.  So for RGB, support 6x2, 4x2
-                *  and for NL YUV420, support 6x2 for Y and 4x2 for UV
-                *
-                * All other cases, have to disable EASF_V and EASF_H
-                *
-                * If optimal no of taps is 5, then set it to 4
-                * If optimal no of taps is 7 or 8, then fine since max tap is 6
-                *
-                */
-               if (spl_scratch->scl_data.taps.v_taps == 5)
-                       spl_scratch->scl_data.taps.v_taps = 4;
-
-               if (spl_scratch->scl_data.taps.v_taps_c == 5)
-                       spl_scratch->scl_data.taps.v_taps_c = 4;
-
-               if (spl_scratch->scl_data.taps.h_taps == 5)
-                       spl_scratch->scl_data.taps.h_taps = 4;
-
-               if (spl_scratch->scl_data.taps.h_taps_c == 5)
-                       spl_scratch->scl_data.taps.h_taps_c = 4;
-
-               if (spl_is_yuv420(spl_in->basic_in.format)) {
-                       if ((spl_scratch->scl_data.taps.h_taps <= 4) ||
-                               (spl_scratch->scl_data.taps.h_taps_c <= 3)) {
-                               *enable_easf_v = false;
-                               *enable_easf_h = false;
-                       } else if ((spl_scratch->scl_data.taps.v_taps <= 3) ||
-                               (spl_scratch->scl_data.taps.v_taps_c <= 3)) {
-                               *enable_easf_v = false;
-                               *enable_easf_h = true;
-                       } else {
-                               *enable_easf_v = true;
-                               *enable_easf_h = true;
-                       }
-                       SPL_ASSERT((spl_scratch->scl_data.taps.v_taps > 1) &&
-                               (spl_scratch->scl_data.taps.v_taps_c > 1));
-               } else { /* RGB */
-                       if (spl_scratch->scl_data.taps.h_taps <= 3) {
-                               *enable_easf_v = false;
-                               *enable_easf_h = false;
-                       } else if (spl_scratch->scl_data.taps.v_taps < 3) {
-                               *enable_easf_v = false;
-                               *enable_easf_h = true;
-                       } else {
-                               *enable_easf_v = true;
-                               *enable_easf_h = true;
-                       }
-                       SPL_ASSERT(spl_scratch->scl_data.taps.v_taps > 1);
-               }
-       } else {
-               *enable_easf_v = false;
-               *enable_easf_h = false;
        } // end of if prefer_easf
-
-       /* Sharpener requires scaler to be enabled, including for 1:1
-        * Check if ISHARP can be enabled
-        * If ISHARP is not enabled, for 1:1, set taps to 1 and disable
-        *  EASF
-        * For case of 2:1 YUV where chroma is 1:1, set taps to 1 if
-        *  EASF is not enabled
-        */
-
-       *enable_isharp = spl_get_isharp_en(spl_in, spl_scratch);
-       if (!*enable_isharp && !spl_in->basic_out.always_scale) {
-               if ((IDENTITY_RATIO(spl_scratch->scl_data.ratios.horz)) &&
-                       (IDENTITY_RATIO(spl_scratch->scl_data.ratios.vert))) {
-                       spl_scratch->scl_data.taps.h_taps = 1;
-                       spl_scratch->scl_data.taps.v_taps = 1;
-
-                       if (IDENTITY_RATIO(spl_scratch->scl_data.ratios.horz_c))
-                               spl_scratch->scl_data.taps.h_taps_c = 1;
-
-                       if (IDENTITY_RATIO(spl_scratch->scl_data.ratios.vert_c))
-                               spl_scratch->scl_data.taps.v_taps_c = 1;
-
-                       *enable_easf_v = false;
-                       *enable_easf_h = false;
-               } else {
-                       if ((!*enable_easf_h) &&
-                               (IDENTITY_RATIO(spl_scratch->scl_data.ratios.horz_c)))
-                               spl_scratch->scl_data.taps.h_taps_c = 1;
-
-                       if ((!*enable_easf_v) &&
-                               (IDENTITY_RATIO(spl_scratch->scl_data.ratios.vert_c)))
-                               spl_scratch->scl_data.taps.v_taps_c = 1;
-               }
+       if (!spl_in->basic_out.always_scale)    {
+               if (IDENTITY_RATIO(spl_out->scl_data.ratios.horz))
+                       spl_out->scl_data.taps.h_taps = 1;
+               if (IDENTITY_RATIO(spl_out->scl_data.ratios.vert))
+                       spl_out->scl_data.taps.v_taps = 1;
+               if (IDENTITY_RATIO(spl_out->scl_data.ratios.horz_c))
+                       spl_out->scl_data.taps.h_taps_c = 1;
+               if (IDENTITY_RATIO(spl_out->scl_data.ratios.vert_c))
+                       spl_out->scl_data.taps.v_taps_c = 1;
        }
        return true;
 }
-
 static void spl_set_black_color_data(enum spl_pixel_format format,
                        struct scl_black_color *scl_black_color)
 {
@@ -1073,38 +848,38 @@ static void spl_set_black_color_data(enum spl_pixel_format format,
 static void spl_set_manual_ratio_init_data(struct dscl_prog_data *dscl_prog_data,
                const struct spl_scaler_data *scl_data)
 {
-       struct spl_fixed31_32 bot;
+       struct fixed31_32 bot;
 
-       dscl_prog_data->ratios.h_scale_ratio = spl_fixpt_u3d19(scl_data->ratios.horz) << 5;
-       dscl_prog_data->ratios.v_scale_ratio = spl_fixpt_u3d19(scl_data->ratios.vert) << 5;
-       dscl_prog_data->ratios.h_scale_ratio_c = spl_fixpt_u3d19(scl_data->ratios.horz_c) << 5;
-       dscl_prog_data->ratios.v_scale_ratio_c = spl_fixpt_u3d19(scl_data->ratios.vert_c) << 5;
+       dscl_prog_data->ratios.h_scale_ratio = dc_fixpt_u3d19(scl_data->ratios.horz) << 5;
+       dscl_prog_data->ratios.v_scale_ratio = dc_fixpt_u3d19(scl_data->ratios.vert) << 5;
+       dscl_prog_data->ratios.h_scale_ratio_c = dc_fixpt_u3d19(scl_data->ratios.horz_c) << 5;
+       dscl_prog_data->ratios.v_scale_ratio_c = dc_fixpt_u3d19(scl_data->ratios.vert_c) << 5;
        /*
         * 0.24 format for fraction, first five bits zeroed
         */
        dscl_prog_data->init.h_filter_init_frac =
-                       spl_fixpt_u0d19(scl_data->inits.h) << 5;
+                       dc_fixpt_u0d19(scl_data->inits.h) << 5;
        dscl_prog_data->init.h_filter_init_int =
-                       spl_fixpt_floor(scl_data->inits.h);
+                       dc_fixpt_floor(scl_data->inits.h);
        dscl_prog_data->init.h_filter_init_frac_c =
-                       spl_fixpt_u0d19(scl_data->inits.h_c) << 5;
+                       dc_fixpt_u0d19(scl_data->inits.h_c) << 5;
        dscl_prog_data->init.h_filter_init_int_c =
-                       spl_fixpt_floor(scl_data->inits.h_c);
+                       dc_fixpt_floor(scl_data->inits.h_c);
        dscl_prog_data->init.v_filter_init_frac =
-                       spl_fixpt_u0d19(scl_data->inits.v) << 5;
+                       dc_fixpt_u0d19(scl_data->inits.v) << 5;
        dscl_prog_data->init.v_filter_init_int =
-                       spl_fixpt_floor(scl_data->inits.v);
+                       dc_fixpt_floor(scl_data->inits.v);
        dscl_prog_data->init.v_filter_init_frac_c =
-                       spl_fixpt_u0d19(scl_data->inits.v_c) << 5;
+                       dc_fixpt_u0d19(scl_data->inits.v_c) << 5;
        dscl_prog_data->init.v_filter_init_int_c =
-                       spl_fixpt_floor(scl_data->inits.v_c);
-
-       bot = spl_fixpt_add(scl_data->inits.v, scl_data->ratios.vert);
-       dscl_prog_data->init.v_filter_init_bot_frac = spl_fixpt_u0d19(bot) << 5;
-       dscl_prog_data->init.v_filter_init_bot_int = spl_fixpt_floor(bot);
-       bot = spl_fixpt_add(scl_data->inits.v_c, scl_data->ratios.vert_c);
-       dscl_prog_data->init.v_filter_init_bot_frac_c = spl_fixpt_u0d19(bot) << 5;
-       dscl_prog_data->init.v_filter_init_bot_int_c = spl_fixpt_floor(bot);
+                       dc_fixpt_floor(scl_data->inits.v_c);
+
+       bot = dc_fixpt_add(scl_data->inits.v, scl_data->ratios.vert);
+       dscl_prog_data->init.v_filter_init_bot_frac = dc_fixpt_u0d19(bot) << 5;
+       dscl_prog_data->init.v_filter_init_bot_int = dc_fixpt_floor(bot);
+       bot = dc_fixpt_add(scl_data->inits.v_c, scl_data->ratios.vert_c);
+       dscl_prog_data->init.v_filter_init_bot_frac_c = dc_fixpt_u0d19(bot) << 5;
+       dscl_prog_data->init.v_filter_init_bot_int_c = dc_fixpt_floor(bot);
 }
 
 static void spl_set_taps_data(struct dscl_prog_data *dscl_prog_data,
@@ -1115,28 +890,79 @@ static void spl_set_taps_data(struct dscl_prog_data *dscl_prog_data,
        dscl_prog_data->taps.v_taps_c = scl_data->taps.v_taps_c - 1;
        dscl_prog_data->taps.h_taps_c = scl_data->taps.h_taps_c - 1;
 }
-
+static const uint16_t *spl_dscl_get_filter_coeffs_64p(int taps, struct fixed31_32 ratio)
+{
+       if (taps == 8)
+               return spl_get_filter_8tap_64p(ratio);
+       else if (taps == 7)
+               return spl_get_filter_7tap_64p(ratio);
+       else if (taps == 6)
+               return spl_get_filter_6tap_64p(ratio);
+       else if (taps == 5)
+               return spl_get_filter_5tap_64p(ratio);
+       else if (taps == 4)
+               return spl_get_filter_4tap_64p(ratio);
+       else if (taps == 3)
+               return spl_get_filter_3tap_64p(ratio);
+       else if (taps == 2)
+               return spl_get_filter_2tap_64p();
+       else if (taps == 1)
+               return NULL;
+       else {
+               /* should never happen, bug */
+               BREAK_TO_DEBUGGER();
+               return NULL;
+       }
+}
+static void spl_set_filters_data(struct dscl_prog_data *dscl_prog_data,
+               const struct spl_scaler_data *data)
+{
+       dscl_prog_data->filter_h = spl_dscl_get_filter_coeffs_64p(
+                               data->taps.h_taps, data->ratios.horz);
+       dscl_prog_data->filter_v = spl_dscl_get_filter_coeffs_64p(
+                               data->taps.v_taps, data->ratios.vert);
+       dscl_prog_data->filter_h_c = spl_dscl_get_filter_coeffs_64p(
+                               data->taps.h_taps_c, data->ratios.horz_c);
+       dscl_prog_data->filter_v_c = spl_dscl_get_filter_coeffs_64p(
+                               data->taps.v_taps_c, data->ratios.vert_c);
+}
+#ifdef CONFIG_DRM_AMD_DC_FP
+static const uint16_t *spl_dscl_get_blur_scale_coeffs_64p(int taps)
+{
+       if ((taps == 3) || (taps == 4) || (taps == 6))
+               return spl_get_filter_isharp_bs_4tap_64p();
+       else {
+               /* should never happen, bug */
+               BREAK_TO_DEBUGGER();
+               return NULL;
+       }
+}
+static void spl_set_blur_scale_data(struct dscl_prog_data *dscl_prog_data,
+               const struct spl_scaler_data *data)
+{
+       dscl_prog_data->filter_blur_scale_h = spl_dscl_get_blur_scale_coeffs_64p(
+                               data->taps.h_taps);
+       dscl_prog_data->filter_blur_scale_v = spl_dscl_get_blur_scale_coeffs_64p(
+                               data->taps.v_taps);
+}
+#endif
 /* Populate dscl prog data structure from scaler data calculated by SPL */
-static void spl_set_dscl_prog_data(struct spl_in *spl_in, struct spl_scratch *spl_scratch,
-       struct spl_out *spl_out, bool enable_easf_v, bool enable_easf_h, bool enable_isharp)
+static void spl_set_dscl_prog_data(struct spl_in *spl_in, struct spl_out *spl_out)
 {
        struct dscl_prog_data *dscl_prog_data = spl_out->dscl_prog_data;
 
-       const struct spl_scaler_data *data = &spl_scratch->scl_data;
+       const struct spl_scaler_data *data = &spl_out->scl_data;
 
        struct scl_black_color *scl_black_color = &dscl_prog_data->scl_black_color;
 
-       bool enable_easf = enable_easf_v || enable_easf_h;
-
        // Set values for recout
-       dscl_prog_data->recout = spl_scratch->scl_data.recout;
+       dscl_prog_data->recout = spl_out->scl_data.recout;
        // Set values for MPC Size
-       dscl_prog_data->mpc_size.width = spl_scratch->scl_data.h_active;
-       dscl_prog_data->mpc_size.height = spl_scratch->scl_data.v_active;
+       dscl_prog_data->mpc_size.width = spl_out->scl_data.h_active;
+       dscl_prog_data->mpc_size.height = spl_out->scl_data.v_active;
 
        // SCL_MODE - Set SCL_MODE data
-       dscl_prog_data->dscl_mode = spl_get_dscl_mode(spl_in, data, enable_isharp,
-               enable_easf);
+       dscl_prog_data->dscl_mode = spl_get_dscl_mode(spl_in, data);
 
        // SCL_BLACK_COLOR
        spl_set_black_color_data(spl_in->basic_in.format, scl_black_color);
@@ -1147,101 +973,103 @@ static void spl_set_dscl_prog_data(struct spl_in *spl_in, struct spl_scratch *sp
        // Set HTaps/VTaps
        spl_set_taps_data(dscl_prog_data, data);
        // Set viewport
-       dscl_prog_data->viewport = spl_scratch->scl_data.viewport;
+       dscl_prog_data->viewport = spl_out->scl_data.viewport;
        // Set viewport_c
-       dscl_prog_data->viewport_c = spl_scratch->scl_data.viewport_c;
+       dscl_prog_data->viewport_c = spl_out->scl_data.viewport_c;
        // Set filters data
-       spl_set_filters_data(dscl_prog_data, data, enable_easf_v, enable_easf_h);
+       spl_set_filters_data(dscl_prog_data, data);
+}
+/* Enable EASF ?*/
+static bool enable_easf(int scale_ratio, int taps,
+               enum linear_light_scaling lls_pref, bool prefer_easf)
+{
+       // Is downscaling > 6:1 ?
+       if (scale_ratio > 6) {
+               // END - No EASF support for downscaling > 6:1
+               return false;
+       }
+       // Is upscaling or downscaling up to 2:1?
+       if (scale_ratio <= 2) {
+               // Is linear scaling or EASF preferred?
+               if (lls_pref == LLS_PREF_YES || prefer_easf)    {
+                       // LB support taps 3, 4, 6
+                       if (taps == 3 || taps == 4 || taps == 6) {
+                               // END - EASF supported
+                               return true;
+                       }
+               }
+       }
+       // END - EASF not supported
+       return false;
 }
-
 /* Set EASF data */
-static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *spl_out, bool enable_easf_v,
-       bool enable_easf_h, enum linear_light_scaling lls_pref,
-       enum spl_pixel_format format, enum system_setup setup)
+static void spl_set_easf_data(struct dscl_prog_data *dscl_prog_data,
+       bool enable_easf_v, bool enable_easf_h, enum linear_light_scaling lls_pref,
+       enum spl_pixel_format format)
 {
-       struct dscl_prog_data *dscl_prog_data = spl_out->dscl_prog_data;
+       if (spl_is_yuv420(format)) /* TODO: 0 = RGB, 1 = YUV */
+               dscl_prog_data->easf_matrix_mode = 1;
+       else
+               dscl_prog_data->easf_matrix_mode = 0;
+
        if (enable_easf_v) {
                dscl_prog_data->easf_v_en = true;
                dscl_prog_data->easf_v_ring = 0;
-               dscl_prog_data->easf_v_sharp_factor = 0;
+               dscl_prog_data->easf_v_sharp_factor = 1;
                dscl_prog_data->easf_v_bf1_en = 1;      // 1-bit, BF1 calculation enable, 0=disable, 1=enable
                dscl_prog_data->easf_v_bf2_mode = 0xF;  // 4-bit, BF2 calculation mode
-               /* 2-bit, BF3 chroma mode correction calculation mode */
-               dscl_prog_data->easf_v_bf3_mode = spl_get_v_bf3_mode(
-                       spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10 [ minCoef ]*/
+               dscl_prog_data->easf_v_bf3_mode = 2;    // 2-bit, BF3 chroma mode correction calculation mode
+               dscl_prog_data->easf_v_bf2_flat1_gain = 4;      // U1.3, BF2 Flat1 Gain control
+               dscl_prog_data->easf_v_bf2_flat2_gain = 8;      // U4.0, BF2 Flat2 Gain control
+               dscl_prog_data->easf_v_bf2_roc_gain = 4;        // U2.2, Rate Of Change control
                dscl_prog_data->easf_v_ringest_3tap_dntilt_uptilt =
-                       spl_get_3tap_dntilt_uptilt_offset(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10 [ upTiltMaxVal ]*/
+                       0x9F00;// FP1.5.10 [minCoef]           (-0.036109167214271)
                dscl_prog_data->easf_v_ringest_3tap_uptilt_max =
-                       spl_get_3tap_uptilt_maxval(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10 [ dnTiltSlope ]*/
+                       0x24FE;       // FP1.5.10 [upTiltMaxVal]      ( 0.904556445553545)
                dscl_prog_data->easf_v_ringest_3tap_dntilt_slope =
-                       spl_get_3tap_dntilt_slope(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10 [ upTilt1Slope ]*/
+                       0x3940;       // FP1.5.10 [dnTiltSlope]       ( 0.910488988173371)
                dscl_prog_data->easf_v_ringest_3tap_uptilt1_slope =
-                       spl_get_3tap_uptilt1_slope(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10 [ upTilt2Slope ]*/
+                       0x359C;       // FP1.5.10 [upTilt1Slope]      ( 0.125620179040899)
                dscl_prog_data->easf_v_ringest_3tap_uptilt2_slope =
-                       spl_get_3tap_uptilt2_slope(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10 [ upTilt2Offset ]*/
+                       0x359C;       // FP1.5.10 [upTilt2Slope]      ( 0.006786817723568)
                dscl_prog_data->easf_v_ringest_3tap_uptilt2_offset =
-                       spl_get_3tap_uptilt2_offset(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10; (2.0) Ring reducer gain for 4 or 6-tap mode [H_REDUCER_GAIN4] */
+                       0x9F00;       // FP1.5.10 [upTilt2Offset]     (-0.006139059716651)
                dscl_prog_data->easf_v_ringest_eventap_reduceg1 =
-                       spl_get_reducer_gain4(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10; (2.5) Ring reducer gain for 6-tap mode [V_REDUCER_GAIN6] */
+                       0x4000;   // FP1.5.10; (2.0) Ring reducer gain for 4 or 6-tap mode [H_REDUCER_GAIN4]
                dscl_prog_data->easf_v_ringest_eventap_reduceg2 =
-                       spl_get_reducer_gain6(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10; (-0.135742) Ring gain for 6-tap set to -139/1024 */
+                       0x4100;   // FP1.5.10; (2.5) Ring reducer gain for 6-tap mode [V_REDUCER_GAIN6]
                dscl_prog_data->easf_v_ringest_eventap_gain1 =
-                       spl_get_gainRing4(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
-               /* FP1.5.10; (-0.024414) Ring gain for 6-tap set to -25/1024 */
+                       0xB058;   // FP1.5.10; (-0.135742) Ring gain for 6-tap set to -139/1024
                dscl_prog_data->easf_v_ringest_eventap_gain2 =
-                       spl_get_gainRing6(spl_scratch->scl_data.taps.v_taps,
-                               spl_scratch->scl_data.recip_ratios.vert);
+                       0xA640;    // FP1.5.10; (-0.024414) Ring gain for 6-tap set to -25/1024
                dscl_prog_data->easf_v_bf_maxa = 63; //Vertical Max BF value A in U0.6 format.Selected if V_FCNTL == 0
                dscl_prog_data->easf_v_bf_maxb = 63; //Vertical Max BF value A in U0.6 format.Selected if V_FCNTL == 1
                dscl_prog_data->easf_v_bf_mina = 0;     //Vertical Min BF value A in U0.6 format.Selected if V_FCNTL == 0
                dscl_prog_data->easf_v_bf_minb = 0;     //Vertical Min BF value A in U0.6 format.Selected if V_FCNTL == 1
+               dscl_prog_data->easf_v_bf1_pwl_in_seg0 = -512;  // S0.10, BF1 PWL Segment 0
+               dscl_prog_data->easf_v_bf1_pwl_base_seg0 = 0;   // U0.6, BF1 Base PWL Segment 0
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg0 = 3;  // S7.3, BF1 Slope PWL Segment 0
+               dscl_prog_data->easf_v_bf1_pwl_in_seg1 = -20;   // S0.10, BF1 PWL Segment 1
+               dscl_prog_data->easf_v_bf1_pwl_base_seg1 = 12;  // U0.6, BF1 Base PWL Segment 1
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg1 = 326;        // S7.3, BF1 Slope PWL Segment 1
+               dscl_prog_data->easf_v_bf1_pwl_in_seg2 = 0;     // S0.10, BF1 PWL Segment 2
+               dscl_prog_data->easf_v_bf1_pwl_base_seg2 = 63;  // U0.6, BF1 Base PWL Segment 2
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg2 = 0;  // S7.3, BF1 Slope PWL Segment 2
+               dscl_prog_data->easf_v_bf1_pwl_in_seg3 = 16;    // S0.10, BF1 PWL Segment 3
+               dscl_prog_data->easf_v_bf1_pwl_base_seg3 = 63;  // U0.6, BF1 Base PWL Segment 3
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg3 = -56;        // S7.3, BF1 Slope PWL Segment 3
+               dscl_prog_data->easf_v_bf1_pwl_in_seg4 = 32;    // S0.10, BF1 PWL Segment 4
+               dscl_prog_data->easf_v_bf1_pwl_base_seg4 = 56;  // U0.6, BF1 Base PWL Segment 4
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg4 = -48;        // S7.3, BF1 Slope PWL Segment 4
+               dscl_prog_data->easf_v_bf1_pwl_in_seg5 = 48;    // S0.10, BF1 PWL Segment 5
+               dscl_prog_data->easf_v_bf1_pwl_base_seg5 = 50;  // U0.6, BF1 Base PWL Segment 5
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg5 = -240;       // S7.3, BF1 Slope PWL Segment 5
+               dscl_prog_data->easf_v_bf1_pwl_in_seg6 = 64;    // S0.10, BF1 PWL Segment 6
+               dscl_prog_data->easf_v_bf1_pwl_base_seg6 = 20;  // U0.6, BF1 Base PWL Segment 6
+               dscl_prog_data->easf_v_bf1_pwl_slope_seg6 = -160;       // S7.3, BF1 Slope PWL Segment 6
+               dscl_prog_data->easf_v_bf1_pwl_in_seg7 = 80;    // S0.10, BF1 PWL Segment 7
+               dscl_prog_data->easf_v_bf1_pwl_base_seg7 = 0;   // U0.6, BF1 Base PWL Segment 7
                if (lls_pref == LLS_PREF_YES)   {
-                       dscl_prog_data->easf_v_bf2_flat1_gain = 4;      // U1.3, BF2 Flat1 Gain control
-                       dscl_prog_data->easf_v_bf2_flat2_gain = 8;      // U4.0, BF2 Flat2 Gain control
-                       dscl_prog_data->easf_v_bf2_roc_gain = 4;        // U2.2, Rate Of Change control
-
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg0 = 0x600; // S0.10, BF1 PWL Segment 0 = -512
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg0 = 0;   // U0.6, BF1 Base PWL Segment 0
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg0 = 3;  // S7.3, BF1 Slope PWL Segment 0
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg1 = 0x7EC; // S0.10, BF1 PWL Segment 1 = -20
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg1 = 12;  // U0.6, BF1 Base PWL Segment 1
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg1 = 326;        // S7.3, BF1 Slope PWL Segment 1
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg2 = 0;     // S0.10, BF1 PWL Segment 2
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg2 = 63;  // U0.6, BF1 Base PWL Segment 2
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg2 = 0;  // S7.3, BF1 Slope PWL Segment 2
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg3 = 16;    // S0.10, BF1 PWL Segment 3
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg3 = 63;  // U0.6, BF1 Base PWL Segment 3
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg3 = 0x7C8;      // S7.3, BF1 Slope PWL Segment 3 = -56
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg4 = 32;    // S0.10, BF1 PWL Segment 4
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg4 = 56;  // U0.6, BF1 Base PWL Segment 4
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg4 = 0x7D0;      // S7.3, BF1 Slope PWL Segment 4 = -48
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg5 = 48;    // S0.10, BF1 PWL Segment 5
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg5 = 50;  // U0.6, BF1 Base PWL Segment 5
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg5 = 0x710;      // S7.3, BF1 Slope PWL Segment 5 = -240
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg6 = 64;    // S0.10, BF1 PWL Segment 6
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg6 = 20;  // U0.6, BF1 Base PWL Segment 6
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg6 = 0x760;      // S7.3, BF1 Slope PWL Segment 6 = -160
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg7 = 80;    // S0.10, BF1 PWL Segment 7
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg7 = 0;   // U0.6, BF1 Base PWL Segment 7
-
                        dscl_prog_data->easf_v_bf3_pwl_in_set0 = 0x000; // FP0.6.6, BF3 Input value PWL Segment 0
                        dscl_prog_data->easf_v_bf3_pwl_base_set0 = 63;  // S0.6, BF3 Base PWL Segment 0
                        dscl_prog_data->easf_v_bf3_pwl_slope_set0 = 0x12C5;     // FP1.6.6, BF3 Slope PWL Segment 0
@@ -1262,41 +1090,13 @@ static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *s
                                0x136B; // FP1.6.6, BF3 Slope PWL Segment 3
                        dscl_prog_data->easf_v_bf3_pwl_in_set4 =
                                0x0C37; // FP0.6.6, BF3 Input value PWL Segment 4 (0.125 * 125^3)
-                       dscl_prog_data->easf_v_bf3_pwl_base_set4 = 0x4E;        // S0.6, BF3 Base PWL Segment 4 = -50
+                       dscl_prog_data->easf_v_bf3_pwl_base_set4 = -50; // S0.6, BF3 Base PWL Segment 4
                        dscl_prog_data->easf_v_bf3_pwl_slope_set4 =
                                0x1200; // FP1.6.6, BF3 Slope PWL Segment 4
                        dscl_prog_data->easf_v_bf3_pwl_in_set5 =
                                0x0CF7; // FP0.6.6, BF3 Input value PWL Segment 5 (1.0 * 125^3)
-                       dscl_prog_data->easf_v_bf3_pwl_base_set5 = 0x41;        // S0.6, BF3 Base PWL Segment 5 = -63
+                       dscl_prog_data->easf_v_bf3_pwl_base_set5 = -63; // S0.6, BF3 Base PWL Segment 5
                }       else    {
-                       dscl_prog_data->easf_v_bf2_flat1_gain = 13;     // U1.3, BF2 Flat1 Gain control
-                       dscl_prog_data->easf_v_bf2_flat2_gain = 15;     // U4.0, BF2 Flat2 Gain control
-                       dscl_prog_data->easf_v_bf2_roc_gain = 14;       // U2.2, Rate Of Change control
-
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg0 = 0x440; // S0.10, BF1 PWL Segment 0 = -960
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg0 = 0;   // U0.6, BF1 Base PWL Segment 0
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg0 = 2;  // S7.3, BF1 Slope PWL Segment 0
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg1 = 0x7C4; // S0.10, BF1 PWL Segment 1 = -60
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg1 = 12;  // U0.6, BF1 Base PWL Segment 1
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg1 = 109;        // S7.3, BF1 Slope PWL Segment 1
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg2 = 0;     // S0.10, BF1 PWL Segment 2
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg2 = 63;  // U0.6, BF1 Base PWL Segment 2
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg2 = 0;  // S7.3, BF1 Slope PWL Segment 2
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg3 = 48;    // S0.10, BF1 PWL Segment 3
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg3 = 63;  // U0.6, BF1 Base PWL Segment 3
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg3 = 0x7ED;      // S7.3, BF1 Slope PWL Segment 3 = -19
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg4 = 96;    // S0.10, BF1 PWL Segment 4
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg4 = 56;  // U0.6, BF1 Base PWL Segment 4
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg4 = 0x7F0;      // S7.3, BF1 Slope PWL Segment 4 = -16
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg5 = 144;   // S0.10, BF1 PWL Segment 5
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg5 = 50;  // U0.6, BF1 Base PWL Segment 5
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg5 = 0x7B0;      // S7.3, BF1 Slope PWL Segment 5 = -80
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg6 = 192;   // S0.10, BF1 PWL Segment 6
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg6 = 20;  // U0.6, BF1 Base PWL Segment 6
-                       dscl_prog_data->easf_v_bf1_pwl_slope_seg6 = 0x7CB;      // S7.3, BF1 Slope PWL Segment 6 = -53
-                       dscl_prog_data->easf_v_bf1_pwl_in_seg7 = 240;   // S0.10, BF1 PWL Segment 7
-                       dscl_prog_data->easf_v_bf1_pwl_base_seg7 = 0;   // U0.6, BF1 Base PWL Segment 7
-
                        dscl_prog_data->easf_v_bf3_pwl_in_set0 = 0x000; // FP0.6.6, BF3 Input value PWL Segment 0
                        dscl_prog_data->easf_v_bf3_pwl_base_set0 = 63;  // S0.6, BF3 Base PWL Segment 0
                        dscl_prog_data->easf_v_bf3_pwl_slope_set0 = 0x0000;     // FP1.6.6, BF3 Slope PWL Segment 0
@@ -1315,11 +1115,11 @@ static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *s
                                0x1878; // FP1.6.6, BF3 Slope PWL Segment 3
                        dscl_prog_data->easf_v_bf3_pwl_in_set4 =
                                0x0761; // FP0.6.6, BF3 Input value PWL Segment 4 (0.375)
-                       dscl_prog_data->easf_v_bf3_pwl_base_set4 = 0x44;        // S0.6, BF3 Base PWL Segment 4 = -60
+                       dscl_prog_data->easf_v_bf3_pwl_base_set4 = -60; // S0.6, BF3 Base PWL Segment 4
                        dscl_prog_data->easf_v_bf3_pwl_slope_set4 = 0x1760;     // FP1.6.6, BF3 Slope PWL Segment 4
                        dscl_prog_data->easf_v_bf3_pwl_in_set5 =
                                0x0780; // FP0.6.6, BF3 Input value PWL Segment 5 (0.5)
-                       dscl_prog_data->easf_v_bf3_pwl_base_set5 = 0x41;        // S0.6, BF3 Base PWL Segment 5 = -63
+                       dscl_prog_data->easf_v_bf3_pwl_base_set5 = -63; // S0.6, BF3 Base PWL Segment 5
                }
        } else
                dscl_prog_data->easf_v_en = false;
@@ -1327,63 +1127,52 @@ static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *s
        if (enable_easf_h) {
                dscl_prog_data->easf_h_en = true;
                dscl_prog_data->easf_h_ring = 0;
-               dscl_prog_data->easf_h_sharp_factor = 0;
+               dscl_prog_data->easf_h_sharp_factor = 1;
                dscl_prog_data->easf_h_bf1_en =
                        1;      // 1-bit, BF1 calculation enable, 0=disable, 1=enable
                dscl_prog_data->easf_h_bf2_mode =
                        0xF;    // 4-bit, BF2 calculation mode
-               /* 2-bit, BF3 chroma mode correction calculation mode */
-               dscl_prog_data->easf_h_bf3_mode = spl_get_h_bf3_mode(
-                       spl_scratch->scl_data.recip_ratios.horz);
-               /* FP1.5.10; (2.0) Ring reducer gain for 4 or 6-tap mode [H_REDUCER_GAIN4] */
+               dscl_prog_data->easf_h_bf3_mode =
+                       2;      // 2-bit, BF3 chroma mode correction calculation mode
+               dscl_prog_data->easf_h_bf2_flat1_gain = 4;      // U1.3, BF2 Flat1 Gain control
+               dscl_prog_data->easf_h_bf2_flat2_gain = 8;      // U4.0, BF2 Flat2 Gain control
+               dscl_prog_data->easf_h_bf2_roc_gain = 4;        // U2.2, Rate Of Change control
                dscl_prog_data->easf_h_ringest_eventap_reduceg1 =
-                       spl_get_reducer_gain4(spl_scratch->scl_data.taps.h_taps,
-                               spl_scratch->scl_data.recip_ratios.horz);
-               /* FP1.5.10; (2.5) Ring reducer gain for 6-tap mode [V_REDUCER_GAIN6] */
+                       0x4000; // FP1.5.10; (2.0) Ring reducer gain for 4 or 6-tap mode [H_REDUCER_GAIN4]
                dscl_prog_data->easf_h_ringest_eventap_reduceg2 =
-                       spl_get_reducer_gain6(spl_scratch->scl_data.taps.h_taps,
-                               spl_scratch->scl_data.recip_ratios.horz);
-               /* FP1.5.10; (-0.135742) Ring gain for 6-tap set to -139/1024 */
+                       0x4100; // FP1.5.10; (2.5) Ring reducer gain for 6-tap mode [V_REDUCER_GAIN6]
                dscl_prog_data->easf_h_ringest_eventap_gain1 =
-                       spl_get_gainRing4(spl_scratch->scl_data.taps.h_taps,
-                               spl_scratch->scl_data.recip_ratios.horz);
-               /* FP1.5.10; (-0.024414) Ring gain for 6-tap set to -25/1024 */
+                       0xB058; // FP1.5.10; (-0.135742) Ring gain for 6-tap set to -139/1024
                dscl_prog_data->easf_h_ringest_eventap_gain2 =
-                       spl_get_gainRing6(spl_scratch->scl_data.taps.h_taps,
-                               spl_scratch->scl_data.recip_ratios.horz);
+                       0xA640; // FP1.5.10; (-0.024414) Ring gain for 6-tap set to -25/1024
                dscl_prog_data->easf_h_bf_maxa = 63; //Horz Max BF value A in U0.6 format.Selected if H_FCNTL==0
                dscl_prog_data->easf_h_bf_maxb = 63; //Horz Max BF value B in U0.6 format.Selected if H_FCNTL==1
                dscl_prog_data->easf_h_bf_mina = 0;     //Horz Min BF value B in U0.6 format.Selected if H_FCNTL==0
                dscl_prog_data->easf_h_bf_minb = 0;     //Horz Min BF value B in U0.6 format.Selected if H_FCNTL==1
+               dscl_prog_data->easf_h_bf1_pwl_in_seg0 = -512;  // S0.10, BF1 PWL Segment 0
+               dscl_prog_data->easf_h_bf1_pwl_base_seg0 = 0;   // U0.6, BF1 Base PWL Segment 0
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg0 = 3;  // S7.3, BF1 Slope PWL Segment 0
+               dscl_prog_data->easf_h_bf1_pwl_in_seg1 = -20;   // S0.10, BF1 PWL Segment 1
+               dscl_prog_data->easf_h_bf1_pwl_base_seg1 = 12;  // U0.6, BF1 Base PWL Segment 1
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg1 = 326;        // S7.3, BF1 Slope PWL Segment 1
+               dscl_prog_data->easf_h_bf1_pwl_in_seg2 = 0;     // S0.10, BF1 PWL Segment 2
+               dscl_prog_data->easf_h_bf1_pwl_base_seg2 = 63;  // U0.6, BF1 Base PWL Segment 2
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg2 = 0;  // S7.3, BF1 Slope PWL Segment 2
+               dscl_prog_data->easf_h_bf1_pwl_in_seg3 = 16;    // S0.10, BF1 PWL Segment 3
+               dscl_prog_data->easf_h_bf1_pwl_base_seg3 = 63;  // U0.6, BF1 Base PWL Segment 3
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg3 = -56;        // S7.3, BF1 Slope PWL Segment 3
+               dscl_prog_data->easf_h_bf1_pwl_in_seg4 = 32;    // S0.10, BF1 PWL Segment 4
+               dscl_prog_data->easf_h_bf1_pwl_base_seg4 = 56;  // U0.6, BF1 Base PWL Segment 4
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg4 = -48;        // S7.3, BF1 Slope PWL Segment 4
+               dscl_prog_data->easf_h_bf1_pwl_in_seg5 = 48;    // S0.10, BF1 PWL Segment 5
+               dscl_prog_data->easf_h_bf1_pwl_base_seg5 = 50;  // U0.6, BF1 Base PWL Segment 5
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg5 = -240;       // S7.3, BF1 Slope PWL Segment 5
+               dscl_prog_data->easf_h_bf1_pwl_in_seg6 = 64;    // S0.10, BF1 PWL Segment 6
+               dscl_prog_data->easf_h_bf1_pwl_base_seg6 = 20;  // U0.6, BF1 Base PWL Segment 6
+               dscl_prog_data->easf_h_bf1_pwl_slope_seg6 = -160;       // S7.3, BF1 Slope PWL Segment 6
+               dscl_prog_data->easf_h_bf1_pwl_in_seg7 = 80;    // S0.10, BF1 PWL Segment 7
+               dscl_prog_data->easf_h_bf1_pwl_base_seg7 = 0;   // U0.6, BF1 Base PWL Segment 7
                if (lls_pref == LLS_PREF_YES)   {
-                       dscl_prog_data->easf_h_bf2_flat1_gain = 4;      // U1.3, BF2 Flat1 Gain control
-                       dscl_prog_data->easf_h_bf2_flat2_gain = 8;      // U4.0, BF2 Flat2 Gain control
-                       dscl_prog_data->easf_h_bf2_roc_gain = 4;        // U2.2, Rate Of Change control
-
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg0 = 0x600; // S0.10, BF1 PWL Segment 0 = -512
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg0 = 0;   // U0.6, BF1 Base PWL Segment 0
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg0 = 3;  // S7.3, BF1 Slope PWL Segment 0
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg1 = 0x7EC; // S0.10, BF1 PWL Segment 1 = -20
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg1 = 12;  // U0.6, BF1 Base PWL Segment 1
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg1 = 326;        // S7.3, BF1 Slope PWL Segment 1
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg2 = 0;     // S0.10, BF1 PWL Segment 2
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg2 = 63;  // U0.6, BF1 Base PWL Segment 2
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg2 = 0;  // S7.3, BF1 Slope PWL Segment 2
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg3 = 16;    // S0.10, BF1 PWL Segment 3
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg3 = 63;  // U0.6, BF1 Base PWL Segment 3
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg3 = 0x7C8;      // S7.3, BF1 Slope PWL Segment 3 = -56
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg4 = 32;    // S0.10, BF1 PWL Segment 4
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg4 = 56;  // U0.6, BF1 Base PWL Segment 4
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg4 = 0x7D0;      // S7.3, BF1 Slope PWL Segment 4 = -48
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg5 = 48;    // S0.10, BF1 PWL Segment 5
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg5 = 50;  // U0.6, BF1 Base PWL Segment 5
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg5 = 0x710;      // S7.3, BF1 Slope PWL Segment 5 = -240
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg6 = 64;    // S0.10, BF1 PWL Segment 6
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg6 = 20;  // U0.6, BF1 Base PWL Segment 6
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg6 = 0x760;      // S7.3, BF1 Slope PWL Segment 6 = -160
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg7 = 80;    // S0.10, BF1 PWL Segment 7
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg7 = 0;   // U0.6, BF1 Base PWL Segment 7
-
                        dscl_prog_data->easf_h_bf3_pwl_in_set0 = 0x000; // FP0.6.6, BF3 Input value PWL Segment 0
                        dscl_prog_data->easf_h_bf3_pwl_base_set0 = 63;  // S0.6, BF3 Base PWL Segment 0
                        dscl_prog_data->easf_h_bf3_pwl_slope_set0 = 0x12C5;     // FP1.6.6, BF3 Slope PWL Segment 0
@@ -1401,40 +1190,12 @@ static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *s
                        dscl_prog_data->easf_h_bf3_pwl_slope_set3 =     0x136B; // FP1.6.6, BF3 Slope PWL Segment 3
                        dscl_prog_data->easf_h_bf3_pwl_in_set4 =
                                0x0C37; // FP0.6.6, BF3 Input value PWL Segment 4 (0.125 * 125^3)
-                       dscl_prog_data->easf_h_bf3_pwl_base_set4 = 0x4E;        // S0.6, BF3 Base PWL Segment 4 = -50
+                       dscl_prog_data->easf_h_bf3_pwl_base_set4 = -50; // S0.6, BF3 Base PWL Segment 4
                        dscl_prog_data->easf_h_bf3_pwl_slope_set4 = 0x1200;     // FP1.6.6, BF3 Slope PWL Segment 4
                        dscl_prog_data->easf_h_bf3_pwl_in_set5 =
                                0x0CF7; // FP0.6.6, BF3 Input value PWL Segment 5 (1.0 * 125^3)
-                       dscl_prog_data->easf_h_bf3_pwl_base_set5 = 0x41;        // S0.6, BF3 Base PWL Segment 5 = -63
+                       dscl_prog_data->easf_h_bf3_pwl_base_set5 = -63; // S0.6, BF3 Base PWL Segment 5
                } else {
-                       dscl_prog_data->easf_h_bf2_flat1_gain = 13;     // U1.3, BF2 Flat1 Gain control
-                       dscl_prog_data->easf_h_bf2_flat2_gain = 15;     // U4.0, BF2 Flat2 Gain control
-                       dscl_prog_data->easf_h_bf2_roc_gain = 14;       // U2.2, Rate Of Change control
-
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg0 = 0x440; // S0.10, BF1 PWL Segment 0 = -960
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg0 = 0;   // U0.6, BF1 Base PWL Segment 0
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg0 = 2;  // S7.3, BF1 Slope PWL Segment 0
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg1 = 0x7C4; // S0.10, BF1 PWL Segment 1 = -60
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg1 = 12;  // U0.6, BF1 Base PWL Segment 1
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg1 = 109;        // S7.3, BF1 Slope PWL Segment 1
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg2 = 0;     // S0.10, BF1 PWL Segment 2
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg2 = 63;  // U0.6, BF1 Base PWL Segment 2
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg2 = 0;  // S7.3, BF1 Slope PWL Segment 2
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg3 = 48;    // S0.10, BF1 PWL Segment 3
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg3 = 63;  // U0.6, BF1 Base PWL Segment 3
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg3 = 0x7ED;      // S7.3, BF1 Slope PWL Segment 3 = -19
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg4 = 96;    // S0.10, BF1 PWL Segment 4
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg4 = 56;  // U0.6, BF1 Base PWL Segment 4
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg4 = 0x7F0;      // S7.3, BF1 Slope PWL Segment 4 = -16
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg5 = 144;   // S0.10, BF1 PWL Segment 5
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg5 = 50;  // U0.6, BF1 Base PWL Segment 5
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg5 = 0x7B0;      // S7.3, BF1 Slope PWL Segment 5 = -80
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg6 = 192;   // S0.10, BF1 PWL Segment 6
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg6 = 20;  // U0.6, BF1 Base PWL Segment 6
-                       dscl_prog_data->easf_h_bf1_pwl_slope_seg6 = 0x7CB;      // S7.3, BF1 Slope PWL Segment 6 = -53
-                       dscl_prog_data->easf_h_bf1_pwl_in_seg7 = 240;   // S0.10, BF1 PWL Segment 7
-                       dscl_prog_data->easf_h_bf1_pwl_base_seg7 = 0;   // U0.6, BF1 Base PWL Segment 7
-
                        dscl_prog_data->easf_h_bf3_pwl_in_set0 = 0x000; // FP0.6.6, BF3 Input value PWL Segment 0
                        dscl_prog_data->easf_h_bf3_pwl_base_set0 = 63;  // S0.6, BF3 Base PWL Segment 0
                        dscl_prog_data->easf_h_bf3_pwl_slope_set0 = 0x0000;     // FP1.6.6, BF3 Slope PWL Segment 0
@@ -1452,36 +1213,25 @@ static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *s
                        dscl_prog_data->easf_h_bf3_pwl_slope_set3 = 0x1878;     // FP1.6.6, BF3 Slope PWL Segment 3
                        dscl_prog_data->easf_h_bf3_pwl_in_set4 =
                                0x0761; // FP0.6.6, BF3 Input value PWL Segment 4 (0.375)
-                       dscl_prog_data->easf_h_bf3_pwl_base_set4 = 0x44;        // S0.6, BF3 Base PWL Segment 4 = -60
+                       dscl_prog_data->easf_h_bf3_pwl_base_set4 = -60; // S0.6, BF3 Base PWL Segment 4
                        dscl_prog_data->easf_h_bf3_pwl_slope_set4 = 0x1760;     // FP1.6.6, BF3 Slope PWL Segment 4
                        dscl_prog_data->easf_h_bf3_pwl_in_set5 =
                                0x0780; // FP0.6.6, BF3 Input value PWL Segment 5 (0.5)
-                       dscl_prog_data->easf_h_bf3_pwl_base_set5 = 0x41;        // S0.6, BF3 Base PWL Segment 5 = -63
+                       dscl_prog_data->easf_h_bf3_pwl_base_set5 = -63; // S0.6, BF3 Base PWL Segment 5
                } // if (lls_pref == LLS_PREF_YES)
        } else
                dscl_prog_data->easf_h_en = false;
 
        if (lls_pref == LLS_PREF_YES)   {
                dscl_prog_data->easf_ltonl_en = 1;      // Linear input
-               if (setup == HDR_L) {
-                       dscl_prog_data->easf_matrix_c0 =
-                               0x504E; // fp1.5.10, C0 coefficient (LN_BT2020:  0.2627 * (2^14)/125 = 34.43750000)
-                       dscl_prog_data->easf_matrix_c1 =
-                               0x558E; // fp1.5.10, C1 coefficient (LN_BT2020:  0.6780 * (2^14)/125 = 88.87500000)
-                       dscl_prog_data->easf_matrix_c2 =
-                               0x47C6; // fp1.5.10, C2 coefficient (LN_BT2020:  0.0593 * (2^14)/125 = 7.77343750)
-                       dscl_prog_data->easf_matrix_c3 =
-                               0x0;    // fp1.5.10, C3 coefficient
-               } else { // SDR_L
-                       dscl_prog_data->easf_matrix_c0 =
-                               0x4EF7; // fp1.5.10, C0 coefficient (LN_rec709:  0.2126 * (2^14)/125 = 27.86590720)
-                       dscl_prog_data->easf_matrix_c1 =
-                               0x55DC; // fp1.5.10, C1 coefficient (LN_rec709:  0.7152 * (2^14)/125 = 93.74269440)
-                       dscl_prog_data->easf_matrix_c2 =
-                               0x48BB; // fp1.5.10, C2 coefficient (LN_rec709:  0.0722 * (2^14)/125 = 9.46339840)
-                       dscl_prog_data->easf_matrix_c3 =
-                               0x0;    // fp1.5.10, C3 coefficient
-               }
+               dscl_prog_data->easf_matrix_c0 =
+                       0x504E; // fp1.5.10, C0 coefficient (LN_BT2020:  0.2627 * (2^14)/125 = 34.43750000)
+               dscl_prog_data->easf_matrix_c1 =
+                       0x558E; // fp1.5.10, C1 coefficient (LN_BT2020:  0.6780 * (2^14)/125 = 88.87500000)
+               dscl_prog_data->easf_matrix_c2 =
+                       0x47C6; // fp1.5.10, C2 coefficient (LN_BT2020:  0.0593 * (2^14)/125 = 7.77343750)
+               dscl_prog_data->easf_matrix_c3 =
+                       0x0;    // fp1.5.10, C3 coefficient
        }       else    {
                dscl_prog_data->easf_ltonl_en = 0;      // Non-Linear input
                dscl_prog_data->easf_matrix_c0 =
@@ -1493,43 +1243,27 @@ static void spl_set_easf_data(struct spl_scratch *spl_scratch, struct spl_out *s
                dscl_prog_data->easf_matrix_c3 =
                        0x0;    // fp1.5.10, C3 coefficient
        }
-
-       if (spl_is_yuv420(format)) { /* TODO: 0 = RGB, 1 = YUV */
-               dscl_prog_data->easf_matrix_mode = 1;
-               /*
-                * 2-bit, BF3 chroma mode correction calculation mode
-                * Needs to be disabled for YUV420 mode
-                * Override lookup value
-                */
-               dscl_prog_data->easf_v_bf3_mode = 0;
-               dscl_prog_data->easf_h_bf3_mode = 0;
-       } else
-               dscl_prog_data->easf_matrix_mode = 0;
-
 }
-
 /*Set isharp noise detection */
-static void spl_set_isharp_noise_det_mode(struct dscl_prog_data *dscl_prog_data,
-       const struct spl_scaler_data *data)
+static void spl_set_isharp_noise_det_mode(struct dscl_prog_data *dscl_prog_data)
 {
        // ISHARP_NOISEDET_MODE
        // 0: 3x5 as VxH
        // 1: 4x5 as VxH
        // 2:
        // 3: 5x5 as VxH
-       if (data->taps.v_taps == 6)
-               dscl_prog_data->isharp_noise_det.mode = 3;
-       else if (data->taps.v_taps == 4)
-               dscl_prog_data->isharp_noise_det.mode = 1;
-       else if (data->taps.v_taps == 3)
-               dscl_prog_data->isharp_noise_det.mode = 0;
+       if (dscl_prog_data->taps.v_taps == 6)
+               dscl_prog_data->isharp_noise_det.mode = 3;      // ISHARP_NOISEDET_MODE
+       else if (dscl_prog_data->taps.h_taps == 4)
+               dscl_prog_data->isharp_noise_det.mode = 1;      // ISHARP_NOISEDET_MODE
+       else if (dscl_prog_data->taps.h_taps == 3)
+               dscl_prog_data->isharp_noise_det.mode = 0;      // ISHARP_NOISEDET_MODE
 };
 /* Set Sharpener data */
 static void spl_set_isharp_data(struct dscl_prog_data *dscl_prog_data,
                struct adaptive_sharpness adp_sharpness, bool enable_isharp,
                enum linear_light_scaling lls_pref, enum spl_pixel_format format,
-               const struct spl_scaler_data *data, struct spl_fixed31_32 ratio,
-               enum system_setup setup)
+               const struct spl_scaler_data *data)
 {
        /* Turn off sharpener if not required */
        if (!enable_isharp) {
@@ -1538,12 +1272,10 @@ static void spl_set_isharp_data(struct dscl_prog_data *dscl_prog_data,
        }
 
        dscl_prog_data->isharp_en = 1;  // ISHARP_EN
+       dscl_prog_data->isharp_noise_det.enable = 1;    // ISHARP_NOISEDET_EN
        // Set ISHARP_NOISEDET_MODE if htaps = 6-tap
-       if (data->taps.h_taps == 6) {
-               dscl_prog_data->isharp_noise_det.enable = 1;    /* ISHARP_NOISEDET_EN */
-               spl_set_isharp_noise_det_mode(dscl_prog_data, data);    /* ISHARP_NOISEDET_MODE */
-       } else
-               dscl_prog_data->isharp_noise_det.enable = 0;    // ISHARP_NOISEDET_EN
+       if (dscl_prog_data->taps.h_taps == 6)
+               spl_set_isharp_noise_det_mode(dscl_prog_data);  // ISHARP_NOISEDET_MODE
        // Program noise detection threshold
        dscl_prog_data->isharp_noise_det.uthreshold = 24;       // ISHARP_NOISEDET_UTHRE
        dscl_prog_data->isharp_noise_det.dthreshold = 4;        // ISHARP_NOISEDET_DTHRE
@@ -1552,67 +1284,50 @@ static void spl_set_isharp_data(struct dscl_prog_data *dscl_prog_data,
        dscl_prog_data->isharp_noise_det.pwl_end_in = 13;       // ISHARP_NOISEDET_PWL_END_IN
        dscl_prog_data->isharp_noise_det.pwl_slope = 1623;      // ISHARP_NOISEDET_PWL_SLOPE
 
-       if (lls_pref == LLS_PREF_NO) /* ISHARP_FMT_MODE */
+       if ((lls_pref == LLS_PREF_NO) && !spl_is_yuv420(format)) /* ISHARP_FMT_MODE */
                dscl_prog_data->isharp_fmt.mode = 1;
        else
                dscl_prog_data->isharp_fmt.mode = 0;
 
        dscl_prog_data->isharp_fmt.norm = 0x3C00;       // ISHARP_FMT_NORM
        dscl_prog_data->isharp_lba.mode = 0;    // ISHARP_LBA_MODE
-       if (setup == SDR_L) {
-               // ISHARP_LBA_PWL_SEG0: ISHARP Local Brightness Adjustment PWL Segment 0
-               dscl_prog_data->isharp_lba.in_seg[0] = 0;       // ISHARP LBA PWL for Seg 0. INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[0] = 0;     // ISHARP LBA PWL for Seg 0. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[0] = 62;   // ISHARP LBA for Seg 0. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG1: ISHARP LBA PWL Segment 1
-               dscl_prog_data->isharp_lba.in_seg[1] = 130;     // ISHARP LBA PWL for Seg 1. INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[1] = 63; // ISHARP LBA PWL for Seg 1. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[1] = 0; // ISHARP LBA for Seg 1. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG2: ISHARP LBA PWL Segment 2
-               dscl_prog_data->isharp_lba.in_seg[2] = 312; // ISHARP LBA PWL for Seg 2. INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[2] = 63; // ISHARP LBA PWL for Seg 2. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[2] = 0x1D9; // ISHARP LBA for Seg 2. SLOPE value in S5.3 format = -39
-               // ISHARP_LBA_PWL_SEG3: ISHARP LBA PWL Segment 3
-               dscl_prog_data->isharp_lba.in_seg[3] = 520; // ISHARP LBA PWL for Seg 3.INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[3] = 0; // ISHARP LBA PWL for Seg 3. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[3] = 0; // ISHARP LBA for Seg 3. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG4: ISHARP LBA PWL Segment 4
-               dscl_prog_data->isharp_lba.in_seg[4] = 520; // ISHARP LBA PWL for Seg 4.INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[4] = 0; // ISHARP LBA PWL for Seg 4. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[4] = 0; // ISHARP LBA for Seg 4. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG5: ISHARP LBA PWL Segment 5
-               dscl_prog_data->isharp_lba.in_seg[5] = 520; // ISHARP LBA PWL for Seg 5.INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[5] = 0;     // ISHARP LBA PWL for Seg 5. BASE value in U0.6 format
-       } else {
-               // ISHARP_LBA_PWL_SEG0: ISHARP Local Brightness Adjustment PWL Segment 0
-               dscl_prog_data->isharp_lba.in_seg[0] = 0;       // ISHARP LBA PWL for Seg 0. INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[0] = 0;     // ISHARP LBA PWL for Seg 0. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[0] = 32;   // ISHARP LBA for Seg 0. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG1: ISHARP LBA PWL Segment 1
-               dscl_prog_data->isharp_lba.in_seg[1] = 256;     // ISHARP LBA PWL for Seg 1. INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[1] = 63; // ISHARP LBA PWL for Seg 1. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[1] = 0; // ISHARP LBA for Seg 1. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG2: ISHARP LBA PWL Segment 2
-               dscl_prog_data->isharp_lba.in_seg[2] = 614; // ISHARP LBA PWL for Seg 2. INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[2] = 63; // ISHARP LBA PWL for Seg 2. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[2] = 0x1EC; // ISHARP LBA for Seg 2. SLOPE value in S5.3 format = -20
-               // ISHARP_LBA_PWL_SEG3: ISHARP LBA PWL Segment 3
-               dscl_prog_data->isharp_lba.in_seg[3] = 1023; // ISHARP LBA PWL for Seg 3.INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[3] = 0; // ISHARP LBA PWL for Seg 3. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[3] = 0; // ISHARP LBA for Seg 3. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG4: ISHARP LBA PWL Segment 4
-               dscl_prog_data->isharp_lba.in_seg[4] = 1023; // ISHARP LBA PWL for Seg 4.INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[4] = 0; // ISHARP LBA PWL for Seg 4. BASE value in U0.6 format
-               dscl_prog_data->isharp_lba.slope_seg[4] = 0; // ISHARP LBA for Seg 4. SLOPE value in S5.3 format
-               // ISHARP_LBA_PWL_SEG5: ISHARP LBA PWL Segment 5
-               dscl_prog_data->isharp_lba.in_seg[5] = 1023; // ISHARP LBA PWL for Seg 5.INPUT value in U0.10 format
-               dscl_prog_data->isharp_lba.base_seg[5] = 0;     // ISHARP LBA PWL for Seg 5. BASE value in U0.6 format
+       // ISHARP_LBA_PWL_SEG0: ISHARP Local Brightness Adjustment PWL Segment 0
+       dscl_prog_data->isharp_lba.in_seg[0] = 0;       // ISHARP LBA PWL for Seg 0. INPUT value in U0.10 format
+       dscl_prog_data->isharp_lba.base_seg[0] = 0;     // ISHARP LBA PWL for Seg 0. BASE value in U0.6 format
+       dscl_prog_data->isharp_lba.slope_seg[0] = 32;   // ISHARP LBA for Seg 0. SLOPE value in S5.3 format
+       // ISHARP_LBA_PWL_SEG1: ISHARP LBA PWL Segment 1
+       dscl_prog_data->isharp_lba.in_seg[1] = 256;     // ISHARP LBA PWL for Seg 1. INPUT value in U0.10 format
+       dscl_prog_data->isharp_lba.base_seg[1] = 63; // ISHARP LBA PWL for Seg 1. BASE value in U0.6 format
+       dscl_prog_data->isharp_lba.slope_seg[1] = 0; // ISHARP LBA for Seg 1. SLOPE value in S5.3 format
+       // ISHARP_LBA_PWL_SEG2: ISHARP LBA PWL Segment 2
+       dscl_prog_data->isharp_lba.in_seg[2] = 614; // ISHARP LBA PWL for Seg 2. INPUT value in U0.10 format
+       dscl_prog_data->isharp_lba.base_seg[2] = 63; // ISHARP LBA PWL for Seg 2. BASE value in U0.6 format
+       dscl_prog_data->isharp_lba.slope_seg[2] = -20; // ISHARP LBA for Seg 2. SLOPE value in S5.3 format
+       // ISHARP_LBA_PWL_SEG3: ISHARP LBA PWL Segment 3
+       dscl_prog_data->isharp_lba.in_seg[3] = 1023; // ISHARP LBA PWL for Seg 3.INPUT value in U0.10 format
+       dscl_prog_data->isharp_lba.base_seg[3] = 0; // ISHARP LBA PWL for Seg 3. BASE value in U0.6 format
+       dscl_prog_data->isharp_lba.slope_seg[3] = 0; // ISHARP LBA for Seg 3. SLOPE value in S5.3 format
+       // ISHARP_LBA_PWL_SEG4: ISHARP LBA PWL Segment 4
+       dscl_prog_data->isharp_lba.in_seg[4] = 1023; // ISHARP LBA PWL for Seg 4.INPUT value in U0.10 format
+       dscl_prog_data->isharp_lba.base_seg[4] = 0; // ISHARP LBA PWL for Seg 4. BASE value in U0.6 format
+       dscl_prog_data->isharp_lba.slope_seg[4] = 0; // ISHARP LBA for Seg 4. SLOPE value in S5.3 format
+       // ISHARP_LBA_PWL_SEG5: ISHARP LBA PWL Segment 5
+       dscl_prog_data->isharp_lba.in_seg[5] = 1023; // ISHARP LBA PWL for Seg 5.INPUT value in U0.10 format
+       dscl_prog_data->isharp_lba.base_seg[5] = 0;     // ISHARP LBA PWL for Seg 5. BASE value in U0.6 format
+       switch (adp_sharpness.sharpness) {
+       case SHARPNESS_LOW:
+               dscl_prog_data->isharp_delta = spl_get_filter_isharp_1D_lut_0p5x();
+               break;
+       case SHARPNESS_MID:
+               dscl_prog_data->isharp_delta = spl_get_filter_isharp_1D_lut_1p0x();
+               break;
+       case SHARPNESS_HIGH:
+               dscl_prog_data->isharp_delta = spl_get_filter_isharp_1D_lut_2p0x();
+               break;
+       default:
+               BREAK_TO_DEBUGGER();
        }
 
-       spl_build_isharp_1dlut_from_reference_curve(ratio, setup);
-       dscl_prog_data->isharp_delta = spl_get_pregen_filter_isharp_1D_lut(
-               adp_sharpness.sharpness);
-
        // Program the nldelta soft clip values
        if (lls_pref == LLS_PREF_YES) {
                dscl_prog_data->isharp_nldelta_sclip.enable_p = 0;      /* ISHARP_NLDELTA_SCLIP_EN_P */
@@ -1631,7 +1346,62 @@ static void spl_set_isharp_data(struct dscl_prog_data *dscl_prog_data,
        }
 
        // Set the values as per lookup table
+#ifdef CONFIG_DRM_AMD_DC_FP
        spl_set_blur_scale_data(dscl_prog_data, data);
+#endif
+}
+static bool spl_get_isharp_en(struct adaptive_sharpness adp_sharpness,
+               int vscale_ratio, int hscale_ratio, struct spl_taps taps,
+               enum spl_pixel_format format)
+{
+       bool enable_isharp = false;
+
+       if (adp_sharpness.enable == false)
+               return enable_isharp; // Return if adaptive sharpness is disabled
+       // Is downscaling ?
+       if (vscale_ratio > 1 || hscale_ratio > 1) {
+               // END - No iSHARP support for downscaling
+               return enable_isharp;
+       }
+       // Scaling is up to 1:1 (no scaling) or upscaling
+
+       /* Only apply sharpness to NV12 and not P010 */
+       if (format != SPL_PIXEL_FORMAT_420BPP8)
+               return enable_isharp;
+
+       // LB support horizontal taps 4,6 or vertical taps 3, 4, 6
+       if (taps.h_taps == 4 || taps.h_taps == 6 ||
+               taps.v_taps == 3 || taps.v_taps == 4 || taps.v_taps == 6) {
+               // END - iSHARP supported
+               enable_isharp = true;
+       }
+       return enable_isharp;
+}
+
+static bool spl_choose_lls_policy(enum spl_pixel_format format,
+       enum spl_transfer_func_type tf_type,
+       enum spl_transfer_func_predefined tf_predefined_type,
+       enum linear_light_scaling *lls_pref)
+{
+       if (spl_is_yuv420(format)) {
+               *lls_pref = LLS_PREF_NO;
+               if ((tf_type == SPL_TF_TYPE_PREDEFINED) || (tf_type == SPL_TF_TYPE_DISTRIBUTED_POINTS))
+                       return true;
+       } else { /* RGB or YUV444 */
+               if (tf_type == SPL_TF_TYPE_PREDEFINED) {
+                       if ((tf_predefined_type == SPL_TRANSFER_FUNCTION_HLG) ||
+                               (tf_predefined_type == SPL_TRANSFER_FUNCTION_HLG12))
+                               *lls_pref = LLS_PREF_NO;
+                       else
+                               *lls_pref = LLS_PREF_YES;
+                       return true;
+               } else if (tf_type == SPL_TF_TYPE_BYPASS) {
+                       *lls_pref = LLS_PREF_YES;
+                       return true;
+               }
+       }
+       *lls_pref = LLS_PREF_NO;
+       return false;
 }
 
 /* Calculate scaler parameters */
@@ -1640,71 +1410,67 @@ bool spl_calculate_scaler_params(struct spl_in *spl_in, struct spl_out *spl_out)
        bool res = false;
        bool enable_easf_v = false;
        bool enable_easf_h = false;
+       bool lls_enable_easf = true;
        int vratio = 0;
        int hratio = 0;
-       struct spl_scratch spl_scratch;
-       struct spl_fixed31_32 isharp_scale_ratio;
-       enum system_setup setup;
-       bool enable_isharp = false;
-       const struct spl_scaler_data *data = &spl_scratch.scl_data;
-
-       memset(&spl_scratch, 0, sizeof(struct spl_scratch));
-       spl_scratch.scl_data.h_active = spl_in->h_active;
-       spl_scratch.scl_data.v_active = spl_in->v_active;
+       const struct spl_scaler_data *data = &spl_out->scl_data;
        // All SPL calls
        /* recout calculation */
        /* depends on h_active */
-       spl_calculate_recout(spl_in, &spl_scratch, spl_out);
+       spl_calculate_recout(spl_in, spl_out);
        /* depends on pixel format */
-       spl_calculate_scaling_ratios(spl_in, &spl_scratch, spl_out);
+       spl_calculate_scaling_ratios(spl_in, spl_out);
        /* depends on scaling ratios and recout, does not calculate offset yet */
-       spl_calculate_viewport_size(spl_in, &spl_scratch);
+       spl_calculate_viewport_size(spl_in, spl_out);
 
        res = spl_get_optimal_number_of_taps(
                          spl_in->basic_out.max_downscale_src_width, spl_in,
-                         &spl_scratch, &spl_in->scaling_quality, &enable_easf_v,
-                         &enable_easf_h, &enable_isharp);
+                         spl_out, &spl_in->scaling_quality);
        /*
         * Depends on recout, scaling ratios, h_active and taps
         * May need to re-check lb size after this in some obscure scenario
         */
        if (res)
-               spl_calculate_inits_and_viewports(spl_in, &spl_scratch);
+               spl_calculate_inits_and_viewports(spl_in, spl_out);
        // Handle 3d recout
-       spl_handle_3d_recout(spl_in, &spl_scratch.scl_data.recout);
+       spl_handle_3d_recout(spl_in, &spl_out->scl_data.recout);
        // Clamp
-       spl_clamp_viewport(&spl_scratch.scl_data.viewport);
+       spl_clamp_viewport(&spl_out->scl_data.viewport);
 
        if (!res)
                return res;
 
+       /*
+        * If lls_pref is LLS_PREF_DONT_CARE, then use pixel format and transfer
+        *  function to determine whether to use LINEAR or NONLINEAR scaling
+        */
+       if (spl_in->lls_pref == LLS_PREF_DONT_CARE)
+               lls_enable_easf = spl_choose_lls_policy(spl_in->basic_in.format,
+                       spl_in->basic_in.tf_type, spl_in->basic_in.tf_predefined_type,
+                       &spl_in->lls_pref);
+
        // Save all calculated parameters in dscl_prog_data structure to program hw registers
-       spl_set_dscl_prog_data(spl_in, &spl_scratch, spl_out, enable_easf_v, enable_easf_h, enable_isharp);
+       spl_set_dscl_prog_data(spl_in, spl_out);
 
-       if (spl_in->lls_pref == LLS_PREF_YES) {
-               if (spl_in->is_hdr_on)
-                       setup = HDR_L;
-               else
-                       setup = SDR_L;
+       vratio = dc_fixpt_ceil(spl_out->scl_data.ratios.vert);
+       hratio = dc_fixpt_ceil(spl_out->scl_data.ratios.horz);
+       if (!lls_enable_easf || spl_in->disable_easf) {
+               enable_easf_v = false;
+               enable_easf_h = false;
        } else {
-               if (spl_in->is_hdr_on)
-                       setup = HDR_NL;
-               else
-                       setup = SDR_NL;
+               /* Enable EASF on vertical? */
+               enable_easf_v = enable_easf(vratio, spl_out->scl_data.taps.v_taps, spl_in->lls_pref, spl_in->prefer_easf);
+               /* Enable EASF on horizontal? */
+               enable_easf_h = enable_easf(hratio, spl_out->scl_data.taps.h_taps, spl_in->lls_pref, spl_in->prefer_easf);
        }
        // Set EASF
-       spl_set_easf_data(&spl_scratch, spl_out, enable_easf_v, enable_easf_h, spl_in->lls_pref,
-               spl_in->basic_in.format, setup);
+       spl_set_easf_data(spl_out->dscl_prog_data, enable_easf_v, enable_easf_h, spl_in->lls_pref,
+               spl_in->basic_in.format);
        // Set iSHARP
-       vratio = spl_fixpt_ceil(spl_scratch.scl_data.ratios.vert);
-       hratio = spl_fixpt_ceil(spl_scratch.scl_data.ratios.horz);
-       if (vratio <= hratio)
-               isharp_scale_ratio = spl_scratch.scl_data.recip_ratios.vert;
-       else
-               isharp_scale_ratio = spl_scratch.scl_data.recip_ratios.horz;
-
+       bool enable_isharp = spl_get_isharp_en(spl_in->adaptive_sharpness, vratio, hratio,
+               spl_out->scl_data.taps, spl_in->basic_in.format);
        spl_set_isharp_data(spl_out->dscl_prog_data, spl_in->adaptive_sharpness, enable_isharp,
-               spl_in->lls_pref, spl_in->basic_in.format, data, isharp_scale_ratio, setup);
+               spl_in->lls_pref, spl_in->basic_in.format, data);
 
        return res;
 }
diff --git a/drivers/gpu/drm/amd/display/dc/spl/dc_spl_filters.c b/drivers/gpu/drm/amd/display/dc/spl/dc_spl_filters.c
deleted file mode 100644 (file)
index 9923864..0000000
+++ /dev/null
@@ -1,15 +0,0 @@
-// SPDX-License-Identifier: MIT
-//
-// Copyright 2024 Advanced Micro Devices, Inc.
-
-#include "dc_spl_filters.h"
-
-void convert_filter_s1_10_to_s1_12(const uint16_t *s1_10_filter,
-       uint16_t *s1_12_filter, int num_taps)
-{
-       int num_entries = NUM_PHASES_COEFF * num_taps;
-       int i;
-
-       for (i = 0; i < num_entries; i++)
-               *(s1_12_filter + i) = *(s1_10_filter + i) * 4;
-}
diff --git a/drivers/gpu/drm/amd/display/dc/spl/dc_spl_filters.h b/drivers/gpu/drm/amd/display/dc/spl/dc_spl_filters.h
deleted file mode 100644 (file)
index 20439cd..0000000
+++ /dev/null
@@ -1,15 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-
-/* Copyright 2024 Advanced Micro Devices, Inc. */
-
-#ifndef __DC_SPL_FILTERS_H__
-#define __DC_SPL_FILTERS_H__
-
-#include "dc_spl_types.h"
-
-#define NUM_PHASES_COEFF 33
-
-void convert_filter_s1_10_to_s1_12(const uint16_t *s1_10_filter,
-       uint16_t *s1_12_filter, int num_taps);
-
-#endif /* __DC_SPL_FILTERS_H__ */
index d483f259512ed0acd886a4fe1598eba147831d6b..a5d9a6223d0619013c6182a06bbb47508c79dc03 100644 (file)
@@ -2,9 +2,6 @@
 //
 // Copyright 2024 Advanced Micro Devices, Inc.
 
-#include "dc_spl_types.h"
-#include "spl_debug.h"
-#include "dc_spl_filters.h"
 #include "dc_spl_isharp_filters.h"
 
 //========================================
@@ -233,53 +230,6 @@ static const uint32_t filter_isharp_1D_lut_2p0x[32] = {
 0x080B0D0E,
 0x00020406,
 };
-//========================================
-// Delta Gain 1DLUT
-// LUT content is packed as 4-bytes into one DWORD/entry
-// A_start = 0.000000
-// A_end   = 10.000000
-// A_gain  = 3.000000
-// B_start = 11.000000
-// B_end   = 127.000000
-// C_start = 40.000000
-// C_end   = 127.000000
-//========================================
-static const uint32_t filter_isharp_1D_lut_3p0x[32] = {
-0x03010000,
-0x0F0B0805,
-0x211E1813,
-0x2B292624,
-0x3533302E,
-0x3E3C3A37,
-0x46444240,
-0x4D4B4A48,
-0x5352504F,
-0x59575655,
-0x5D5C5B5A,
-0x61605F5E,
-0x64646362,
-0x66666565,
-0x68686767,
-0x68686868,
-0x68686868,
-0x67676868,
-0x65656666,
-0x62636464,
-0x5E5F6061,
-0x5A5B5C5D,
-0x55565759,
-0x4F505253,
-0x484A4B4D,
-0x40424446,
-0x373A3C3E,
-0x2E303335,
-0x2426292B,
-0x191B1E21,
-0x0D101316,
-0x0003060A,
-};
-
-//========================================
 // Wide scaler coefficients
 //========================================================
 // <using>                     gen_scaler_coeffs.m
@@ -334,7 +284,7 @@ static const uint16_t filter_isharp_wide_6tap_64p[198] = {
 // <CoefType>          Blur & Scale LPF
 // <CoefQuant>         S1.10
 //========================================================
-static const uint16_t filter_isharp_bs_4tap_in_6_64p[198] = {
+static const uint16_t filter_isharp_bs_4tap_64p[198] = {
 0x0000, 0x00E5, 0x0237, 0x00E4, 0x0000, 0x0000,
 0x0000, 0x00DE, 0x0237, 0x00EB, 0x0000, 0x0000,
 0x0000, 0x00D7, 0x0236, 0x00F2, 0x0001, 0x0000,
@@ -369,228 +319,6 @@ static const uint16_t filter_isharp_bs_4tap_in_6_64p[198] = {
 0x0000, 0x003B, 0x01CF, 0x01C2, 0x0034, 0x0000,
 0x0000, 0x0037, 0x01C9, 0x01C9, 0x0037, 0x0000
 };
-//========================================================
-// <using>                     gen_BlurScale_coeffs.m
-// <date>                      25-Apr-2022
-// <num_taps>          4
-// <num_phases>                64
-// <CoefType>          Blur & Scale LPF
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t filter_isharp_bs_4tap_64p[132] = {
-0x00E5, 0x0237, 0x00E4, 0x0000,
-0x00DE, 0x0237, 0x00EB, 0x0000,
-0x00D7, 0x0236, 0x00F2, 0x0001,
-0x00D0, 0x0235, 0x00FA, 0x0001,
-0x00C9, 0x0234, 0x0101, 0x0002,
-0x00C2, 0x0233, 0x0108, 0x0003,
-0x00BB, 0x0232, 0x0110, 0x0003,
-0x00B5, 0x0230, 0x0117, 0x0004,
-0x00AE, 0x022E, 0x011F, 0x0005,
-0x00A8, 0x022C, 0x0126, 0x0006,
-0x00A2, 0x022A, 0x012D, 0x0007,
-0x009C, 0x0228, 0x0134, 0x0008,
-0x0096, 0x0225, 0x013C, 0x0009,
-0x0090, 0x0222, 0x0143, 0x000B,
-0x008A, 0x021F, 0x014B, 0x000C,
-0x0085, 0x021C, 0x0151, 0x000E,
-0x007F, 0x0218, 0x015A, 0x000F,
-0x007A, 0x0215, 0x0160, 0x0011,
-0x0074, 0x0211, 0x0168, 0x0013,
-0x006F, 0x020D, 0x016F, 0x0015,
-0x006A, 0x0209, 0x0176, 0x0017,
-0x0065, 0x0204, 0x017E, 0x0019,
-0x0060, 0x0200, 0x0185, 0x001B,
-0x005C, 0x01FB, 0x018C, 0x001D,
-0x0057, 0x01F6, 0x0193, 0x0020,
-0x0053, 0x01F1, 0x019A, 0x0022,
-0x004E, 0x01EC, 0x01A1, 0x0025,
-0x004A, 0x01E6, 0x01A8, 0x0028,
-0x0046, 0x01E1, 0x01AF, 0x002A,
-0x0042, 0x01DB, 0x01B6, 0x002D,
-0x003F, 0x01D5, 0x01BB, 0x0031,
-0x003B, 0x01CF, 0x01C2, 0x0034,
-0x0037, 0x01C9, 0x01C9, 0x0037,
-};
-//========================================================
-// <using>                     gen_BlurScale_coeffs.m
-// <date>                      09-Jun-2022
-// <num_taps>          3
-// <num_phases>                64
-// <CoefType>          Blur & Scale LPF
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t filter_isharp_bs_3tap_64p[99] = {
-0x0200, 0x0200, 0x0000,
-0x01F6, 0x0206, 0x0004,
-0x01EC, 0x020B, 0x0009,
-0x01E2, 0x0211, 0x000D,
-0x01D8, 0x0216, 0x0012,
-0x01CE, 0x021C, 0x0016,
-0x01C4, 0x0221, 0x001B,
-0x01BA, 0x0226, 0x0020,
-0x01B0, 0x022A, 0x0026,
-0x01A6, 0x022F, 0x002B,
-0x019C, 0x0233, 0x0031,
-0x0192, 0x0238, 0x0036,
-0x0188, 0x023C, 0x003C,
-0x017E, 0x0240, 0x0042,
-0x0174, 0x0244, 0x0048,
-0x016A, 0x0248, 0x004E,
-0x0161, 0x024A, 0x0055,
-0x0157, 0x024E, 0x005B,
-0x014D, 0x0251, 0x0062,
-0x0144, 0x0253, 0x0069,
-0x013A, 0x0256, 0x0070,
-0x0131, 0x0258, 0x0077,
-0x0127, 0x025B, 0x007E,
-0x011E, 0x025C, 0x0086,
-0x0115, 0x025E, 0x008D,
-0x010B, 0x0260, 0x0095,
-0x0102, 0x0262, 0x009C,
-0x00F9, 0x0263, 0x00A4,
-0x00F0, 0x0264, 0x00AC,
-0x00E7, 0x0265, 0x00B4,
-0x00DF, 0x0264, 0x00BD,
-0x00D6, 0x0265, 0x00C5,
-0x00CD, 0x0266, 0x00CD,
-};
-
-/* Converted Blur & Scale coeff tables from S1.10 to S1.12 */
-static uint16_t filter_isharp_bs_4tap_in_6_64p_s1_12[198];
-static uint16_t filter_isharp_bs_4tap_64p_s1_12[132];
-static uint16_t filter_isharp_bs_3tap_64p_s1_12[99];
-
-struct scale_ratio_to_sharpness_level_lookup scale_to_sharp_sdr_nl[3][6] = {
-       { /* LOW */
-               {1125, 1000, 75, 100},
-               {11, 10, 6, 10},
-               {1075, 1000, 45, 100},
-               {105, 100, 3, 10},
-               {1025, 1000, 15, 100},
-               {1, 1, 0, 1},
-       },
-       { /* MID */
-               {1125, 1000, 2, 1},
-               {11, 10, 175, 100},
-               {1075, 1000, 15, 10},
-               {105, 100, 125, 100},
-               {1025, 1000, 1, 1},
-               {1, 1, 75, 100},
-       },
-       { /* HIGH */
-               {1125, 1000, 35, 10},
-               {11, 10, 32, 10},
-               {1075, 1000, 29, 10},
-               {105, 100, 26, 10},
-               {1025, 1000, 23, 10},
-               {1, 1, 2, 1},
-       },
-};
-
-struct scale_ratio_to_sharpness_level_lookup scale_to_sharp_sdr_l[3][6] = {
-       { /* LOW */
-               {1125, 1000, 75, 100},
-               {11, 10, 6, 10},
-               {1075, 1000, 45, 100},
-               {105, 100, 3, 10},
-               {1025, 1000, 15, 100},
-               {1, 1, 0, 1},
-       },
-       { /* MID */
-               {1125, 1000, 15, 10},
-               {11, 10, 135, 100},
-               {1075, 1000, 12, 10},
-               {105, 100, 105, 100},
-               {1025, 1000, 9, 10},
-               {1, 1, 75, 100},
-       },
-       { /* HIGH */
-               {1125, 1000, 25, 10},
-               {11, 10, 23, 10},
-               {1075, 1000, 21, 10},
-               {105, 100, 19, 10},
-               {1025, 1000, 17, 10},
-               {1, 1, 15, 10},
-       },
-};
-
-struct scale_ratio_to_sharpness_level_lookup scale_to_sharp_hdr_nl[3][6] = {
-       { /* LOW */
-               {1125, 1000, 5, 10},
-               {11, 10, 4, 10},
-               {1075, 1000, 3, 10},
-               {105, 100, 2, 10},
-               {1025, 1000, 1, 10},
-               {1, 1, 0, 1},
-       },
-       { /* MID */
-               {1125, 1000, 1, 1},
-               {11, 10, 9, 10},
-               {1075, 1000, 8, 10},
-               {105, 100, 7, 10},
-               {1025, 1000, 6, 10},
-               {1, 1, 5, 10},
-       },
-       { /* HIGH */
-               {1125, 1000, 15, 10},
-               {11, 10, 14, 10},
-               {1075, 1000, 13, 10},
-               {105, 100, 12, 10},
-               {1025, 1000, 11, 10},
-               {1, 1, 1, 1},
-       },
-};
-
-struct scale_ratio_to_sharpness_level_lookup scale_to_sharp_hdr_l[3][6] = {
-       { /* LOW */
-               {1125, 1000, 75, 100},
-               {11, 10, 6, 10},
-               {1075, 1000, 45, 100},
-               {105, 100, 3, 10},
-               {1025, 1000, 15, 100},
-               {1, 1, 0, 1},
-       },
-       { /* MID */
-               {1125, 1000, 15, 10},
-               {11, 10, 135, 100},
-               {1075, 1000, 12, 10},
-               {105, 100, 105, 100},
-               {1025, 1000, 9, 10},
-               {1, 1, 75, 100},
-       },
-       { /* HIGH */
-               {1125, 1000, 25, 10},
-               {11, 10, 23, 10},
-               {1075, 1000, 21, 10},
-               {105, 100, 19, 10},
-               {1025, 1000, 17, 10},
-               {1, 1, 15, 10},
-       },
-};
-
-/* Pre-generated 1DLUT for LOW for given setup and sharpness level */
-uint32_t filter_isharp_1D_lut_pregen[3][32] = {
-       {
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-       },
-       {
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-       },
-       {
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-               0, 0, 0, 0, 0, 0, 0, 0,
-       },
-};
-
 const uint32_t *spl_get_filter_isharp_1D_lut_0(void)
 {
        return filter_isharp_1D_lut_0;
@@ -611,160 +339,11 @@ const uint32_t *spl_get_filter_isharp_1D_lut_2p0x(void)
 {
        return filter_isharp_1D_lut_2p0x;
 }
-const uint32_t *spl_get_filter_isharp_1D_lut_3p0x(void)
-{
-       return filter_isharp_1D_lut_3p0x;
-}
 const uint16_t *spl_get_filter_isharp_wide_6tap_64p(void)
 {
        return filter_isharp_wide_6tap_64p;
 }
-uint16_t *spl_get_filter_isharp_bs_4tap_in_6_64p(void)
-{
-       return filter_isharp_bs_4tap_in_6_64p_s1_12;
-}
-uint16_t *spl_get_filter_isharp_bs_4tap_64p(void)
-{
-       return filter_isharp_bs_4tap_64p_s1_12;
-}
-uint16_t *spl_get_filter_isharp_bs_3tap_64p(void)
-{
-       return filter_isharp_bs_3tap_64p_s1_12;
-}
-
-void spl_build_isharp_1dlut_from_reference_curve(struct spl_fixed31_32 ratio, enum system_setup setup)
-{
-       uint8_t *byte_ptr_1dlut_src, *byte_ptr_1dlut_dst;
-       struct spl_fixed31_32 sharp_base, sharp_calc, sharp_level, ratio_level;
-       int i, j;
-       struct scale_ratio_to_sharpness_level_lookup *setup_lookup_ptr;
-       int num_sharp_ramp_levels;
-       int size_1dlut;
-       int sharp_calc_int;
-       uint32_t filter_pregen_store[32];
-
-       /*
-        * Given scaling ratio and current system setup, build pregenerated
-        * 1DLUT tables for three sharpness levels - LOW, MID, HIGH
-        */
-       for (i = 0; i < 3; i++) {
-               /*
-                * Based on setup ( HDR/SDR, L/NL ), get base scale ratio to
-                *  sharpness curve
-                */
-               switch (setup) {
-               case HDR_L:
-                       setup_lookup_ptr = scale_to_sharp_hdr_l[i];
-                       num_sharp_ramp_levels = sizeof(scale_to_sharp_hdr_l[i])/
-                               sizeof(struct scale_ratio_to_sharpness_level_lookup);
-                       break;
-               case HDR_NL:
-                       setup_lookup_ptr = scale_to_sharp_hdr_nl[i];
-                       num_sharp_ramp_levels = sizeof(scale_to_sharp_hdr_nl[i])/
-                               sizeof(struct scale_ratio_to_sharpness_level_lookup);
-                       break;
-               case SDR_L:
-                       setup_lookup_ptr = scale_to_sharp_sdr_l[i];
-                       num_sharp_ramp_levels = sizeof(scale_to_sharp_sdr_l[i])/
-                               sizeof(struct scale_ratio_to_sharpness_level_lookup);
-                       break;
-               case SDR_NL:
-               default:
-                       setup_lookup_ptr = scale_to_sharp_sdr_nl[i];
-                       num_sharp_ramp_levels = sizeof(scale_to_sharp_sdr_nl[i])/
-                               sizeof(struct scale_ratio_to_sharpness_level_lookup);
-                       break;
-               }
-
-               /*
-                * Compare desired scaling ratio and find adjusted sharpness from
-                *  base scale ratio to sharpness curve
-                */
-               j = 0;
-               sharp_level = spl_fixpt_zero;
-               while (j < num_sharp_ramp_levels) {
-                       ratio_level = spl_fixpt_from_fraction(setup_lookup_ptr->ratio_numer,
-                               setup_lookup_ptr->ratio_denom);
-                       if (ratio.value >= ratio_level.value) {
-                               sharp_level = spl_fixpt_from_fraction(setup_lookup_ptr->sharpness_numer,
-                                       setup_lookup_ptr->sharpness_denom);
-                               break;
-                       }
-                       setup_lookup_ptr++;
-                       j++;
-               }
-
-               /*
-                * Calculate LUT_128_gained with this equation:
-                *
-                * LUT_128_gained[i] = (uint8)(0.5 + min(255,(double)(LUT_128[i])*sharpLevel/iGain))
-                *  where LUT_128[i] is contents of 3p0x isharp 1dlut
-                *  where sharpLevel is desired sharpness level
-                *  where iGain is base sharpness level 3.0
-                *  where LUT_128_gained[i] is adjusted 1dlut value based on desired sharpness level
-                */
-               byte_ptr_1dlut_src = (uint8_t *)filter_isharp_1D_lut_3p0x;
-               byte_ptr_1dlut_dst = (uint8_t *)filter_pregen_store;
-               size_1dlut = sizeof(filter_isharp_1D_lut_3p0x);
-               memset(byte_ptr_1dlut_dst, 0, size_1dlut);
-               for (j = 0; j < size_1dlut; j++) {
-                       sharp_base = spl_fixpt_from_int((int)*byte_ptr_1dlut_src);
-                       sharp_calc = spl_fixpt_mul(sharp_base, sharp_level);
-                       sharp_calc = spl_fixpt_div(sharp_calc, spl_fixpt_from_int(3));
-                       sharp_calc = spl_fixpt_min(spl_fixpt_from_int(255), sharp_calc);
-                       sharp_calc = spl_fixpt_add(sharp_calc, spl_fixpt_from_fraction(1, 2));
-                       sharp_calc_int = spl_fixpt_floor(sharp_calc);
-                       if (sharp_calc_int > 255)
-                               sharp_calc_int = 255;
-                       *byte_ptr_1dlut_dst = (uint8_t)sharp_calc_int;
-
-                       byte_ptr_1dlut_src++;
-                       byte_ptr_1dlut_dst++;
-               }
-
-               /* Compare if filter has change, if so update */
-               if (memcmp((void *)filter_isharp_1D_lut_pregen[i], (void *)filter_pregen_store, size_1dlut) != 0)
-                       memcpy((void *)filter_isharp_1D_lut_pregen[i], (void *)filter_pregen_store, size_1dlut);
-       }
-}
-
-uint32_t *spl_get_pregen_filter_isharp_1D_lut(enum explicit_sharpness sharpness)
-{
-       return filter_isharp_1D_lut_pregen[sharpness];
-}
-
-void spl_init_blur_scale_coeffs(void)
+const uint16_t *spl_get_filter_isharp_bs_4tap_64p(void)
 {
-       convert_filter_s1_10_to_s1_12(filter_isharp_bs_3tap_64p,
-               filter_isharp_bs_3tap_64p_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(filter_isharp_bs_4tap_64p,
-               filter_isharp_bs_4tap_64p_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(filter_isharp_bs_4tap_in_6_64p,
-               filter_isharp_bs_4tap_in_6_64p_s1_12, 6);
+       return filter_isharp_bs_4tap_64p;
 }
-
-uint16_t *spl_dscl_get_blur_scale_coeffs_64p(int taps)
-{
-       if (taps == 3)
-               return spl_get_filter_isharp_bs_3tap_64p();
-       else if (taps == 4)
-               return spl_get_filter_isharp_bs_4tap_64p();
-       else if (taps == 6)
-               return spl_get_filter_isharp_bs_4tap_in_6_64p();
-       else {
-               /* should never happen, bug */
-               SPL_BREAK_TO_DEBUGGER();
-               return NULL;
-       }
-}
-
-void spl_set_blur_scale_data(struct dscl_prog_data *dscl_prog_data,
-               const struct spl_scaler_data *data)
-{
-       dscl_prog_data->filter_blur_scale_h =
-               spl_dscl_get_blur_scale_coeffs_64p(data->taps.h_taps);
-
-       dscl_prog_data->filter_blur_scale_v =
-               spl_dscl_get_blur_scale_coeffs_64p(data->taps.v_taps);
-}
-
index 6cb000bf9d5303328dbf8d8aba6fb1852dbeb8bd..1aaf4c50c1bc2e6c4e1ab89d430457eba4d8a41b 100644 (file)
@@ -12,37 +12,6 @@ const uint32_t *spl_get_filter_isharp_1D_lut_0p5x(void);
 const uint32_t *spl_get_filter_isharp_1D_lut_1p0x(void);
 const uint32_t *spl_get_filter_isharp_1D_lut_1p5x(void);
 const uint32_t *spl_get_filter_isharp_1D_lut_2p0x(void);
-const uint32_t *spl_get_filter_isharp_1D_lut_3p0x(void);
-uint16_t *spl_get_filter_isharp_bs_4tap_in_6_64p(void);
-uint16_t *spl_get_filter_isharp_bs_4tap_64p(void);
-uint16_t *spl_get_filter_isharp_bs_3tap_64p(void);
+const uint16_t *spl_get_filter_isharp_bs_4tap_64p(void);
 const uint16_t *spl_get_filter_isharp_wide_6tap_64p(void);
-uint16_t *spl_dscl_get_blur_scale_coeffs_64p(int taps);
-
-struct scale_ratio_to_sharpness_level_lookup {
-       unsigned int ratio_numer;
-       unsigned int ratio_denom;
-       unsigned int sharpness_numer;
-       unsigned int sharpness_denom;
-};
-
-struct sharpness_level_mapping {
-       unsigned int level;
-       unsigned int level_numer;
-       unsigned int level_denom;
-};
-
-enum system_setup {
-       SDR_NL = 0,
-       SDR_L,
-       HDR_NL,
-       HDR_L
-};
-
-void spl_init_blur_scale_coeffs(void);
-void spl_set_blur_scale_data(struct dscl_prog_data *dscl_prog_data,
-       const struct spl_scaler_data *data);
-
-void spl_build_isharp_1dlut_from_reference_curve(struct spl_fixed31_32 ratio, enum system_setup setup);
-uint32_t *spl_get_pregen_filter_isharp_1D_lut(enum explicit_sharpness sharpness);
 #endif /* __DC_SPL_ISHARP_FILTERS_H__ */
diff --git a/drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_easf_filters.c b/drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_easf_filters.c
deleted file mode 100644 (file)
index 09bf82f..0000000
+++ /dev/null
@@ -1,1726 +0,0 @@
-// SPDX-License-Identifier: MIT
-//
-// Copyright 2024 Advanced Micro Devices, Inc.
-
-#include "spl_debug.h"
-#include "dc_spl_filters.h"
-#include "dc_spl_scl_filters.h"
-#include "dc_spl_scl_easf_filters.h"
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.3_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.300000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_30[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01F6, 0x0206, 0x0004,
-       0x01EC, 0x020B, 0x0009,
-       0x01E2, 0x0211, 0x000D,
-       0x01D8, 0x0216, 0x0012,
-       0x01CE, 0x021C, 0x0016,
-       0x01C4, 0x0221, 0x001B,
-       0x01BA, 0x0226, 0x0020,
-       0x01B0, 0x022A, 0x0026,
-       0x01A6, 0x022F, 0x002B,
-       0x019C, 0x0233, 0x0031,
-       0x0192, 0x0238, 0x0036,
-       0x0188, 0x023C, 0x003C,
-       0x017E, 0x0240, 0x0042,
-       0x0174, 0x0244, 0x0048,
-       0x016A, 0x0248, 0x004E,
-       0x0161, 0x024A, 0x0055,
-       0x0157, 0x024E, 0x005B,
-       0x014D, 0x0251, 0x0062,
-       0x0144, 0x0253, 0x0069,
-       0x013A, 0x0256, 0x0070,
-       0x0131, 0x0258, 0x0077,
-       0x0127, 0x025B, 0x007E,
-       0x011E, 0x025C, 0x0086,
-       0x0115, 0x025E, 0x008D,
-       0x010B, 0x0260, 0x0095,
-       0x0102, 0x0262, 0x009C,
-       0x00F9, 0x0263, 0x00A4,
-       0x00F0, 0x0264, 0x00AC,
-       0x00E7, 0x0265, 0x00B4,
-       0x00DF, 0x0264, 0x00BD,
-       0x00D6, 0x0265, 0x00C5,
-       0x00CD, 0x0266, 0x00CD,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.4_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.400000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_40[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01F6, 0x0206, 0x0004,
-       0x01EB, 0x020E, 0x0007,
-       0x01E1, 0x0214, 0x000B,
-       0x01D7, 0x021A, 0x000F,
-       0x01CD, 0x0220, 0x0013,
-       0x01C2, 0x0226, 0x0018,
-       0x01B8, 0x022C, 0x001C,
-       0x01AE, 0x0231, 0x0021,
-       0x01A3, 0x0237, 0x0026,
-       0x0199, 0x023C, 0x002B,
-       0x018F, 0x0240, 0x0031,
-       0x0185, 0x0245, 0x0036,
-       0x017A, 0x024A, 0x003C,
-       0x0170, 0x024F, 0x0041,
-       0x0166, 0x0253, 0x0047,
-       0x015C, 0x0257, 0x004D,
-       0x0152, 0x025A, 0x0054,
-       0x0148, 0x025E, 0x005A,
-       0x013E, 0x0261, 0x0061,
-       0x0134, 0x0264, 0x0068,
-       0x012B, 0x0266, 0x006F,
-       0x0121, 0x0269, 0x0076,
-       0x0117, 0x026C, 0x007D,
-       0x010E, 0x026E, 0x0084,
-       0x0104, 0x0270, 0x008C,
-       0x00FB, 0x0271, 0x0094,
-       0x00F2, 0x0272, 0x009C,
-       0x00E9, 0x0273, 0x00A4,
-       0x00E0, 0x0274, 0x00AC,
-       0x00D7, 0x0275, 0x00B4,
-       0x00CE, 0x0275, 0x00BD,
-       0x00C5, 0x0276, 0x00C5,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.5_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.500000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_50[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01F5, 0x0209, 0x0002,
-       0x01EA, 0x0211, 0x0005,
-       0x01DF, 0x021A, 0x0007,
-       0x01D4, 0x0222, 0x000A,
-       0x01C9, 0x022A, 0x000D,
-       0x01BE, 0x0232, 0x0010,
-       0x01B3, 0x0239, 0x0014,
-       0x01A8, 0x0241, 0x0017,
-       0x019D, 0x0248, 0x001B,
-       0x0192, 0x024F, 0x001F,
-       0x0187, 0x0255, 0x0024,
-       0x017C, 0x025C, 0x0028,
-       0x0171, 0x0262, 0x002D,
-       0x0166, 0x0268, 0x0032,
-       0x015B, 0x026E, 0x0037,
-       0x0150, 0x0273, 0x003D,
-       0x0146, 0x0278, 0x0042,
-       0x013B, 0x027D, 0x0048,
-       0x0130, 0x0282, 0x004E,
-       0x0126, 0x0286, 0x0054,
-       0x011B, 0x028A, 0x005B,
-       0x0111, 0x028D, 0x0062,
-       0x0107, 0x0290, 0x0069,
-       0x00FD, 0x0293, 0x0070,
-       0x00F3, 0x0296, 0x0077,
-       0x00E9, 0x0298, 0x007F,
-       0x00DF, 0x029A, 0x0087,
-       0x00D5, 0x029C, 0x008F,
-       0x00CC, 0x029D, 0x0097,
-       0x00C3, 0x029E, 0x009F,
-       0x00BA, 0x029E, 0x00A8,
-       0x00B1, 0x029E, 0x00B1,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.6_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.600000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_60[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01F4, 0x020B, 0x0001,
-       0x01E8, 0x0216, 0x0002,
-       0x01DC, 0x0221, 0x0003,
-       0x01D0, 0x022B, 0x0005,
-       0x01C4, 0x0235, 0x0007,
-       0x01B8, 0x0240, 0x0008,
-       0x01AC, 0x0249, 0x000B,
-       0x01A0, 0x0253, 0x000D,
-       0x0194, 0x025C, 0x0010,
-       0x0188, 0x0265, 0x0013,
-       0x017C, 0x026E, 0x0016,
-       0x0170, 0x0277, 0x0019,
-       0x0164, 0x027F, 0x001D,
-       0x0158, 0x0287, 0x0021,
-       0x014C, 0x028F, 0x0025,
-       0x0140, 0x0297, 0x0029,
-       0x0135, 0x029D, 0x002E,
-       0x0129, 0x02A4, 0x0033,
-       0x011D, 0x02AB, 0x0038,
-       0x0112, 0x02B0, 0x003E,
-       0x0107, 0x02B5, 0x0044,
-       0x00FC, 0x02BA, 0x004A,
-       0x00F1, 0x02BF, 0x0050,
-       0x00E6, 0x02C3, 0x0057,
-       0x00DB, 0x02C7, 0x005E,
-       0x00D1, 0x02CA, 0x0065,
-       0x00C7, 0x02CC, 0x006D,
-       0x00BD, 0x02CE, 0x0075,
-       0x00B3, 0x02D0, 0x007D,
-       0x00A9, 0x02D2, 0x0085,
-       0x00A0, 0x02D2, 0x008E,
-       0x0097, 0x02D2, 0x0097,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.7_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.700000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_70[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01F3, 0x020D, 0x0000,
-       0x01E5, 0x021B, 0x0000,
-       0x01D8, 0x0228, 0x0000,
-       0x01CB, 0x0235, 0x0000,
-       0x01BD, 0x0243, 0x0000,
-       0x01B0, 0x024F, 0x0001,
-       0x01A2, 0x025C, 0x0002,
-       0x0195, 0x0268, 0x0003,
-       0x0187, 0x0275, 0x0004,
-       0x017A, 0x0280, 0x0006,
-       0x016D, 0x028C, 0x0007,
-       0x015F, 0x0298, 0x0009,
-       0x0152, 0x02A2, 0x000C,
-       0x0145, 0x02AD, 0x000E,
-       0x0138, 0x02B7, 0x0011,
-       0x012B, 0x02C0, 0x0015,
-       0x011E, 0x02CA, 0x0018,
-       0x0111, 0x02D3, 0x001C,
-       0x0105, 0x02DB, 0x0020,
-       0x00F8, 0x02E3, 0x0025,
-       0x00EC, 0x02EA, 0x002A,
-       0x00E0, 0x02F1, 0x002F,
-       0x00D5, 0x02F6, 0x0035,
-       0x00C9, 0x02FC, 0x003B,
-       0x00BE, 0x0301, 0x0041,
-       0x00B3, 0x0305, 0x0048,
-       0x00A8, 0x0309, 0x004F,
-       0x009E, 0x030C, 0x0056,
-       0x0094, 0x030E, 0x005E,
-       0x008A, 0x0310, 0x0066,
-       0x0081, 0x0310, 0x006F,
-       0x0077, 0x0312, 0x0077,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.8_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.800000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_80[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01F1, 0x0210, 0x0FFF,
-       0x01E2, 0x0220, 0x0FFE,
-       0x01D2, 0x0232, 0x0FFC,
-       0x01C3, 0x0241, 0x0FFC,
-       0x01B4, 0x0251, 0x0FFB,
-       0x01A4, 0x0262, 0x0FFA,
-       0x0195, 0x0271, 0x0FFA,
-       0x0186, 0x0281, 0x0FF9,
-       0x0176, 0x0291, 0x0FF9,
-       0x0167, 0x02A0, 0x0FF9,
-       0x0158, 0x02AE, 0x0FFA,
-       0x0149, 0x02BD, 0x0FFA,
-       0x013A, 0x02CB, 0x0FFB,
-       0x012C, 0x02D7, 0x0FFD,
-       0x011D, 0x02E5, 0x0FFE,
-       0x010F, 0x02F1, 0x0000,
-       0x0101, 0x02FD, 0x0002,
-       0x00F3, 0x0308, 0x0005,
-       0x00E5, 0x0313, 0x0008,
-       0x00D8, 0x031D, 0x000B,
-       0x00CB, 0x0326, 0x000F,
-       0x00BE, 0x032F, 0x0013,
-       0x00B2, 0x0337, 0x0017,
-       0x00A6, 0x033E, 0x001C,
-       0x009A, 0x0345, 0x0021,
-       0x008F, 0x034A, 0x0027,
-       0x0084, 0x034F, 0x002D,
-       0x0079, 0x0353, 0x0034,
-       0x006F, 0x0356, 0x003B,
-       0x0065, 0x0358, 0x0043,
-       0x005C, 0x0359, 0x004B,
-       0x0053, 0x035A, 0x0053,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_0.9_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.900000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_0_90[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01EE, 0x0214, 0x0FFE,
-       0x01DC, 0x0228, 0x0FFC,
-       0x01CA, 0x023C, 0x0FFA,
-       0x01B9, 0x024F, 0x0FF8,
-       0x01A7, 0x0262, 0x0FF7,
-       0x0195, 0x0276, 0x0FF5,
-       0x0183, 0x028A, 0x0FF3,
-       0x0172, 0x029C, 0x0FF2,
-       0x0160, 0x02AF, 0x0FF1,
-       0x014F, 0x02C2, 0x0FEF,
-       0x013E, 0x02D4, 0x0FEE,
-       0x012D, 0x02E5, 0x0FEE,
-       0x011C, 0x02F7, 0x0FED,
-       0x010C, 0x0307, 0x0FED,
-       0x00FB, 0x0318, 0x0FED,
-       0x00EC, 0x0327, 0x0FED,
-       0x00DC, 0x0336, 0x0FEE,
-       0x00CD, 0x0344, 0x0FEF,
-       0x00BE, 0x0352, 0x0FF0,
-       0x00B0, 0x035E, 0x0FF2,
-       0x00A2, 0x036A, 0x0FF4,
-       0x0095, 0x0375, 0x0FF6,
-       0x0088, 0x037F, 0x0FF9,
-       0x007B, 0x0388, 0x0FFD,
-       0x006F, 0x0391, 0x0000,
-       0x0064, 0x0397, 0x0005,
-       0x0059, 0x039D, 0x000A,
-       0x004E, 0x03A3, 0x000F,
-       0x0045, 0x03A6, 0x0015,
-       0x003B, 0x03A9, 0x001C,
-       0x0033, 0x03AA, 0x0023,
-       0x002A, 0x03AC, 0x002A,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      3t_64p_LanczosEd_p_1_p_10qb_
-// <num_taps>          3
-// <num_phases>                64
-// <scale_ratio>        input/output = 1.000000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_3tap_64p_ratio_1_00[99] = {
-       0x0200, 0x0200, 0x0000,
-       0x01EB, 0x0217, 0x0FFE,
-       0x01D5, 0x022F, 0x0FFC,
-       0x01C0, 0x0247, 0x0FF9,
-       0x01AB, 0x025E, 0x0FF7,
-       0x0196, 0x0276, 0x0FF4,
-       0x0181, 0x028D, 0x0FF2,
-       0x016C, 0x02A5, 0x0FEF,
-       0x0158, 0x02BB, 0x0FED,
-       0x0144, 0x02D1, 0x0FEB,
-       0x0130, 0x02E8, 0x0FE8,
-       0x011C, 0x02FE, 0x0FE6,
-       0x0109, 0x0313, 0x0FE4,
-       0x00F6, 0x0328, 0x0FE2,
-       0x00E4, 0x033C, 0x0FE0,
-       0x00D2, 0x034F, 0x0FDF,
-       0x00C0, 0x0363, 0x0FDD,
-       0x00B0, 0x0374, 0x0FDC,
-       0x009F, 0x0385, 0x0FDC,
-       0x0090, 0x0395, 0x0FDB,
-       0x0081, 0x03A4, 0x0FDB,
-       0x0072, 0x03B3, 0x0FDB,
-       0x0064, 0x03C0, 0x0FDC,
-       0x0057, 0x03CC, 0x0FDD,
-       0x004B, 0x03D6, 0x0FDF,
-       0x003F, 0x03E0, 0x0FE1,
-       0x0034, 0x03E8, 0x0FE4,
-       0x002A, 0x03EF, 0x0FE7,
-       0x0020, 0x03F5, 0x0FEB,
-       0x0017, 0x03FA, 0x0FEF,
-       0x000F, 0x03FD, 0x0FF4,
-       0x0007, 0x03FF, 0x0FFA,
-       0x0000, 0x0400, 0x0000,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.3_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.300000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_30[132] = {
-       0x0104, 0x01F8, 0x0104, 0x0000,
-       0x00FE, 0x01F7, 0x010A, 0x0001,
-       0x00F8, 0x01F6, 0x010F, 0x0003,
-       0x00F2, 0x01F5, 0x0114, 0x0005,
-       0x00EB, 0x01F4, 0x011B, 0x0006,
-       0x00E5, 0x01F3, 0x0120, 0x0008,
-       0x00DF, 0x01F2, 0x0125, 0x000A,
-       0x00DA, 0x01F0, 0x012A, 0x000C,
-       0x00D4, 0x01EE, 0x0130, 0x000E,
-       0x00CE, 0x01ED, 0x0135, 0x0010,
-       0x00C8, 0x01EB, 0x013A, 0x0013,
-       0x00C2, 0x01E9, 0x0140, 0x0015,
-       0x00BD, 0x01E7, 0x0145, 0x0017,
-       0x00B7, 0x01E5, 0x014A, 0x001A,
-       0x00B1, 0x01E2, 0x0151, 0x001C,
-       0x00AC, 0x01E0, 0x0155, 0x001F,
-       0x00A7, 0x01DD, 0x015A, 0x0022,
-       0x00A1, 0x01DB, 0x015F, 0x0025,
-       0x009C, 0x01D8, 0x0165, 0x0027,
-       0x0097, 0x01D5, 0x016A, 0x002A,
-       0x0092, 0x01D2, 0x016E, 0x002E,
-       0x008C, 0x01CF, 0x0174, 0x0031,
-       0x0087, 0x01CC, 0x0179, 0x0034,
-       0x0083, 0x01C9, 0x017D, 0x0037,
-       0x007E, 0x01C5, 0x0182, 0x003B,
-       0x0079, 0x01C2, 0x0187, 0x003E,
-       0x0074, 0x01BE, 0x018C, 0x0042,
-       0x0070, 0x01BA, 0x0190, 0x0046,
-       0x006B, 0x01B7, 0x0195, 0x0049,
-       0x0066, 0x01B3, 0x019A, 0x004D,
-       0x0062, 0x01AF, 0x019E, 0x0051,
-       0x005E, 0x01AB, 0x01A2, 0x0055,
-       0x005A, 0x01A6, 0x01A6, 0x005A,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.4_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.400000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_40[132] = {
-       0x00FB, 0x0209, 0x00FC, 0x0000,
-       0x00F5, 0x0209, 0x0101, 0x0001,
-       0x00EE, 0x0208, 0x0108, 0x0002,
-       0x00E8, 0x0207, 0x010E, 0x0003,
-       0x00E2, 0x0206, 0x0114, 0x0004,
-       0x00DB, 0x0205, 0x011A, 0x0006,
-       0x00D5, 0x0204, 0x0120, 0x0007,
-       0x00CF, 0x0203, 0x0125, 0x0009,
-       0x00C9, 0x0201, 0x012C, 0x000A,
-       0x00C3, 0x01FF, 0x0132, 0x000C,
-       0x00BD, 0x01FD, 0x0138, 0x000E,
-       0x00B7, 0x01FB, 0x013E, 0x0010,
-       0x00B1, 0x01F9, 0x0144, 0x0012,
-       0x00AC, 0x01F7, 0x0149, 0x0014,
-       0x00A6, 0x01F4, 0x0150, 0x0016,
-       0x00A0, 0x01F2, 0x0156, 0x0018,
-       0x009B, 0x01EF, 0x015C, 0x001A,
-       0x0095, 0x01EC, 0x0162, 0x001D,
-       0x0090, 0x01E9, 0x0168, 0x001F,
-       0x008B, 0x01E6, 0x016D, 0x0022,
-       0x0085, 0x01E3, 0x0173, 0x0025,
-       0x0080, 0x01DF, 0x0179, 0x0028,
-       0x007B, 0x01DC, 0x017E, 0x002B,
-       0x0076, 0x01D8, 0x0184, 0x002E,
-       0x0071, 0x01D4, 0x018A, 0x0031,
-       0x006D, 0x01D1, 0x018E, 0x0034,
-       0x0068, 0x01CD, 0x0193, 0x0038,
-       0x0063, 0x01C8, 0x019A, 0x003B,
-       0x005F, 0x01C4, 0x019E, 0x003F,
-       0x005B, 0x01C0, 0x01A3, 0x0042,
-       0x0056, 0x01BB, 0x01A9, 0x0046,
-       0x0052, 0x01B7, 0x01AD, 0x004A,
-       0x004E, 0x01B2, 0x01B2, 0x004E,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.5_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.500000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_50[132] = {
-       0x00E5, 0x0236, 0x00E5, 0x0000,
-       0x00DE, 0x0235, 0x00ED, 0x0000,
-       0x00D7, 0x0235, 0x00F4, 0x0000,
-       0x00D0, 0x0235, 0x00FB, 0x0000,
-       0x00C9, 0x0234, 0x0102, 0x0001,
-       0x00C2, 0x0233, 0x010A, 0x0001,
-       0x00BC, 0x0232, 0x0111, 0x0001,
-       0x00B5, 0x0230, 0x0119, 0x0002,
-       0x00AE, 0x022F, 0x0121, 0x0002,
-       0x00A8, 0x022D, 0x0128, 0x0003,
-       0x00A2, 0x022B, 0x012F, 0x0004,
-       0x009B, 0x0229, 0x0137, 0x0005,
-       0x0095, 0x0226, 0x013F, 0x0006,
-       0x008F, 0x0224, 0x0146, 0x0007,
-       0x0089, 0x0221, 0x014E, 0x0008,
-       0x0083, 0x021E, 0x0155, 0x000A,
-       0x007E, 0x021B, 0x015C, 0x000B,
-       0x0078, 0x0217, 0x0164, 0x000D,
-       0x0072, 0x0213, 0x016D, 0x000E,
-       0x006D, 0x0210, 0x0173, 0x0010,
-       0x0068, 0x020C, 0x017A, 0x0012,
-       0x0063, 0x0207, 0x0182, 0x0014,
-       0x005E, 0x0203, 0x0189, 0x0016,
-       0x0059, 0x01FE, 0x0191, 0x0018,
-       0x0054, 0x01F9, 0x0198, 0x001B,
-       0x0050, 0x01F4, 0x019F, 0x001D,
-       0x004B, 0x01EF, 0x01A6, 0x0020,
-       0x0047, 0x01EA, 0x01AC, 0x0023,
-       0x0043, 0x01E4, 0x01B3, 0x0026,
-       0x003F, 0x01DF, 0x01B9, 0x0029,
-       0x003B, 0x01D9, 0x01C0, 0x002C,
-       0x0037, 0x01D3, 0x01C6, 0x0030,
-       0x0033, 0x01CD, 0x01CD, 0x0033,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.6_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.600000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_60[132] = {
-       0x00C8, 0x026F, 0x00C9, 0x0000,
-       0x00C0, 0x0270, 0x00D1, 0x0FFF,
-       0x00B8, 0x0270, 0x00D9, 0x0FFF,
-       0x00B1, 0x0270, 0x00E1, 0x0FFE,
-       0x00A9, 0x026F, 0x00EB, 0x0FFD,
-       0x00A2, 0x026E, 0x00F3, 0x0FFD,
-       0x009A, 0x026D, 0x00FD, 0x0FFC,
-       0x0093, 0x026C, 0x0105, 0x0FFC,
-       0x008C, 0x026A, 0x010F, 0x0FFB,
-       0x0085, 0x0268, 0x0118, 0x0FFB,
-       0x007E, 0x0265, 0x0122, 0x0FFB,
-       0x0078, 0x0263, 0x012A, 0x0FFB,
-       0x0071, 0x0260, 0x0134, 0x0FFB,
-       0x006B, 0x025C, 0x013E, 0x0FFB,
-       0x0065, 0x0259, 0x0147, 0x0FFB,
-       0x005F, 0x0255, 0x0151, 0x0FFB,
-       0x0059, 0x0251, 0x015A, 0x0FFC,
-       0x0054, 0x024D, 0x0163, 0x0FFC,
-       0x004E, 0x0248, 0x016D, 0x0FFD,
-       0x0049, 0x0243, 0x0176, 0x0FFE,
-       0x0044, 0x023E, 0x017F, 0x0FFF,
-       0x003F, 0x0238, 0x0189, 0x0000,
-       0x003A, 0x0232, 0x0193, 0x0001,
-       0x0036, 0x022C, 0x019C, 0x0002,
-       0x0031, 0x0226, 0x01A5, 0x0004,
-       0x002D, 0x021F, 0x01AF, 0x0005,
-       0x0029, 0x0218, 0x01B8, 0x0007,
-       0x0025, 0x0211, 0x01C1, 0x0009,
-       0x0022, 0x020A, 0x01C9, 0x000B,
-       0x001E, 0x0203, 0x01D2, 0x000D,
-       0x001B, 0x01FB, 0x01DA, 0x0010,
-       0x0018, 0x01F3, 0x01E3, 0x0012,
-       0x0015, 0x01EB, 0x01EB, 0x0015,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.7_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.700000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_70[132] = {
-       0x00A3, 0x02B9, 0x00A4, 0x0000,
-       0x009A, 0x02BA, 0x00AD, 0x0FFF,
-       0x0092, 0x02BA, 0x00B6, 0x0FFE,
-       0x0089, 0x02BA, 0x00C1, 0x0FFC,
-       0x0081, 0x02B9, 0x00CB, 0x0FFB,
-       0x0079, 0x02B8, 0x00D5, 0x0FFA,
-       0x0071, 0x02B7, 0x00DF, 0x0FF9,
-       0x0069, 0x02B5, 0x00EA, 0x0FF8,
-       0x0062, 0x02B3, 0x00F4, 0x0FF7,
-       0x005B, 0x02B0, 0x00FF, 0x0FF6,
-       0x0054, 0x02AD, 0x010B, 0x0FF4,
-       0x004D, 0x02A9, 0x0117, 0x0FF3,
-       0x0046, 0x02A5, 0x0123, 0x0FF2,
-       0x0040, 0x02A1, 0x012D, 0x0FF2,
-       0x003A, 0x029C, 0x0139, 0x0FF1,
-       0x0034, 0x0297, 0x0145, 0x0FF0,
-       0x002F, 0x0292, 0x0150, 0x0FEF,
-       0x0029, 0x028C, 0x015C, 0x0FEF,
-       0x0024, 0x0285, 0x0169, 0x0FEE,
-       0x001F, 0x027F, 0x0174, 0x0FEE,
-       0x001B, 0x0278, 0x017F, 0x0FEE,
-       0x0016, 0x0270, 0x018D, 0x0FED,
-       0x0012, 0x0268, 0x0199, 0x0FED,
-       0x000E, 0x0260, 0x01A4, 0x0FEE,
-       0x000B, 0x0258, 0x01AF, 0x0FEE,
-       0x0007, 0x024F, 0x01BC, 0x0FEE,
-       0x0004, 0x0246, 0x01C7, 0x0FEF,
-       0x0001, 0x023D, 0x01D3, 0x0FEF,
-       0x0FFE, 0x0233, 0x01DF, 0x0FF0,
-       0x0FFC, 0x0229, 0x01EA, 0x0FF1,
-       0x0FFA, 0x021F, 0x01F4, 0x0FF3,
-       0x0FF8, 0x0215, 0x01FF, 0x0FF4,
-       0x0FF6, 0x020A, 0x020A, 0x0FF6,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.8_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.800000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_80[132] = {
-       0x0075, 0x0315, 0x0076, 0x0000,
-       0x006C, 0x0316, 0x007F, 0x0FFF,
-       0x0062, 0x0316, 0x008A, 0x0FFE,
-       0x0059, 0x0315, 0x0096, 0x0FFC,
-       0x0050, 0x0314, 0x00A1, 0x0FFB,
-       0x0048, 0x0312, 0x00AD, 0x0FF9,
-       0x0040, 0x0310, 0x00B8, 0x0FF8,
-       0x0038, 0x030D, 0x00C5, 0x0FF6,
-       0x0030, 0x030A, 0x00D1, 0x0FF5,
-       0x0029, 0x0306, 0x00DE, 0x0FF3,
-       0x0022, 0x0301, 0x00EB, 0x0FF2,
-       0x001C, 0x02FC, 0x00F8, 0x0FF0,
-       0x0015, 0x02F7, 0x0106, 0x0FEE,
-       0x0010, 0x02F1, 0x0112, 0x0FED,
-       0x000A, 0x02EA, 0x0121, 0x0FEB,
-       0x0005, 0x02E3, 0x012F, 0x0FE9,
-       0x0000, 0x02DB, 0x013D, 0x0FE8,
-       0x0FFB, 0x02D3, 0x014C, 0x0FE6,
-       0x0FF7, 0x02CA, 0x015A, 0x0FE5,
-       0x0FF3, 0x02C1, 0x0169, 0x0FE3,
-       0x0FF0, 0x02B7, 0x0177, 0x0FE2,
-       0x0FEC, 0x02AD, 0x0186, 0x0FE1,
-       0x0FE9, 0x02A2, 0x0196, 0x0FDF,
-       0x0FE7, 0x0297, 0x01A4, 0x0FDE,
-       0x0FE4, 0x028C, 0x01B3, 0x0FDD,
-       0x0FE2, 0x0280, 0x01C2, 0x0FDC,
-       0x0FE0, 0x0274, 0x01D0, 0x0FDC,
-       0x0FDF, 0x0268, 0x01DE, 0x0FDB,
-       0x0FDD, 0x025B, 0x01EE, 0x0FDA,
-       0x0FDC, 0x024E, 0x01FC, 0x0FDA,
-       0x0FDB, 0x0241, 0x020A, 0x0FDA,
-       0x0FDB, 0x0233, 0x0218, 0x0FDA,
-       0x0FDA, 0x0226, 0x0226, 0x0FDA,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_0.9_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.900000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_0_90[132] = {
-       0x003F, 0x0383, 0x003E, 0x0000,
-       0x0034, 0x0383, 0x004A, 0x0FFF,
-       0x002B, 0x0383, 0x0054, 0x0FFE,
-       0x0021, 0x0381, 0x0061, 0x0FFD,
-       0x0019, 0x037F, 0x006C, 0x0FFC,
-       0x0010, 0x037C, 0x0079, 0x0FFB,
-       0x0008, 0x0378, 0x0086, 0x0FFA,
-       0x0001, 0x0374, 0x0093, 0x0FF8,
-       0x0FFA, 0x036E, 0x00A1, 0x0FF7,
-       0x0FF3, 0x0368, 0x00B0, 0x0FF5,
-       0x0FED, 0x0361, 0x00BF, 0x0FF3,
-       0x0FE8, 0x035A, 0x00CD, 0x0FF1,
-       0x0FE2, 0x0352, 0x00DC, 0x0FF0,
-       0x0FDE, 0x0349, 0x00EB, 0x0FEE,
-       0x0FD9, 0x033F, 0x00FC, 0x0FEC,
-       0x0FD5, 0x0335, 0x010D, 0x0FE9,
-       0x0FD2, 0x032A, 0x011D, 0x0FE7,
-       0x0FCF, 0x031E, 0x012E, 0x0FE5,
-       0x0FCC, 0x0312, 0x013F, 0x0FE3,
-       0x0FCA, 0x0305, 0x0150, 0x0FE1,
-       0x0FC8, 0x02F8, 0x0162, 0x0FDE,
-       0x0FC6, 0x02EA, 0x0174, 0x0FDC,
-       0x0FC5, 0x02DC, 0x0185, 0x0FDA,
-       0x0FC4, 0x02CD, 0x0197, 0x0FD8,
-       0x0FC3, 0x02BE, 0x01AA, 0x0FD5,
-       0x0FC3, 0x02AF, 0x01BB, 0x0FD3,
-       0x0FC3, 0x029F, 0x01CD, 0x0FD1,
-       0x0FC3, 0x028E, 0x01E0, 0x0FCF,
-       0x0FC3, 0x027E, 0x01F2, 0x0FCD,
-       0x0FC4, 0x026D, 0x0203, 0x0FCC,
-       0x0FC5, 0x025C, 0x0215, 0x0FCA,
-       0x0FC6, 0x024B, 0x0227, 0x0FC8,
-       0x0FC7, 0x0239, 0x0239, 0x0FC7,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      03-Apr-2024
-// <coeffDescrip>      4t_64p_LanczosEd_p_1_p_10qb_
-// <num_taps>          4
-// <num_phases>                64
-// <scale_ratio>        input/output = 1.000000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_4tap_64p_ratio_1_00[132] = {
-       0x0000, 0x0400, 0x0000, 0x0000,
-       0x0FF6, 0x03FF, 0x000B, 0x0000,
-       0x0FED, 0x03FE, 0x0015, 0x0000,
-       0x0FE4, 0x03FB, 0x0022, 0x0FFF,
-       0x0FDC, 0x03F7, 0x002E, 0x0FFF,
-       0x0FD5, 0x03F2, 0x003B, 0x0FFE,
-       0x0FCE, 0x03EC, 0x0048, 0x0FFE,
-       0x0FC8, 0x03E5, 0x0056, 0x0FFD,
-       0x0FC3, 0x03DC, 0x0065, 0x0FFC,
-       0x0FBE, 0x03D3, 0x0075, 0x0FFA,
-       0x0FB9, 0x03C9, 0x0085, 0x0FF9,
-       0x0FB6, 0x03BE, 0x0094, 0x0FF8,
-       0x0FB2, 0x03B2, 0x00A6, 0x0FF6,
-       0x0FB0, 0x03A5, 0x00B7, 0x0FF4,
-       0x0FAD, 0x0397, 0x00CA, 0x0FF2,
-       0x0FAB, 0x0389, 0x00DC, 0x0FF0,
-       0x0FAA, 0x0379, 0x00EF, 0x0FEE,
-       0x0FA9, 0x0369, 0x0102, 0x0FEC,
-       0x0FA9, 0x0359, 0x0115, 0x0FE9,
-       0x0FA9, 0x0348, 0x0129, 0x0FE6,
-       0x0FA9, 0x0336, 0x013D, 0x0FE4,
-       0x0FA9, 0x0323, 0x0153, 0x0FE1,
-       0x0FAA, 0x0310, 0x0168, 0x0FDE,
-       0x0FAC, 0x02FD, 0x017C, 0x0FDB,
-       0x0FAD, 0x02E9, 0x0192, 0x0FD8,
-       0x0FAF, 0x02D5, 0x01A7, 0x0FD5,
-       0x0FB1, 0x02C0, 0x01BD, 0x0FD2,
-       0x0FB3, 0x02AC, 0x01D2, 0x0FCF,
-       0x0FB5, 0x0296, 0x01E9, 0x0FCC,
-       0x0FB8, 0x0281, 0x01FE, 0x0FC9,
-       0x0FBA, 0x026C, 0x0214, 0x0FC6,
-       0x0FBD, 0x0256, 0x022A, 0x0FC3,
-       0x0FC0, 0x0240, 0x0240, 0x0FC0,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.3_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.300000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_30[198] = {
-       0x004B, 0x0100, 0x0169, 0x0101, 0x004B, 0x0000,
-       0x0049, 0x00FD, 0x0169, 0x0103, 0x004E, 0x0000,
-       0x0047, 0x00FA, 0x0169, 0x0106, 0x0050, 0x0000,
-       0x0045, 0x00F7, 0x0168, 0x0109, 0x0052, 0x0001,
-       0x0043, 0x00F5, 0x0168, 0x010B, 0x0054, 0x0001,
-       0x0040, 0x00F2, 0x0168, 0x010E, 0x0057, 0x0001,
-       0x003E, 0x00EF, 0x0168, 0x0110, 0x0059, 0x0002,
-       0x003C, 0x00EC, 0x0167, 0x0113, 0x005C, 0x0002,
-       0x003A, 0x00E9, 0x0167, 0x0116, 0x005E, 0x0002,
-       0x0038, 0x00E6, 0x0166, 0x0118, 0x0061, 0x0003,
-       0x0036, 0x00E3, 0x0165, 0x011C, 0x0063, 0x0003,
-       0x0034, 0x00E0, 0x0165, 0x011D, 0x0066, 0x0004,
-       0x0033, 0x00DD, 0x0164, 0x0120, 0x0068, 0x0004,
-       0x0031, 0x00DA, 0x0163, 0x0122, 0x006B, 0x0005,
-       0x002F, 0x00D7, 0x0163, 0x0125, 0x006D, 0x0005,
-       0x002D, 0x00D3, 0x0162, 0x0128, 0x0070, 0x0006,
-       0x002B, 0x00D0, 0x0161, 0x012A, 0x0073, 0x0007,
-       0x002A, 0x00CD, 0x0160, 0x012D, 0x0075, 0x0007,
-       0x0028, 0x00CA, 0x015F, 0x012F, 0x0078, 0x0008,
-       0x0026, 0x00C7, 0x015E, 0x0131, 0x007B, 0x0009,
-       0x0025, 0x00C4, 0x015D, 0x0133, 0x007E, 0x0009,
-       0x0023, 0x00C1, 0x015C, 0x0136, 0x0080, 0x000A,
-       0x0022, 0x00BE, 0x015A, 0x0138, 0x0083, 0x000B,
-       0x0020, 0x00BB, 0x0159, 0x013A, 0x0086, 0x000C,
-       0x001F, 0x00B8, 0x0158, 0x013B, 0x0089, 0x000D,
-       0x001E, 0x00B5, 0x0156, 0x013E, 0x008C, 0x000D,
-       0x001C, 0x00B2, 0x0155, 0x0140, 0x008F, 0x000E,
-       0x001B, 0x00AF, 0x0153, 0x0143, 0x0091, 0x000F,
-       0x0019, 0x00AC, 0x0152, 0x0145, 0x0094, 0x0010,
-       0x0018, 0x00A9, 0x0150, 0x0147, 0x0097, 0x0011,
-       0x0017, 0x00A6, 0x014F, 0x0148, 0x009A, 0x0012,
-       0x0016, 0x00A3, 0x014D, 0x0149, 0x009D, 0x0014,
-       0x0015, 0x00A0, 0x014B, 0x014B, 0x00A0, 0x0015,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.4_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.400000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_40[198] = {
-       0x0028, 0x0106, 0x01A3, 0x0107, 0x0028, 0x0000,
-       0x0026, 0x0102, 0x01A3, 0x010A, 0x002B, 0x0000,
-       0x0024, 0x00FE, 0x01A3, 0x010F, 0x002D, 0x0FFF,
-       0x0022, 0x00FA, 0x01A3, 0x0113, 0x002F, 0x0FFF,
-       0x0021, 0x00F6, 0x01A3, 0x0116, 0x0031, 0x0FFF,
-       0x001F, 0x00F2, 0x01A2, 0x011B, 0x0034, 0x0FFE,
-       0x001D, 0x00EE, 0x01A2, 0x011F, 0x0036, 0x0FFE,
-       0x001B, 0x00EA, 0x01A1, 0x0123, 0x0039, 0x0FFE,
-       0x0019, 0x00E6, 0x01A1, 0x0127, 0x003B, 0x0FFE,
-       0x0018, 0x00E2, 0x01A0, 0x012A, 0x003E, 0x0FFE,
-       0x0016, 0x00DE, 0x01A0, 0x012E, 0x0041, 0x0FFD,
-       0x0015, 0x00DA, 0x019F, 0x0132, 0x0043, 0x0FFD,
-       0x0013, 0x00D6, 0x019E, 0x0136, 0x0046, 0x0FFD,
-       0x0012, 0x00D2, 0x019D, 0x0139, 0x0049, 0x0FFD,
-       0x0010, 0x00CE, 0x019C, 0x013D, 0x004C, 0x0FFD,
-       0x000F, 0x00CA, 0x019A, 0x0141, 0x004F, 0x0FFD,
-       0x000E, 0x00C6, 0x0199, 0x0144, 0x0052, 0x0FFD,
-       0x000D, 0x00C2, 0x0197, 0x0148, 0x0055, 0x0FFD,
-       0x000B, 0x00BE, 0x0196, 0x014C, 0x0058, 0x0FFD,
-       0x000A, 0x00BA, 0x0195, 0x014F, 0x005B, 0x0FFD,
-       0x0009, 0x00B6, 0x0193, 0x0153, 0x005E, 0x0FFD,
-       0x0008, 0x00B2, 0x0191, 0x0157, 0x0061, 0x0FFD,
-       0x0007, 0x00AE, 0x0190, 0x015A, 0x0064, 0x0FFD,
-       0x0006, 0x00AA, 0x018E, 0x015D, 0x0068, 0x0FFD,
-       0x0005, 0x00A6, 0x018C, 0x0161, 0x006B, 0x0FFD,
-       0x0005, 0x00A2, 0x0189, 0x0164, 0x006F, 0x0FFD,
-       0x0004, 0x009E, 0x0187, 0x0167, 0x0072, 0x0FFE,
-       0x0003, 0x009A, 0x0185, 0x016B, 0x0075, 0x0FFE,
-       0x0002, 0x0096, 0x0183, 0x016E, 0x0079, 0x0FFE,
-       0x0002, 0x0093, 0x0180, 0x016F, 0x007D, 0x0FFF,
-       0x0001, 0x008F, 0x017E, 0x0173, 0x0080, 0x0FFF,
-       0x0001, 0x008B, 0x017B, 0x0175, 0x0084, 0x0000,
-       0x0000, 0x0087, 0x0179, 0x0179, 0x0087, 0x0000,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.5_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.500000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_50[198] = {
-       0x0000, 0x0107, 0x01F3, 0x0106, 0x0000, 0x0000,
-       0x0FFE, 0x0101, 0x01F3, 0x010D, 0x0002, 0x0FFF,
-       0x0FFD, 0x00FB, 0x01F3, 0x0113, 0x0003, 0x0FFF,
-       0x0FFC, 0x00F6, 0x01F3, 0x0118, 0x0005, 0x0FFE,
-       0x0FFA, 0x00F0, 0x01F3, 0x011E, 0x0007, 0x0FFE,
-       0x0FF9, 0x00EB, 0x01F2, 0x0124, 0x0009, 0x0FFD,
-       0x0FF8, 0x00E5, 0x01F2, 0x0129, 0x000B, 0x0FFD,
-       0x0FF7, 0x00E0, 0x01F1, 0x012F, 0x000D, 0x0FFC,
-       0x0FF6, 0x00DA, 0x01F0, 0x0135, 0x0010, 0x0FFB,
-       0x0FF5, 0x00D4, 0x01EF, 0x013B, 0x0012, 0x0FFB,
-       0x0FF4, 0x00CF, 0x01EE, 0x0141, 0x0014, 0x0FFA,
-       0x0FF3, 0x00C9, 0x01ED, 0x0147, 0x0017, 0x0FF9,
-       0x0FF2, 0x00C4, 0x01EB, 0x014C, 0x001A, 0x0FF9,
-       0x0FF1, 0x00BF, 0x01EA, 0x0152, 0x001C, 0x0FF8,
-       0x0FF1, 0x00B9, 0x01E8, 0x0157, 0x001F, 0x0FF8,
-       0x0FF0, 0x00B4, 0x01E6, 0x015D, 0x0022, 0x0FF7,
-       0x0FF0, 0x00AE, 0x01E4, 0x0163, 0x0025, 0x0FF6,
-       0x0FEF, 0x00A9, 0x01E2, 0x0168, 0x0028, 0x0FF6,
-       0x0FEF, 0x00A4, 0x01DF, 0x016E, 0x002B, 0x0FF5,
-       0x0FEF, 0x009F, 0x01DD, 0x0172, 0x002E, 0x0FF5,
-       0x0FEE, 0x009A, 0x01DA, 0x0178, 0x0032, 0x0FF4,
-       0x0FEE, 0x0094, 0x01D8, 0x017E, 0x0035, 0x0FF3,
-       0x0FEE, 0x008F, 0x01D5, 0x0182, 0x0039, 0x0FF3,
-       0x0FEE, 0x008A, 0x01D2, 0x0188, 0x003C, 0x0FF2,
-       0x0FEE, 0x0085, 0x01CF, 0x018C, 0x0040, 0x0FF2,
-       0x0FEE, 0x0081, 0x01CB, 0x0191, 0x0044, 0x0FF1,
-       0x0FEE, 0x007C, 0x01C8, 0x0196, 0x0047, 0x0FF1,
-       0x0FEE, 0x0077, 0x01C4, 0x019C, 0x004B, 0x0FF0,
-       0x0FEE, 0x0072, 0x01C1, 0x01A0, 0x004F, 0x0FF0,
-       0x0FEE, 0x006E, 0x01BD, 0x01A4, 0x0053, 0x0FF0,
-       0x0FEE, 0x0069, 0x01B9, 0x01A9, 0x0058, 0x0FEF,
-       0x0FEE, 0x0065, 0x01B5, 0x01AD, 0x005C, 0x0FEF,
-       0x0FEF, 0x0060, 0x01B1, 0x01B1, 0x0060, 0x0FEF,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.6_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.600000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_60[198] = {
-       0x0FD9, 0x00FB, 0x0258, 0x00FB, 0x0FD9, 0x0000,
-       0x0FD9, 0x00F3, 0x0258, 0x0102, 0x0FDA, 0x0000,
-       0x0FD8, 0x00EB, 0x0258, 0x010B, 0x0FDB, 0x0FFF,
-       0x0FD8, 0x00E3, 0x0258, 0x0112, 0x0FDC, 0x0FFF,
-       0x0FD8, 0x00DC, 0x0257, 0x011B, 0x0FDC, 0x0FFE,
-       0x0FD7, 0x00D4, 0x0256, 0x0123, 0x0FDE, 0x0FFE,
-       0x0FD7, 0x00CD, 0x0255, 0x012B, 0x0FDF, 0x0FFD,
-       0x0FD7, 0x00C5, 0x0254, 0x0133, 0x0FE0, 0x0FFD,
-       0x0FD7, 0x00BE, 0x0252, 0x013C, 0x0FE1, 0x0FFC,
-       0x0FD7, 0x00B6, 0x0251, 0x0143, 0x0FE3, 0x0FFC,
-       0x0FD8, 0x00AF, 0x024F, 0x014B, 0x0FE4, 0x0FFB,
-       0x0FD8, 0x00A8, 0x024C, 0x0154, 0x0FE6, 0x0FFA,
-       0x0FD8, 0x00A1, 0x024A, 0x015B, 0x0FE8, 0x0FFA,
-       0x0FD9, 0x009A, 0x0247, 0x0163, 0x0FEA, 0x0FF9,
-       0x0FD9, 0x0093, 0x0244, 0x016C, 0x0FEC, 0x0FF8,
-       0x0FD9, 0x008C, 0x0241, 0x0174, 0x0FEF, 0x0FF7,
-       0x0FDA, 0x0085, 0x023E, 0x017B, 0x0FF1, 0x0FF7,
-       0x0FDB, 0x007F, 0x023A, 0x0183, 0x0FF3, 0x0FF6,
-       0x0FDB, 0x0078, 0x0237, 0x018B, 0x0FF6, 0x0FF5,
-       0x0FDC, 0x0072, 0x0233, 0x0192, 0x0FF9, 0x0FF4,
-       0x0FDD, 0x006C, 0x022F, 0x0199, 0x0FFC, 0x0FF3,
-       0x0FDD, 0x0065, 0x022A, 0x01A3, 0x0FFF, 0x0FF2,
-       0x0FDE, 0x005F, 0x0226, 0x01AA, 0x0002, 0x0FF1,
-       0x0FDF, 0x005A, 0x0221, 0x01B0, 0x0006, 0x0FF0,
-       0x0FE0, 0x0054, 0x021C, 0x01B7, 0x0009, 0x0FF0,
-       0x0FE1, 0x004E, 0x0217, 0x01BE, 0x000D, 0x0FEF,
-       0x0FE2, 0x0048, 0x0212, 0x01C6, 0x0010, 0x0FEE,
-       0x0FE3, 0x0043, 0x020C, 0x01CD, 0x0014, 0x0FED,
-       0x0FE4, 0x003E, 0x0207, 0x01D3, 0x0018, 0x0FEC,
-       0x0FE5, 0x0039, 0x0200, 0x01DA, 0x001D, 0x0FEB,
-       0x0FE6, 0x0034, 0x01FA, 0x01E1, 0x0021, 0x0FEA,
-       0x0FE7, 0x002F, 0x01F5, 0x01E7, 0x0025, 0x0FE9,
-       0x0FE8, 0x002A, 0x01EE, 0x01EE, 0x002A, 0x0FE8,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.7_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.700000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_70[198] = {
-       0x0FC0, 0x00DA, 0x02CC, 0x00DA, 0x0FC0, 0x0000,
-       0x0FC1, 0x00D0, 0x02CC, 0x00E4, 0x0FBF, 0x0000,
-       0x0FC2, 0x00C6, 0x02CB, 0x00EF, 0x0FBE, 0x0000,
-       0x0FC3, 0x00BC, 0x02CA, 0x00F9, 0x0FBE, 0x0000,
-       0x0FC4, 0x00B2, 0x02C9, 0x0104, 0x0FBD, 0x0000,
-       0x0FC5, 0x00A8, 0x02C7, 0x010F, 0x0FBD, 0x0000,
-       0x0FC7, 0x009F, 0x02C5, 0x0119, 0x0FBC, 0x0000,
-       0x0FC8, 0x0095, 0x02C3, 0x0124, 0x0FBC, 0x0000,
-       0x0FC9, 0x008C, 0x02C0, 0x012F, 0x0FBC, 0x0000,
-       0x0FCB, 0x0083, 0x02BD, 0x0139, 0x0FBC, 0x0000,
-       0x0FCC, 0x007A, 0x02BA, 0x0144, 0x0FBC, 0x0000,
-       0x0FCE, 0x0072, 0x02B6, 0x014D, 0x0FBD, 0x0000,
-       0x0FD0, 0x0069, 0x02B2, 0x0159, 0x0FBD, 0x0FFF,
-       0x0FD1, 0x0061, 0x02AD, 0x0164, 0x0FBE, 0x0FFF,
-       0x0FD3, 0x0059, 0x02A9, 0x016E, 0x0FBF, 0x0FFE,
-       0x0FD4, 0x0051, 0x02A4, 0x017A, 0x0FBF, 0x0FFE,
-       0x0FD6, 0x0049, 0x029E, 0x0184, 0x0FC1, 0x0FFE,
-       0x0FD8, 0x0042, 0x0299, 0x018E, 0x0FC2, 0x0FFD,
-       0x0FD9, 0x003A, 0x0293, 0x019B, 0x0FC3, 0x0FFC,
-       0x0FDB, 0x0033, 0x028D, 0x01A4, 0x0FC5, 0x0FFC,
-       0x0FDC, 0x002D, 0x0286, 0x01AF, 0x0FC7, 0x0FFB,
-       0x0FDE, 0x0026, 0x0280, 0x01BA, 0x0FC8, 0x0FFA,
-       0x0FE0, 0x001F, 0x0279, 0x01C4, 0x0FCB, 0x0FF9,
-       0x0FE1, 0x0019, 0x0272, 0x01CE, 0x0FCD, 0x0FF9,
-       0x0FE3, 0x0013, 0x026A, 0x01D9, 0x0FCF, 0x0FF8,
-       0x0FE4, 0x000D, 0x0263, 0x01E3, 0x0FD2, 0x0FF7,
-       0x0FE6, 0x0008, 0x025B, 0x01EC, 0x0FD5, 0x0FF6,
-       0x0FE7, 0x0002, 0x0253, 0x01F7, 0x0FD8, 0x0FF5,
-       0x0FE9, 0x0FFD, 0x024A, 0x0202, 0x0FDB, 0x0FF3,
-       0x0FEA, 0x0FF8, 0x0242, 0x020B, 0x0FDF, 0x0FF2,
-       0x0FEC, 0x0FF3, 0x0239, 0x0215, 0x0FE2, 0x0FF1,
-       0x0FED, 0x0FEF, 0x0230, 0x021E, 0x0FE6, 0x0FF0,
-       0x0FEF, 0x0FEB, 0x0226, 0x0226, 0x0FEB, 0x0FEF,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.8_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.800000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_80[198] = {
-       0x0FBF, 0x00A1, 0x0340, 0x00A1, 0x0FBF, 0x0000,
-       0x0FC1, 0x0095, 0x0340, 0x00AD, 0x0FBC, 0x0001,
-       0x0FC4, 0x0089, 0x033E, 0x00BA, 0x0FBA, 0x0001,
-       0x0FC6, 0x007D, 0x033D, 0x00C6, 0x0FB8, 0x0002,
-       0x0FC9, 0x0072, 0x033A, 0x00D3, 0x0FB6, 0x0002,
-       0x0FCC, 0x0067, 0x0338, 0x00DF, 0x0FB3, 0x0003,
-       0x0FCE, 0x005C, 0x0334, 0x00EE, 0x0FB1, 0x0003,
-       0x0FD1, 0x0051, 0x0331, 0x00FA, 0x0FAF, 0x0004,
-       0x0FD3, 0x0047, 0x032D, 0x0108, 0x0FAD, 0x0004,
-       0x0FD6, 0x003D, 0x0328, 0x0116, 0x0FAB, 0x0004,
-       0x0FD8, 0x0033, 0x0323, 0x0123, 0x0FAA, 0x0005,
-       0x0FDB, 0x002A, 0x031D, 0x0131, 0x0FA8, 0x0005,
-       0x0FDD, 0x0021, 0x0317, 0x013F, 0x0FA7, 0x0005,
-       0x0FDF, 0x0018, 0x0311, 0x014D, 0x0FA5, 0x0006,
-       0x0FE2, 0x0010, 0x030A, 0x015A, 0x0FA4, 0x0006,
-       0x0FE4, 0x0008, 0x0302, 0x0169, 0x0FA3, 0x0006,
-       0x0FE6, 0x0000, 0x02FB, 0x0177, 0x0FA2, 0x0006,
-       0x0FE8, 0x0FF9, 0x02F3, 0x0185, 0x0FA1, 0x0006,
-       0x0FEB, 0x0FF1, 0x02EA, 0x0193, 0x0FA1, 0x0006,
-       0x0FED, 0x0FEB, 0x02E1, 0x01A1, 0x0FA0, 0x0006,
-       0x0FEE, 0x0FE4, 0x02D8, 0x01B0, 0x0FA0, 0x0006,
-       0x0FF0, 0x0FDE, 0x02CE, 0x01BE, 0x0FA0, 0x0006,
-       0x0FF2, 0x0FD8, 0x02C5, 0x01CB, 0x0FA0, 0x0006,
-       0x0FF4, 0x0FD3, 0x02BA, 0x01D8, 0x0FA1, 0x0006,
-       0x0FF6, 0x0FCD, 0x02B0, 0x01E7, 0x0FA1, 0x0005,
-       0x0FF7, 0x0FC8, 0x02A5, 0x01F5, 0x0FA2, 0x0005,
-       0x0FF9, 0x0FC4, 0x029A, 0x0202, 0x0FA3, 0x0004,
-       0x0FFA, 0x0FC0, 0x028E, 0x0210, 0x0FA4, 0x0004,
-       0x0FFB, 0x0FBC, 0x0283, 0x021D, 0x0FA6, 0x0003,
-       0x0FFD, 0x0FB8, 0x0276, 0x022A, 0x0FA8, 0x0003,
-       0x0FFE, 0x0FB4, 0x026B, 0x0237, 0x0FAA, 0x0002,
-       0x0FFF, 0x0FB1, 0x025E, 0x0245, 0x0FAC, 0x0001,
-       0x0000, 0x0FAE, 0x0252, 0x0252, 0x0FAE, 0x0000,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_0.9_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 0.900000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_0_90[198] = {
-       0x0FD8, 0x0055, 0x03A7, 0x0054, 0x0FD8, 0x0000,
-       0x0FDB, 0x0047, 0x03A7, 0x0063, 0x0FD4, 0x0000,
-       0x0FDF, 0x003B, 0x03A5, 0x006F, 0x0FD1, 0x0001,
-       0x0FE2, 0x002E, 0x03A3, 0x007E, 0x0FCD, 0x0002,
-       0x0FE5, 0x0022, 0x03A0, 0x008D, 0x0FCA, 0x0002,
-       0x0FE8, 0x0017, 0x039D, 0x009B, 0x0FC6, 0x0003,
-       0x0FEB, 0x000C, 0x0398, 0x00AC, 0x0FC2, 0x0003,
-       0x0FEE, 0x0001, 0x0394, 0x00BA, 0x0FBF, 0x0004,
-       0x0FF1, 0x0FF7, 0x038E, 0x00CA, 0x0FBB, 0x0005,
-       0x0FF4, 0x0FED, 0x0388, 0x00DA, 0x0FB8, 0x0005,
-       0x0FF6, 0x0FE4, 0x0381, 0x00EB, 0x0FB4, 0x0006,
-       0x0FF9, 0x0FDB, 0x037A, 0x00FA, 0x0FB1, 0x0007,
-       0x0FFB, 0x0FD3, 0x0372, 0x010B, 0x0FAD, 0x0008,
-       0x0FFD, 0x0FCB, 0x0369, 0x011D, 0x0FAA, 0x0008,
-       0x0000, 0x0FC3, 0x0360, 0x012E, 0x0FA6, 0x0009,
-       0x0002, 0x0FBC, 0x0356, 0x013F, 0x0FA3, 0x000A,
-       0x0003, 0x0FB6, 0x034C, 0x0150, 0x0FA0, 0x000B,
-       0x0005, 0x0FB0, 0x0341, 0x0162, 0x0F9D, 0x000B,
-       0x0007, 0x0FAA, 0x0336, 0x0173, 0x0F9A, 0x000C,
-       0x0008, 0x0FA5, 0x032A, 0x0185, 0x0F97, 0x000D,
-       0x000A, 0x0FA0, 0x031E, 0x0197, 0x0F94, 0x000D,
-       0x000B, 0x0F9B, 0x0311, 0x01A9, 0x0F92, 0x000E,
-       0x000C, 0x0F97, 0x0303, 0x01BC, 0x0F8F, 0x000F,
-       0x000D, 0x0F94, 0x02F6, 0x01CD, 0x0F8D, 0x000F,
-       0x000E, 0x0F91, 0x02E8, 0x01DE, 0x0F8B, 0x0010,
-       0x000F, 0x0F8E, 0x02D9, 0x01F1, 0x0F89, 0x0010,
-       0x0010, 0x0F8B, 0x02CA, 0x0202, 0x0F88, 0x0011,
-       0x0010, 0x0F89, 0x02BB, 0x0214, 0x0F87, 0x0011,
-       0x0011, 0x0F87, 0x02AB, 0x0226, 0x0F86, 0x0011,
-       0x0011, 0x0F86, 0x029C, 0x0236, 0x0F85, 0x0012,
-       0x0011, 0x0F85, 0x028B, 0x0249, 0x0F84, 0x0012,
-       0x0012, 0x0F84, 0x027B, 0x0259, 0x0F84, 0x0012,
-       0x0012, 0x0F84, 0x026A, 0x026A, 0x0F84, 0x0012,
-};
-
-//========================================================
-// <using>                     gen_scaler_coeffs_cnf_file.m
-// <using>                     make_test_script.m
-// <date>                      02-Apr-2024
-// <coeffDescrip>      6t_64p_LanczosEd_p_1_p_10qb_
-// <num_taps>          6
-// <num_phases>                64
-// <scale_ratio>        input/output = 1.000000000000
-// <CoefType>          LanczosEd
-// <CoefQuant>         S1.10
-//========================================================
-static const uint16_t easf_filter_6tap_64p_ratio_1_00[198] = {
-       0x0000, 0x0000, 0x0400, 0x0000, 0x0000, 0x0000,
-       0x0003, 0x0FF3, 0x0400, 0x000D, 0x0FFD, 0x0000,
-       0x0006, 0x0FE7, 0x03FE, 0x001C, 0x0FF9, 0x0000,
-       0x0009, 0x0FDB, 0x03FC, 0x002B, 0x0FF5, 0x0000,
-       0x000C, 0x0FD0, 0x03F9, 0x003A, 0x0FF1, 0x0000,
-       0x000E, 0x0FC5, 0x03F5, 0x004A, 0x0FED, 0x0001,
-       0x0011, 0x0FBB, 0x03F0, 0x005A, 0x0FE9, 0x0001,
-       0x0013, 0x0FB2, 0x03EB, 0x006A, 0x0FE5, 0x0001,
-       0x0015, 0x0FA9, 0x03E4, 0x007B, 0x0FE1, 0x0002,
-       0x0017, 0x0FA1, 0x03DD, 0x008D, 0x0FDC, 0x0002,
-       0x0018, 0x0F99, 0x03D4, 0x00A0, 0x0FD8, 0x0003,
-       0x001A, 0x0F92, 0x03CB, 0x00B2, 0x0FD3, 0x0004,
-       0x001B, 0x0F8C, 0x03C1, 0x00C6, 0x0FCE, 0x0004,
-       0x001C, 0x0F86, 0x03B7, 0x00D9, 0x0FC9, 0x0005,
-       0x001D, 0x0F80, 0x03AB, 0x00EE, 0x0FC4, 0x0006,
-       0x001E, 0x0F7C, 0x039F, 0x0101, 0x0FBF, 0x0007,
-       0x001F, 0x0F78, 0x0392, 0x0115, 0x0FBA, 0x0008,
-       0x001F, 0x0F74, 0x0385, 0x012B, 0x0FB5, 0x0008,
-       0x0020, 0x0F71, 0x0376, 0x0140, 0x0FB0, 0x0009,
-       0x0020, 0x0F6E, 0x0367, 0x0155, 0x0FAB, 0x000B,
-       0x0020, 0x0F6C, 0x0357, 0x016B, 0x0FA6, 0x000C,
-       0x0020, 0x0F6A, 0x0347, 0x0180, 0x0FA2, 0x000D,
-       0x0020, 0x0F69, 0x0336, 0x0196, 0x0F9D, 0x000E,
-       0x0020, 0x0F69, 0x0325, 0x01AB, 0x0F98, 0x000F,
-       0x001F, 0x0F68, 0x0313, 0x01C3, 0x0F93, 0x0010,
-       0x001F, 0x0F69, 0x0300, 0x01D8, 0x0F8F, 0x0011,
-       0x001E, 0x0F69, 0x02ED, 0x01EF, 0x0F8B, 0x0012,
-       0x001D, 0x0F6A, 0x02D9, 0x0205, 0x0F87, 0x0014,
-       0x001D, 0x0F6C, 0x02C5, 0x021A, 0x0F83, 0x0015,
-       0x001C, 0x0F6E, 0x02B1, 0x0230, 0x0F7F, 0x0016,
-       0x001B, 0x0F70, 0x029C, 0x0247, 0x0F7B, 0x0017,
-       0x001A, 0x0F72, 0x0287, 0x025D, 0x0F78, 0x0018,
-       0x0019, 0x0F75, 0x0272, 0x0272, 0x0F75, 0x0019,
-};
-
-/* Converted scaler coeff tables from S1.10 to S1.12 */
-static uint16_t easf_filter_3tap_64p_ratio_0_30_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_0_40_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_0_50_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_0_60_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_0_70_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_0_80_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_0_90_s1_12[99];
-static uint16_t easf_filter_3tap_64p_ratio_1_00_s1_12[99];
-static uint16_t easf_filter_4tap_64p_ratio_0_30_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_0_40_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_0_50_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_0_60_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_0_70_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_0_80_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_0_90_s1_12[132];
-static uint16_t easf_filter_4tap_64p_ratio_1_00_s1_12[132];
-static uint16_t easf_filter_6tap_64p_ratio_0_30_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_0_40_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_0_50_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_0_60_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_0_70_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_0_80_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_0_90_s1_12[198];
-static uint16_t easf_filter_6tap_64p_ratio_1_00_s1_12[198];
-
-struct scale_ratio_to_reg_value_lookup easf_v_bf3_mode_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x0000},
-       {9, 10, 0x0000},
-       {1, 1, 0x0000},
-       {-1, -1, 0x0002},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_h_bf3_mode_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x0000},
-       {9, 10, 0x0000},
-       {1, 1, 0x0000},
-       {-1, -1, 0x0002},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_reducer_gain6_6tap_lookup[] = {
-       {3, 10, 0x4100},
-       {4, 10, 0x4100},
-       {5, 10, 0x4100},
-       {6, 10, 0x4100},
-       {7, 10, 0x4100},
-       {8, 10, 0x4100},
-       {9, 10, 0x4100},
-       {1, 1, 0x4100},
-       {-1, -1, 0x4100},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_reducer_gain4_6tap_lookup[] = {
-       {3, 10, 0x4000},
-       {4, 10, 0x4000},
-       {5, 10, 0x4000},
-       {6, 10, 0x4000},
-       {7, 10, 0x4000},
-       {8, 10, 0x4000},
-       {9, 10, 0x4000},
-       {1, 1, 0x4000},
-       {-1, -1, 0x4000},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_gain_ring6_6tap_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x251F},
-       {5, 10, 0x291F},
-       {6, 10, 0xA51F},
-       {7, 10, 0xA51F},
-       {8, 10, 0xAA66},
-       {9, 10, 0xA51F},
-       {1, 1, 0xA640},
-       {-1, -1, 0xA640},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_gain_ring4_6tap_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x9600},
-       {5, 10, 0xA460},
-       {6, 10, 0xA8E0},
-       {7, 10, 0xAC00},
-       {8, 10, 0xAD20},
-       {9, 10, 0xAFC0},
-       {1, 1, 0xB058},
-       {-1, -1, 0xB058},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_reducer_gain6_4tap_lookup[] = {
-       {3, 10, 0x4100},
-       {4, 10, 0x4100},
-       {5, 10, 0x4100},
-       {6, 10, 0x4100},
-       {7, 10, 0x4100},
-       {8, 10, 0x4100},
-       {9, 10, 0x4100},
-       {1, 1, 0x4100},
-       {-1, -1, 0x4100},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_reducer_gain4_4tap_lookup[] = {
-       {3, 10, 0x4000},
-       {4, 10, 0x4000},
-       {5, 10, 0x4000},
-       {6, 10, 0x4000},
-       {7, 10, 0x4000},
-       {8, 10, 0x4000},
-       {9, 10, 0x4000},
-       {1, 1, 0x4000},
-       {-1, -1, 0x4000},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_gain_ring6_4tap_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x0000},
-       {9, 10, 0x0000},
-       {1, 1, 0x0000},
-       {-1, -1, 0x0000},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_gain_ring4_4tap_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x9900},
-       {7, 10, 0xA100},
-       {8, 10, 0xA8C0},
-       {9, 10, 0xAB20},
-       {1, 1, 0xAC00},
-       {-1, -1, 0xAC00},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_3tap_dntilt_uptilt_offset_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x4100},
-       {9, 10, 0x9F00},
-       {1, 1, 0xA4C0},
-       {-1, -1, 0xA8D8},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_3tap_uptilt_maxval_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x4000},
-       {9, 10, 0x24FE},
-       {1, 1, 0x2D64},
-       {-1, -1, 0x3ADB},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_3tap_dntilt_slope_lookup[] = {
-       {3, 10, 0x3800},
-       {4, 10, 0x3800},
-       {5, 10, 0x3800},
-       {6, 10, 0x3800},
-       {7, 10, 0x3800},
-       {8, 10, 0x3886},
-       {9, 10, 0x3940},
-       {1, 1, 0x3A4E},
-       {-1, -1, 0x3B66},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_3tap_uptilt1_slope_lookup[] = {
-       {3, 10, 0x3800},
-       {4, 10, 0x3800},
-       {5, 10, 0x3800},
-       {6, 10, 0x3800},
-       {7, 10, 0x3800},
-       {8, 10, 0x36F4},
-       {9, 10, 0x359C},
-       {1, 1, 0x3360},
-       {-1, -1, 0x2F20},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_3tap_uptilt2_slope_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x0000},
-       {9, 10, 0x359C},
-       {1, 1, 0x31F0},
-       {-1, -1, 0x1F00},
-};
-
-struct scale_ratio_to_reg_value_lookup easf_3tap_uptilt2_offset_lookup[] = {
-       {3, 10, 0x0000},
-       {4, 10, 0x0000},
-       {5, 10, 0x0000},
-       {6, 10, 0x0000},
-       {7, 10, 0x0000},
-       {8, 10, 0x0000},
-       {9, 10, 0x9F00},
-       {1, 1, 0xA400},
-       {-1, -1, 0x9E00},
-};
-
-void spl_init_easf_filter_coeffs(void)
-{
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_30,
-               easf_filter_3tap_64p_ratio_0_30_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_40,
-               easf_filter_3tap_64p_ratio_0_40_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_50,
-               easf_filter_3tap_64p_ratio_0_50_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_60,
-               easf_filter_3tap_64p_ratio_0_60_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_70,
-               easf_filter_3tap_64p_ratio_0_70_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_80,
-               easf_filter_3tap_64p_ratio_0_80_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_0_90,
-               easf_filter_3tap_64p_ratio_0_90_s1_12, 3);
-       convert_filter_s1_10_to_s1_12(easf_filter_3tap_64p_ratio_1_00,
-               easf_filter_3tap_64p_ratio_1_00_s1_12, 3);
-
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_30,
-               easf_filter_4tap_64p_ratio_0_30_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_40,
-               easf_filter_4tap_64p_ratio_0_40_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_50,
-               easf_filter_4tap_64p_ratio_0_50_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_60,
-               easf_filter_4tap_64p_ratio_0_60_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_70,
-               easf_filter_4tap_64p_ratio_0_70_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_80,
-               easf_filter_4tap_64p_ratio_0_80_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_0_90,
-               easf_filter_4tap_64p_ratio_0_90_s1_12, 4);
-       convert_filter_s1_10_to_s1_12(easf_filter_4tap_64p_ratio_1_00,
-               easf_filter_4tap_64p_ratio_1_00_s1_12, 4);
-
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_30,
-               easf_filter_6tap_64p_ratio_0_30_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_40,
-               easf_filter_6tap_64p_ratio_0_40_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_50,
-               easf_filter_6tap_64p_ratio_0_50_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_60,
-               easf_filter_6tap_64p_ratio_0_60_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_70,
-               easf_filter_6tap_64p_ratio_0_70_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_80,
-               easf_filter_6tap_64p_ratio_0_80_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_0_90,
-               easf_filter_6tap_64p_ratio_0_90_s1_12, 6);
-       convert_filter_s1_10_to_s1_12(easf_filter_6tap_64p_ratio_1_00,
-               easf_filter_6tap_64p_ratio_1_00_s1_12, 6);
-}
-
-uint16_t *spl_get_easf_filter_3tap_64p(struct spl_fixed31_32 ratio)
-{
-       if (ratio.value < spl_fixpt_from_fraction(3, 10).value)
-               return easf_filter_3tap_64p_ratio_0_30_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 10).value)
-               return easf_filter_3tap_64p_ratio_0_40_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 10).value)
-               return easf_filter_3tap_64p_ratio_0_50_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(6, 10).value)
-               return easf_filter_3tap_64p_ratio_0_60_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(7, 10).value)
-               return easf_filter_3tap_64p_ratio_0_70_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(8, 10).value)
-               return easf_filter_3tap_64p_ratio_0_80_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(9, 10).value)
-               return easf_filter_3tap_64p_ratio_0_90_s1_12;
-       else
-               return easf_filter_3tap_64p_ratio_1_00_s1_12;
-}
-
-uint16_t *spl_get_easf_filter_4tap_64p(struct spl_fixed31_32 ratio)
-{
-       if (ratio.value < spl_fixpt_from_fraction(3, 10).value)
-               return easf_filter_4tap_64p_ratio_0_30_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 10).value)
-               return easf_filter_4tap_64p_ratio_0_40_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 10).value)
-               return easf_filter_4tap_64p_ratio_0_50_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(6, 10).value)
-               return easf_filter_4tap_64p_ratio_0_60_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(7, 10).value)
-               return easf_filter_4tap_64p_ratio_0_70_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(8, 10).value)
-               return easf_filter_4tap_64p_ratio_0_80_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(9, 10).value)
-               return easf_filter_4tap_64p_ratio_0_90_s1_12;
-       else
-               return easf_filter_4tap_64p_ratio_1_00_s1_12;
-}
-
-uint16_t *spl_get_easf_filter_6tap_64p(struct spl_fixed31_32 ratio)
-{
-       if (ratio.value < spl_fixpt_from_fraction(3, 10).value)
-               return easf_filter_6tap_64p_ratio_0_30_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 10).value)
-               return easf_filter_6tap_64p_ratio_0_40_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 10).value)
-               return easf_filter_6tap_64p_ratio_0_50_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(6, 10).value)
-               return easf_filter_6tap_64p_ratio_0_60_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(7, 10).value)
-               return easf_filter_6tap_64p_ratio_0_70_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(8, 10).value)
-               return easf_filter_6tap_64p_ratio_0_80_s1_12;
-       else if (ratio.value < spl_fixpt_from_fraction(9, 10).value)
-               return easf_filter_6tap_64p_ratio_0_90_s1_12;
-       else
-               return easf_filter_6tap_64p_ratio_1_00_s1_12;
-}
-
-uint16_t *spl_dscl_get_easf_filter_coeffs_64p(int taps, struct spl_fixed31_32 ratio)
-{
-       if (taps == 6)
-               return spl_get_easf_filter_6tap_64p(ratio);
-       else if (taps == 4)
-               return spl_get_easf_filter_4tap_64p(ratio);
-       else if (taps == 3)
-               return spl_get_easf_filter_3tap_64p(ratio);
-       else {
-               /* should never happen, bug */
-               SPL_BREAK_TO_DEBUGGER();
-               return NULL;
-       }
-}
-
-void spl_set_filters_data(struct dscl_prog_data *dscl_prog_data,
-               const struct spl_scaler_data *data, bool enable_easf_v,
-               bool enable_easf_h)
-{
-       /*
-        * Old coefficients calculated scaling ratio = input / output
-        * New coefficients are calculated based on = output / input
-        */
-       if (enable_easf_h) {
-               dscl_prog_data->filter_h = spl_dscl_get_easf_filter_coeffs_64p(
-                       data->taps.h_taps, data->recip_ratios.horz);
-
-               dscl_prog_data->filter_h_c = spl_dscl_get_easf_filter_coeffs_64p(
-                       data->taps.h_taps_c, data->recip_ratios.horz_c);
-       } else {
-               dscl_prog_data->filter_h = spl_dscl_get_filter_coeffs_64p(
-                       data->taps.h_taps, data->ratios.horz);
-
-               dscl_prog_data->filter_h_c = spl_dscl_get_filter_coeffs_64p(
-                       data->taps.h_taps_c, data->ratios.horz_c);
-       }
-       if (enable_easf_v) {
-               dscl_prog_data->filter_v = spl_dscl_get_easf_filter_coeffs_64p(
-                       data->taps.v_taps, data->recip_ratios.vert);
-
-               dscl_prog_data->filter_v_c = spl_dscl_get_easf_filter_coeffs_64p(
-                       data->taps.v_taps_c, data->recip_ratios.vert_c);
-       } else {
-               dscl_prog_data->filter_v = spl_dscl_get_filter_coeffs_64p(
-                       data->taps.v_taps, data->ratios.vert);
-
-               dscl_prog_data->filter_v_c = spl_dscl_get_filter_coeffs_64p(
-                       data->taps.v_taps_c, data->ratios.vert_c);
-       }
-}
-
-static uint32_t spl_easf_get_scale_ratio_to_reg_value(struct spl_fixed31_32 ratio,
-       struct scale_ratio_to_reg_value_lookup *lookup_table_base_ptr,
-       unsigned int num_entries)
-{
-       unsigned int count = 0;
-       uint32_t value = 0;
-       struct scale_ratio_to_reg_value_lookup *lookup_table_index_ptr;
-
-       lookup_table_index_ptr = (lookup_table_base_ptr + num_entries - 1);
-       value = lookup_table_index_ptr->reg_value;
-
-       while (count < num_entries) {
-
-               lookup_table_index_ptr = (lookup_table_base_ptr + count);
-               if (lookup_table_index_ptr->numer < 0)
-                       break;
-
-               if (ratio.value < spl_fixpt_from_fraction(
-                       lookup_table_index_ptr->numer,
-                       lookup_table_index_ptr->denom).value) {
-                       value = lookup_table_index_ptr->reg_value;
-                       break;
-               }
-
-               count++;
-       }
-       return value;
-}
-uint32_t spl_get_v_bf3_mode(struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries = sizeof(easf_v_bf3_mode_lookup) /
-               sizeof(struct scale_ratio_to_reg_value_lookup);
-       value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-               easf_v_bf3_mode_lookup, num_entries);
-       return value;
-}
-uint32_t spl_get_h_bf3_mode(struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries = sizeof(easf_h_bf3_mode_lookup) /
-               sizeof(struct scale_ratio_to_reg_value_lookup);
-       value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-               easf_h_bf3_mode_lookup, num_entries);
-       return value;
-}
-uint32_t spl_get_reducer_gain6(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 4) {
-               num_entries = sizeof(easf_reducer_gain6_4tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_reducer_gain6_4tap_lookup, num_entries);
-       } else if (taps == 6) {
-               num_entries = sizeof(easf_reducer_gain6_6tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_reducer_gain6_6tap_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_reducer_gain4(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 4) {
-               num_entries = sizeof(easf_reducer_gain4_4tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_reducer_gain4_4tap_lookup, num_entries);
-       } else if (taps == 6) {
-               num_entries = sizeof(easf_reducer_gain4_6tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_reducer_gain4_6tap_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_gainRing6(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 4) {
-               num_entries = sizeof(easf_gain_ring6_4tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_gain_ring6_4tap_lookup, num_entries);
-       } else if (taps == 6) {
-               num_entries = sizeof(easf_gain_ring6_6tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_gain_ring6_6tap_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_gainRing4(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 4) {
-               num_entries = sizeof(easf_gain_ring4_4tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_gain_ring4_4tap_lookup, num_entries);
-       } else if (taps == 6) {
-               num_entries = sizeof(easf_gain_ring4_6tap_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_gain_ring4_6tap_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_3tap_dntilt_uptilt_offset(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 3) {
-               num_entries = sizeof(easf_3tap_dntilt_uptilt_offset_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_3tap_dntilt_uptilt_offset_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_3tap_uptilt_maxval(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 3) {
-               num_entries = sizeof(easf_3tap_uptilt_maxval_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_3tap_uptilt_maxval_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_3tap_dntilt_slope(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 3) {
-               num_entries = sizeof(easf_3tap_dntilt_slope_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_3tap_dntilt_slope_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_3tap_uptilt1_slope(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 3) {
-               num_entries = sizeof(easf_3tap_uptilt1_slope_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_3tap_uptilt1_slope_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_3tap_uptilt2_slope(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 3) {
-               num_entries = sizeof(easf_3tap_uptilt2_slope_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_3tap_uptilt2_slope_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
-uint32_t spl_get_3tap_uptilt2_offset(int taps, struct spl_fixed31_32 ratio)
-{
-       uint32_t value;
-       unsigned int num_entries;
-
-       if (taps == 3) {
-               num_entries = sizeof(easf_3tap_uptilt2_offset_lookup) /
-                       sizeof(struct scale_ratio_to_reg_value_lookup);
-               value = spl_easf_get_scale_ratio_to_reg_value(ratio,
-                       easf_3tap_uptilt2_offset_lookup, num_entries);
-       } else
-               value = 0;
-       return value;
-}
diff --git a/drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_easf_filters.h b/drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_easf_filters.h
deleted file mode 100644 (file)
index 8bb2b81..0000000
+++ /dev/null
@@ -1,38 +0,0 @@
-/* SPDX-License-Identifier: MIT */
-
-/* Copyright 2024 Advanced Micro Devices, Inc. */
-
-#ifndef __DC_SPL_SCL_EASF_FILTERS_H__
-#define __DC_SPL_SCL_EASF_FILTERS_H__
-
-#include "dc_spl_types.h"
-
-struct scale_ratio_to_reg_value_lookup {
-       int numer;
-       int denom;
-       const uint32_t reg_value;
-};
-
-void spl_init_easf_filter_coeffs(void);
-uint16_t *spl_get_easf_filter_3tap_64p(struct spl_fixed31_32 ratio);
-uint16_t *spl_get_easf_filter_4tap_64p(struct spl_fixed31_32 ratio);
-uint16_t *spl_get_easf_filter_6tap_64p(struct spl_fixed31_32 ratio);
-uint16_t *spl_dscl_get_easf_filter_coeffs_64p(int taps, struct spl_fixed31_32 ratio);
-void spl_set_filters_data(struct dscl_prog_data *dscl_prog_data,
-       const struct spl_scaler_data *data, bool enable_easf_v,
-       bool enable_easf_h);
-
-uint32_t spl_get_v_bf3_mode(struct spl_fixed31_32 ratio);
-uint32_t spl_get_h_bf3_mode(struct spl_fixed31_32 ratio);
-uint32_t spl_get_reducer_gain6(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_reducer_gain4(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_gainRing6(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_gainRing4(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_3tap_dntilt_uptilt_offset(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_3tap_uptilt_maxval(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_3tap_dntilt_slope(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_3tap_uptilt1_slope(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_3tap_uptilt2_slope(int taps, struct spl_fixed31_32 ratio);
-uint32_t spl_get_3tap_uptilt2_offset(int taps, struct spl_fixed31_32 ratio);
-
-#endif /* __DC_SPL_SCL_EASF_FILTERS_H__ */
index b9a7b77a7167aa1ed2fe1388d2176be61b3aeaa8..e2baaf5841396b8298c21f38dc3a2f8a37247b4a 100644 (file)
@@ -2,8 +2,6 @@
 //
 // Copyright 2024 Advanced Micro Devices, Inc.
 
-#include "dc_spl_types.h"
-#include "spl_debug.h"
 #include "dc_spl_scl_filters.h"
 //=========================================
 // <num_taps>    = 2
@@ -1319,97 +1317,97 @@ static const uint16_t filter_8tap_64p_183[264] = {
                0x3FD4, 0x3F84, 0x0214, 0x0694, 0x0694, 0x0214, 0x3F84, 0x3FD4
 };
 
-const uint16_t *spl_get_filter_3tap_16p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_3tap_16p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_3tap_16p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_3tap_16p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_3tap_16p_149;
        else
                return filter_3tap_16p_183;
 }
 
-const uint16_t *spl_get_filter_3tap_64p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_3tap_64p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_3tap_64p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_3tap_64p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_3tap_64p_149;
        else
                return filter_3tap_64p_183;
 }
 
-const uint16_t *spl_get_filter_4tap_16p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_4tap_16p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_4tap_16p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_4tap_16p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_4tap_16p_149;
        else
                return filter_4tap_16p_183;
 }
 
-const uint16_t *spl_get_filter_4tap_64p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_4tap_64p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_4tap_64p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_4tap_64p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_4tap_64p_149;
        else
                return filter_4tap_64p_183;
 }
 
-const uint16_t *spl_get_filter_5tap_64p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_5tap_64p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_5tap_64p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_5tap_64p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_5tap_64p_149;
        else
                return filter_5tap_64p_183;
 }
 
-const uint16_t *spl_get_filter_6tap_64p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_6tap_64p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_6tap_64p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_6tap_64p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_6tap_64p_149;
        else
                return filter_6tap_64p_183;
 }
 
-const uint16_t *spl_get_filter_7tap_64p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_7tap_64p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_7tap_64p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_7tap_64p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_7tap_64p_149;
        else
                return filter_7tap_64p_183;
 }
 
-const uint16_t *spl_get_filter_8tap_64p(struct spl_fixed31_32 ratio)
+const uint16_t *spl_get_filter_8tap_64p(struct fixed31_32 ratio)
 {
-       if (ratio.value < spl_fixpt_one.value)
+       if (ratio.value < dc_fixpt_one.value)
                return filter_8tap_64p_upscale;
-       else if (ratio.value < spl_fixpt_from_fraction(4, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(4, 3).value)
                return filter_8tap_64p_116;
-       else if (ratio.value < spl_fixpt_from_fraction(5, 3).value)
+       else if (ratio.value < dc_fixpt_from_fraction(5, 3).value)
                return filter_8tap_64p_149;
        else
                return filter_8tap_64p_183;
@@ -1424,29 +1422,3 @@ const uint16_t *spl_get_filter_2tap_64p(void)
 {
        return filter_2tap_64p;
 }
-
-const uint16_t *spl_dscl_get_filter_coeffs_64p(int taps, struct spl_fixed31_32 ratio)
-{
-       if (taps == 8)
-               return spl_get_filter_8tap_64p(ratio);
-       else if (taps == 7)
-               return spl_get_filter_7tap_64p(ratio);
-       else if (taps == 6)
-               return spl_get_filter_6tap_64p(ratio);
-       else if (taps == 5)
-               return spl_get_filter_5tap_64p(ratio);
-       else if (taps == 4)
-               return spl_get_filter_4tap_64p(ratio);
-       else if (taps == 3)
-               return spl_get_filter_3tap_64p(ratio);
-       else if (taps == 2)
-               return spl_get_filter_2tap_64p();
-       else if (taps == 1)
-               return NULL;
-       else {
-               /* should never happen, bug */
-               SPL_BREAK_TO_DEBUGGER();
-               return NULL;
-       }
-}
-
index 48202bc4f81e81bcf94d91aa9495b460e2a54ee8..6d96aca53b24dbaff5ae56fbde418af803c3e9bf 100644 (file)
@@ -7,16 +7,53 @@
 
 #include "dc_spl_types.h"
 
-const uint16_t *spl_get_filter_3tap_16p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_3tap_64p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_4tap_16p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_4tap_64p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_5tap_64p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_6tap_64p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_7tap_64p(struct spl_fixed31_32 ratio);
-const uint16_t *spl_get_filter_8tap_64p(struct spl_fixed31_32 ratio);
+const uint16_t *spl_get_filter_3tap_16p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_3tap_64p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_4tap_16p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_4tap_64p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_5tap_64p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_6tap_64p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_7tap_64p(struct fixed31_32 ratio);
+const uint16_t *spl_get_filter_8tap_64p(struct fixed31_32 ratio);
 const uint16_t *spl_get_filter_2tap_16p(void);
 const uint16_t *spl_get_filter_2tap_64p(void);
-const uint16_t *spl_dscl_get_filter_coeffs_64p(int taps, struct spl_fixed31_32 ratio);
+const uint16_t *spl_get_filter_3tap_16p_upscale(void);
+const uint16_t *spl_get_filter_3tap_16p_116(void);
+const uint16_t *spl_get_filter_3tap_16p_149(void);
+const uint16_t *spl_get_filter_3tap_16p_183(void);
 
+const uint16_t *spl_get_filter_4tap_16p_upscale(void);
+const uint16_t *spl_get_filter_4tap_16p_116(void);
+const uint16_t *spl_get_filter_4tap_16p_149(void);
+const uint16_t *spl_get_filter_4tap_16p_183(void);
+
+const uint16_t *spl_get_filter_3tap_64p_upscale(void);
+const uint16_t *spl_get_filter_3tap_64p_116(void);
+const uint16_t *spl_get_filter_3tap_64p_149(void);
+const uint16_t *spl_get_filter_3tap_64p_183(void);
+
+const uint16_t *spl_get_filter_4tap_64p_upscale(void);
+const uint16_t *spl_get_filter_4tap_64p_116(void);
+const uint16_t *spl_get_filter_4tap_64p_149(void);
+const uint16_t *spl_get_filter_4tap_64p_183(void);
+
+const uint16_t *spl_get_filter_5tap_64p_upscale(void);
+const uint16_t *spl_get_filter_5tap_64p_116(void);
+const uint16_t *spl_get_filter_5tap_64p_149(void);
+const uint16_t *spl_get_filter_5tap_64p_183(void);
+
+const uint16_t *spl_get_filter_6tap_64p_upscale(void);
+const uint16_t *spl_get_filter_6tap_64p_116(void);
+const uint16_t *spl_get_filter_6tap_64p_149(void);
+const uint16_t *spl_get_filter_6tap_64p_183(void);
+
+const uint16_t *spl_get_filter_7tap_64p_upscale(void);
+const uint16_t *spl_get_filter_7tap_64p_116(void);
+const uint16_t *spl_get_filter_7tap_64p_149(void);
+const uint16_t *spl_get_filter_7tap_64p_183(void);
+
+const uint16_t *spl_get_filter_8tap_64p_upscale(void);
+const uint16_t *spl_get_filter_8tap_64p_116(void);
+const uint16_t *spl_get_filter_8tap_64p_149(void);
+const uint16_t *spl_get_filter_8tap_64p_183(void);
 #endif /* __DC_SPL_SCL_FILTERS_H__ */
diff --git a/drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_filters_old.c b/drivers/gpu/drm/amd/display/dc/spl/dc_spl_scl_filters_old.c
new file mode 100644 (file)
index 0000000..bb0e1b8
--- /dev/null
@@ -0,0 +1,25 @@
+/*
+ * Copyright 2012-16 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors: AMD
+ *
+ */
+
index 1438a86826a4e420c9718d1c2fa57b8401cc64be..caaa9ced2ec4a4edf8df09a6371c8127dceaa578 100644 (file)
@@ -2,15 +2,14 @@
 //
 // Copyright 2024 Advanced Micro Devices, Inc.
 
+#include "os_types.h"   // swap
+#ifndef ASSERT
+#define ASSERT(_bool) ((void *)0)
+#endif
+#include "include/fixed31_32.h"        // fixed31_32 and related functions
 #ifndef __DC_SPL_TYPES_H__
 #define __DC_SPL_TYPES_H__
 
-#include "spl_os_types.h"   // swap
-#ifndef SPL_ASSERT
-#define SPL_ASSERT(_bool) ((void *)0)
-#endif
-#include "spl_fixpt31_32.h"    // fixed31_32 and related functions
-
 enum lb_memory_config {
        /* Enable all 3 pieces of memory */
        LB_MEMORY_CONFIG_0 = 0,
@@ -39,16 +38,16 @@ struct spl_rect     {
 };
 
 struct spl_ratios {
-       struct spl_fixed31_32 horz;
-       struct spl_fixed31_32 vert;
-       struct spl_fixed31_32 horz_c;
-       struct spl_fixed31_32 vert_c;
+       struct fixed31_32 horz;
+       struct fixed31_32 vert;
+       struct fixed31_32 horz_c;
+       struct fixed31_32 vert_c;
 };
 struct spl_inits {
-       struct spl_fixed31_32 h;
-       struct spl_fixed31_32 h_c;
-       struct spl_fixed31_32 v;
-       struct spl_fixed31_32 v_c;
+       struct fixed31_32 h;
+       struct fixed31_32 h_c;
+       struct fixed31_32 v;
+       struct fixed31_32 v_c;
 };
 
 struct spl_taps        {
@@ -81,8 +80,6 @@ enum spl_pixel_format {
        SPL_PIXEL_FORMAT_420BPP10,
        /*end of pixel format definition*/
        SPL_PIXEL_FORMAT_INVALID,
-       SPL_PIXEL_FORMAT_422BPP8,
-       SPL_PIXEL_FORMAT_422BPP10,
        SPL_PIXEL_FORMAT_GRPH_BEGIN = SPL_PIXEL_FORMAT_INDEX8,
        SPL_PIXEL_FORMAT_GRPH_END = SPL_PIXEL_FORMAT_FP16,
        SPL_PIXEL_FORMAT_VIDEO_BEGIN = SPL_PIXEL_FORMAT_420BPP8,
@@ -138,7 +135,6 @@ struct spl_scaler_data {
        struct spl_rect viewport_c;
        struct spl_rect recout;
        struct spl_ratios ratios;
-       struct spl_ratios recip_ratios;
        struct spl_inits inits;
 };
 
@@ -408,16 +404,11 @@ struct dscl_prog_data {
        const uint16_t *filter_blur_scale_h;
 };
 
-/* SPL input and output definitions */
-// SPL scratch struct
-struct spl_scratch {
-       // Pack all SPL outputs in scl_data
-       struct spl_scaler_data scl_data;
-};
-
 /* SPL input and output definitions */
 // SPL outputs struct
 struct spl_out {
+       // Pack all SPL outputs in scl_data
+       struct spl_scaler_data scl_data;
        // Pack all output need to program hw registers
        struct dscl_prog_data *dscl_prog_data;
 };
@@ -500,10 +491,6 @@ struct spl_in      {
        bool prefer_easf;
        bool disable_easf;
        struct spl_debug debug;
-       bool is_fullscreen;
-       bool is_hdr_on;
-       int h_active;
-       int v_active;
 };
 // end of SPL inputs
 
diff --git a/drivers/gpu/drm/amd/display/dc/spl/spl_debug.h b/drivers/gpu/drm/amd/display/dc/spl/spl_debug.h
deleted file mode 100644 (file)
index a36239a..0000000
+++ /dev/null
@@ -1,23 +0,0 @@
-/* Copyright Â© 1997-2004 Advanced Micro Devices, Inc. All rights reserved. */
-
-#ifndef SPL_DEBUG_H
-#define SPL_DEBUG_H
-
-#ifdef SPL_ASSERT
-#undef SPL_ASSERT
-#endif
-#define SPL_ASSERT(b)
-
-#define SPL_ASSERT_CRITICAL(expr)  do {if (expr)/* Do nothing */; } while (0)
-
-#ifdef SPL_DALMSG
-#undef SPL_DALMSG
-#endif
-#define SPL_DALMSG(b)
-
-#ifdef SPL_DAL_ASSERT_MSG
-#undef SPL_DAL_ASSERT_MSG
-#endif
-#define SPL_DAL_ASSERT_MSG(b, m)
-
-#endif  // SPL_DEBUG_H
diff --git a/drivers/gpu/drm/amd/display/dc/spl/spl_fixpt31_32.c b/drivers/gpu/drm/amd/display/dc/spl/spl_fixpt31_32.c
deleted file mode 100644 (file)
index 2bb1de8..0000000
+++ /dev/null
@@ -1,518 +0,0 @@
-/*
- * Copyright 2012-15 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: AMD
- *
- */
-
-#include "spl_fixpt31_32.h"
-
-static const struct spl_fixed31_32 spl_fixpt_two_pi = { 26986075409LL };
-static const struct spl_fixed31_32 spl_fixpt_ln2 = { 2977044471LL };
-static const struct spl_fixed31_32 spl_fixpt_ln2_div_2 = { 1488522236LL };
-
-static inline unsigned long long abs_i64(
-       long long arg)
-{
-       if (arg > 0)
-               return (unsigned long long)arg;
-       else
-               return (unsigned long long)(-arg);
-}
-
-/*
- * @brief
- * result = dividend / divisor
- * *remainder = dividend % divisor
- */
-static inline unsigned long long complete_integer_division_u64(
-       unsigned long long dividend,
-       unsigned long long divisor,
-       unsigned long long *remainder)
-{
-       unsigned long long result;
-
-       ASSERT(divisor);
-
-       result = spl_div64_u64_rem(dividend, divisor, remainder);
-
-       return result;
-}
-
-
-#define FRACTIONAL_PART_MASK \
-       ((1ULL << FIXED31_32_BITS_PER_FRACTIONAL_PART) - 1)
-
-#define GET_INTEGER_PART(x) \
-       ((x) >> FIXED31_32_BITS_PER_FRACTIONAL_PART)
-
-#define GET_FRACTIONAL_PART(x) \
-       (FRACTIONAL_PART_MASK & (x))
-
-struct spl_fixed31_32 spl_fixpt_from_fraction(long long numerator, long long denominator)
-{
-       struct spl_fixed31_32 res;
-
-       bool arg1_negative = numerator < 0;
-       bool arg2_negative = denominator < 0;
-
-       unsigned long long arg1_value = arg1_negative ? -numerator : numerator;
-       unsigned long long arg2_value = arg2_negative ? -denominator : denominator;
-
-       unsigned long long remainder;
-
-       /* determine integer part */
-
-       unsigned long long res_value = complete_integer_division_u64(
-               arg1_value, arg2_value, &remainder);
-
-       ASSERT(res_value <= LONG_MAX);
-
-       /* determine fractional part */
-       {
-               unsigned int i = FIXED31_32_BITS_PER_FRACTIONAL_PART;
-
-               do {
-                       remainder <<= 1;
-
-                       res_value <<= 1;
-
-                       if (remainder >= arg2_value) {
-                               res_value |= 1;
-                               remainder -= arg2_value;
-                       }
-               } while (--i != 0);
-       }
-
-       /* round up LSB */
-       {
-               unsigned long long summand = (remainder << 1) >= arg2_value;
-
-               ASSERT(res_value <= LLONG_MAX - summand);
-
-               res_value += summand;
-       }
-
-       res.value = (long long)res_value;
-
-       if (arg1_negative ^ arg2_negative)
-               res.value = -res.value;
-
-       return res;
-}
-
-struct spl_fixed31_32 spl_fixpt_mul(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       struct spl_fixed31_32 res;
-
-       bool arg1_negative = arg1.value < 0;
-       bool arg2_negative = arg2.value < 0;
-
-       unsigned long long arg1_value = arg1_negative ? -arg1.value : arg1.value;
-       unsigned long long arg2_value = arg2_negative ? -arg2.value : arg2.value;
-
-       unsigned long long arg1_int = GET_INTEGER_PART(arg1_value);
-       unsigned long long arg2_int = GET_INTEGER_PART(arg2_value);
-
-       unsigned long long arg1_fra = GET_FRACTIONAL_PART(arg1_value);
-       unsigned long long arg2_fra = GET_FRACTIONAL_PART(arg2_value);
-
-       unsigned long long tmp;
-
-       res.value = arg1_int * arg2_int;
-
-       ASSERT(res.value <= (long long)LONG_MAX);
-
-       res.value <<= FIXED31_32_BITS_PER_FRACTIONAL_PART;
-
-       tmp = arg1_int * arg2_fra;
-
-       ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
-
-       res.value += tmp;
-
-       tmp = arg2_int * arg1_fra;
-
-       ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
-
-       res.value += tmp;
-
-       tmp = arg1_fra * arg2_fra;
-
-       tmp = (tmp >> FIXED31_32_BITS_PER_FRACTIONAL_PART) +
-               (tmp >= (unsigned long long)spl_fixpt_half.value);
-
-       ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
-
-       res.value += tmp;
-
-       if (arg1_negative ^ arg2_negative)
-               res.value = -res.value;
-
-       return res;
-}
-
-struct spl_fixed31_32 spl_fixpt_sqr(struct spl_fixed31_32 arg)
-{
-       struct spl_fixed31_32 res;
-
-       unsigned long long arg_value = abs_i64(arg.value);
-
-       unsigned long long arg_int = GET_INTEGER_PART(arg_value);
-
-       unsigned long long arg_fra = GET_FRACTIONAL_PART(arg_value);
-
-       unsigned long long tmp;
-
-       res.value = arg_int * arg_int;
-
-       ASSERT(res.value <= (long long)LONG_MAX);
-
-       res.value <<= FIXED31_32_BITS_PER_FRACTIONAL_PART;
-
-       tmp = arg_int * arg_fra;
-
-       ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
-
-       res.value += tmp;
-
-       ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
-
-       res.value += tmp;
-
-       tmp = arg_fra * arg_fra;
-
-       tmp = (tmp >> FIXED31_32_BITS_PER_FRACTIONAL_PART) +
-               (tmp >= (unsigned long long)spl_fixpt_half.value);
-
-       ASSERT(tmp <= (unsigned long long)(LLONG_MAX - res.value));
-
-       res.value += tmp;
-
-       return res;
-}
-
-struct spl_fixed31_32 spl_fixpt_recip(struct spl_fixed31_32 arg)
-{
-       /*
-        * @note
-        * Good idea to use Newton's method
-        */
-
-       ASSERT(arg.value);
-
-       return spl_fixpt_from_fraction(
-               spl_fixpt_one.value,
-               arg.value);
-}
-
-struct spl_fixed31_32 spl_fixpt_sinc(struct spl_fixed31_32 arg)
-{
-       struct spl_fixed31_32 square;
-
-       struct spl_fixed31_32 res = spl_fixpt_one;
-
-       int n = 27;
-
-       struct spl_fixed31_32 arg_norm = arg;
-
-       if (spl_fixpt_le(
-               spl_fixpt_two_pi,
-               spl_fixpt_abs(arg))) {
-               arg_norm = spl_fixpt_sub(
-                       arg_norm,
-                       spl_fixpt_mul_int(
-                               spl_fixpt_two_pi,
-                               (int)spl_div64_s64(
-                                       arg_norm.value,
-                                       spl_fixpt_two_pi.value)));
-       }
-
-       square = spl_fixpt_sqr(arg_norm);
-
-       do {
-               res = spl_fixpt_sub(
-                       spl_fixpt_one,
-                       spl_fixpt_div_int(
-                               spl_fixpt_mul(
-                                       square,
-                                       res),
-                               n * (n - 1)));
-
-               n -= 2;
-       } while (n > 2);
-
-       if (arg.value != arg_norm.value)
-               res = spl_fixpt_div(
-                       spl_fixpt_mul(res, arg_norm),
-                       arg);
-
-       return res;
-}
-
-struct spl_fixed31_32 spl_fixpt_sin(struct spl_fixed31_32 arg)
-{
-       return spl_fixpt_mul(
-               arg,
-               spl_fixpt_sinc(arg));
-}
-
-struct spl_fixed31_32 spl_fixpt_cos(struct spl_fixed31_32 arg)
-{
-       /* TODO implement argument normalization */
-
-       const struct spl_fixed31_32 square = spl_fixpt_sqr(arg);
-
-       struct spl_fixed31_32 res = spl_fixpt_one;
-
-       int n = 26;
-
-       do {
-               res = spl_fixpt_sub(
-                       spl_fixpt_one,
-                       spl_fixpt_div_int(
-                               spl_fixpt_mul(
-                                       square,
-                                       res),
-                               n * (n - 1)));
-
-               n -= 2;
-       } while (n != 0);
-
-       return res;
-}
-
-/*
- * @brief
- * result = exp(arg),
- * where abs(arg) < 1
- *
- * Calculated as Taylor series.
- */
-static struct spl_fixed31_32 fixed31_32_exp_from_taylor_series(struct spl_fixed31_32 arg)
-{
-       unsigned int n = 9;
-
-       struct spl_fixed31_32 res = spl_fixpt_from_fraction(
-               n + 2,
-               n + 1);
-       /* TODO find correct res */
-
-       ASSERT(spl_fixpt_lt(arg, spl_fixpt_one));
-
-       do
-               res = spl_fixpt_add(
-                       spl_fixpt_one,
-                       spl_fixpt_div_int(
-                               spl_fixpt_mul(
-                                       arg,
-                                       res),
-                               n));
-       while (--n != 1);
-
-       return spl_fixpt_add(
-               spl_fixpt_one,
-               spl_fixpt_mul(
-                       arg,
-                       res));
-}
-
-struct spl_fixed31_32 spl_fixpt_exp(struct spl_fixed31_32 arg)
-{
-       /*
-        * @brief
-        * Main equation is:
-        * exp(x) = exp(r + m * ln(2)) = (1 << m) * exp(r),
-        * where m = round(x / ln(2)), r = x - m * ln(2)
-        */
-
-       if (spl_fixpt_le(
-               spl_fixpt_ln2_div_2,
-               spl_fixpt_abs(arg))) {
-               int m = spl_fixpt_round(
-                       spl_fixpt_div(
-                               arg,
-                               spl_fixpt_ln2));
-
-               struct spl_fixed31_32 r = spl_fixpt_sub(
-                       arg,
-                       spl_fixpt_mul_int(
-                               spl_fixpt_ln2,
-                               m));
-
-               ASSERT(m != 0);
-
-               ASSERT(spl_fixpt_lt(
-                       spl_fixpt_abs(r),
-                       spl_fixpt_one));
-
-               if (m > 0)
-                       return spl_fixpt_shl(
-                               fixed31_32_exp_from_taylor_series(r),
-                               (unsigned char)m);
-               else
-                       return spl_fixpt_div_int(
-                               fixed31_32_exp_from_taylor_series(r),
-                               1LL << -m);
-       } else if (arg.value != 0)
-               return fixed31_32_exp_from_taylor_series(arg);
-       else
-               return spl_fixpt_one;
-}
-
-struct spl_fixed31_32 spl_fixpt_log(struct spl_fixed31_32 arg)
-{
-       struct spl_fixed31_32 res = spl_fixpt_neg(spl_fixpt_one);
-       /* TODO improve 1st estimation */
-
-       struct spl_fixed31_32 error;
-
-       ASSERT(arg.value > 0);
-       /* TODO if arg is negative, return NaN */
-       /* TODO if arg is zero, return -INF */
-
-       do {
-               struct spl_fixed31_32 res1 = spl_fixpt_add(
-                       spl_fixpt_sub(
-                               res,
-                               spl_fixpt_one),
-                       spl_fixpt_div(
-                               arg,
-                               spl_fixpt_exp(res)));
-
-               error = spl_fixpt_sub(
-                       res,
-                       res1);
-
-               res = res1;
-               /* TODO determine max_allowed_error based on quality of exp() */
-       } while (abs_i64(error.value) > 100ULL);
-
-       return res;
-}
-
-
-/* this function is a generic helper to translate fixed point value to
- * specified integer format that will consist of integer_bits integer part and
- * fractional_bits fractional part. For example it is used in
- * spl_fixpt_u2d19 to receive 2 bits integer part and 19 bits fractional
- * part in 32 bits. It is used in hw programming (scaler)
- */
-
-static inline unsigned int ux_dy(
-       long long value,
-       unsigned int integer_bits,
-       unsigned int fractional_bits)
-{
-       /* 1. create mask of integer part */
-       unsigned int result = (1 << integer_bits) - 1;
-       /* 2. mask out fractional part */
-       unsigned int fractional_part = FRACTIONAL_PART_MASK & value;
-       /* 3. shrink fixed point integer part to be of integer_bits width*/
-       result &= GET_INTEGER_PART(value);
-       /* 4. make space for fractional part to be filled in after integer */
-       result <<= fractional_bits;
-       /* 5. shrink fixed point fractional part to of fractional_bits width*/
-       fractional_part >>= FIXED31_32_BITS_PER_FRACTIONAL_PART - fractional_bits;
-       /* 6. merge the result */
-       return result | fractional_part;
-}
-
-static inline unsigned int clamp_ux_dy(
-       long long value,
-       unsigned int integer_bits,
-       unsigned int fractional_bits,
-       unsigned int min_clamp)
-{
-       unsigned int truncated_val = ux_dy(value, integer_bits, fractional_bits);
-
-       if (value >= (1LL << (integer_bits + FIXED31_32_BITS_PER_FRACTIONAL_PART)))
-               return (1 << (integer_bits + fractional_bits)) - 1;
-       else if (truncated_val > min_clamp)
-               return truncated_val;
-       else
-               return min_clamp;
-}
-
-unsigned int spl_fixpt_u4d19(struct spl_fixed31_32 arg)
-{
-       return ux_dy(arg.value, 4, 19);
-}
-
-unsigned int spl_fixpt_u3d19(struct spl_fixed31_32 arg)
-{
-       return ux_dy(arg.value, 3, 19);
-}
-
-unsigned int spl_fixpt_u2d19(struct spl_fixed31_32 arg)
-{
-       return ux_dy(arg.value, 2, 19);
-}
-
-unsigned int spl_fixpt_u0d19(struct spl_fixed31_32 arg)
-{
-       return ux_dy(arg.value, 0, 19);
-}
-
-unsigned int spl_fixpt_clamp_u0d14(struct spl_fixed31_32 arg)
-{
-       return clamp_ux_dy(arg.value, 0, 14, 1);
-}
-
-unsigned int spl_fixpt_clamp_u0d10(struct spl_fixed31_32 arg)
-{
-       return clamp_ux_dy(arg.value, 0, 10, 1);
-}
-
-int spl_fixpt_s4d19(struct spl_fixed31_32 arg)
-{
-       if (arg.value < 0)
-               return -(int)ux_dy(spl_fixpt_abs(arg).value, 4, 19);
-       else
-               return ux_dy(arg.value, 4, 19);
-}
-
-struct spl_fixed31_32 spl_fixpt_from_ux_dy(unsigned int value,
-       unsigned int integer_bits,
-       unsigned int fractional_bits)
-{
-       struct spl_fixed31_32 fixpt_value = spl_fixpt_zero;
-       struct spl_fixed31_32 fixpt_int_value = spl_fixpt_zero;
-       long long frac_mask = ((long long)1 << (long long)integer_bits) - 1;
-
-       fixpt_value.value = (long long)value << (FIXED31_32_BITS_PER_FRACTIONAL_PART - fractional_bits);
-       frac_mask = frac_mask << fractional_bits;
-       fixpt_int_value.value = value & frac_mask;
-       fixpt_int_value.value <<= (FIXED31_32_BITS_PER_FRACTIONAL_PART - fractional_bits);
-       fixpt_value.value |= fixpt_int_value.value;
-       return fixpt_value;
-}
-
-struct spl_fixed31_32 spl_fixpt_from_int_dy(unsigned int int_value,
-       unsigned int frac_value,
-       unsigned int integer_bits,
-       unsigned int fractional_bits)
-{
-       struct spl_fixed31_32 fixpt_value = spl_fixpt_from_int(int_value);
-
-       fixpt_value.value |= (long long)frac_value << (FIXED31_32_BITS_PER_FRACTIONAL_PART - fractional_bits);
-       return fixpt_value;
-}
diff --git a/drivers/gpu/drm/amd/display/dc/spl/spl_fixpt31_32.h b/drivers/gpu/drm/amd/display/dc/spl/spl_fixpt31_32.h
deleted file mode 100644 (file)
index 27ec6d4..0000000
+++ /dev/null
@@ -1,546 +0,0 @@
-/*
- * Copyright 2012-15 Advanced Micro Devices, Inc.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
- * OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors: AMD
- *
- */
-
-#ifndef __SPL_FIXED31_32_H__
-#define __SPL_FIXED31_32_H__
-
-#include "os_types.h"
-#include "spl_os_types.h"   // swap
-#ifndef ASSERT
-#define ASSERT(_bool) ((void *)0)
-#endif
-
-#ifndef LLONG_MAX
-#define LLONG_MAX 9223372036854775807ll
-#endif
-#ifndef LLONG_MIN
-#define LLONG_MIN (-LLONG_MAX - 1ll)
-#endif
-
-#define FIXED31_32_BITS_PER_FRACTIONAL_PART 32
-#ifndef LLONG_MIN
-#define LLONG_MIN (1LL<<63)
-#endif
-#ifndef LLONG_MAX
-#define LLONG_MAX (-1LL>>1)
-#endif
-
-/*
- * @brief
- * Arithmetic operations on real numbers
- * represented as fixed-point numbers.
- * There are: 1 bit for sign,
- * 31 bit for integer part,
- * 32 bits for fractional part.
- *
- * @note
- * Currently, overflows and underflows are asserted;
- * no special result returned.
- */
-
-struct spl_fixed31_32 {
-       long long value;
-};
-
-
-/*
- * @brief
- * Useful constants
- */
-
-static const struct spl_fixed31_32 spl_fixpt_zero = { 0 };
-static const struct spl_fixed31_32 spl_fixpt_epsilon = { 1LL };
-static const struct spl_fixed31_32 spl_fixpt_half = { 0x80000000LL };
-static const struct spl_fixed31_32 spl_fixpt_one = { 0x100000000LL };
-
-/*
- * @brief
- * Initialization routines
- */
-
-/*
- * @brief
- * result = numerator / denominator
- */
-struct spl_fixed31_32 spl_fixpt_from_fraction(long long numerator, long long denominator);
-
-/*
- * @brief
- * result = arg
- */
-static inline struct spl_fixed31_32 spl_fixpt_from_int(int arg)
-{
-       struct spl_fixed31_32 res;
-
-       res.value = (long long) arg << FIXED31_32_BITS_PER_FRACTIONAL_PART;
-
-       return res;
-}
-
-/*
- * @brief
- * Unary operators
- */
-
-/*
- * @brief
- * result = -arg
- */
-static inline struct spl_fixed31_32 spl_fixpt_neg(struct spl_fixed31_32 arg)
-{
-       struct spl_fixed31_32 res;
-
-       res.value = -arg.value;
-
-       return res;
-}
-
-/*
- * @brief
- * result = abs(arg) := (arg >= 0) ? arg : -arg
- */
-static inline struct spl_fixed31_32 spl_fixpt_abs(struct spl_fixed31_32 arg)
-{
-       if (arg.value < 0)
-               return spl_fixpt_neg(arg);
-       else
-               return arg;
-}
-
-/*
- * @brief
- * Binary relational operators
- */
-
-/*
- * @brief
- * result = arg1 < arg2
- */
-static inline bool spl_fixpt_lt(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       return arg1.value < arg2.value;
-}
-
-/*
- * @brief
- * result = arg1 <= arg2
- */
-static inline bool spl_fixpt_le(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       return arg1.value <= arg2.value;
-}
-
-/*
- * @brief
- * result = arg1 == arg2
- */
-static inline bool spl_fixpt_eq(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       return arg1.value == arg2.value;
-}
-
-/*
- * @brief
- * result = min(arg1, arg2) := (arg1 <= arg2) ? arg1 : arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_min(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       if (arg1.value <= arg2.value)
-               return arg1;
-       else
-               return arg2;
-}
-
-/*
- * @brief
- * result = max(arg1, arg2) := (arg1 <= arg2) ? arg2 : arg1
- */
-static inline struct spl_fixed31_32 spl_fixpt_max(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       if (arg1.value <= arg2.value)
-               return arg2;
-       else
-               return arg1;
-}
-
-/*
- * @brief
- *          | min_value, when arg <= min_value
- * result = | arg, when min_value < arg < max_value
- *          | max_value, when arg >= max_value
- */
-static inline struct spl_fixed31_32 spl_fixpt_clamp(
-       struct spl_fixed31_32 arg,
-       struct spl_fixed31_32 min_value,
-       struct spl_fixed31_32 max_value)
-{
-       if (spl_fixpt_le(arg, min_value))
-               return min_value;
-       else if (spl_fixpt_le(max_value, arg))
-               return max_value;
-       else
-               return arg;
-}
-
-/*
- * @brief
- * Binary shift operators
- */
-
-/*
- * @brief
- * result = arg << shift
- */
-static inline struct spl_fixed31_32 spl_fixpt_shl(struct spl_fixed31_32 arg, unsigned char shift)
-{
-       ASSERT(((arg.value >= 0) && (arg.value <= LLONG_MAX >> shift)) ||
-               ((arg.value < 0) && (arg.value >= ~(LLONG_MAX >> shift))));
-
-       arg.value = arg.value << shift;
-
-       return arg;
-}
-
-/*
- * @brief
- * result = arg >> shift
- */
-static inline struct spl_fixed31_32 spl_fixpt_shr(struct spl_fixed31_32 arg, unsigned char shift)
-{
-       bool negative = arg.value < 0;
-
-       if (negative)
-               arg.value = -arg.value;
-       arg.value = arg.value >> shift;
-       if (negative)
-               arg.value = -arg.value;
-       return arg;
-}
-
-/*
- * @brief
- * Binary additive operators
- */
-
-/*
- * @brief
- * result = arg1 + arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_add(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       struct spl_fixed31_32 res;
-
-       ASSERT(((arg1.value >= 0) && (LLONG_MAX - arg1.value >= arg2.value)) ||
-               ((arg1.value < 0) && (LLONG_MIN - arg1.value <= arg2.value)));
-
-       res.value = arg1.value + arg2.value;
-
-       return res;
-}
-
-/*
- * @brief
- * result = arg1 + arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_add_int(struct spl_fixed31_32 arg1, int arg2)
-{
-       return spl_fixpt_add(arg1, spl_fixpt_from_int(arg2));
-}
-
-/*
- * @brief
- * result = arg1 - arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_sub(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       struct spl_fixed31_32 res;
-
-       ASSERT(((arg2.value >= 0) && (LLONG_MIN + arg2.value <= arg1.value)) ||
-               ((arg2.value < 0) && (LLONG_MAX + arg2.value >= arg1.value)));
-
-       res.value = arg1.value - arg2.value;
-
-       return res;
-}
-
-/*
- * @brief
- * result = arg1 - arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_sub_int(struct spl_fixed31_32 arg1, int arg2)
-{
-       return spl_fixpt_sub(arg1, spl_fixpt_from_int(arg2));
-}
-
-
-/*
- * @brief
- * Binary multiplicative operators
- */
-
-/*
- * @brief
- * result = arg1 * arg2
- */
-struct spl_fixed31_32 spl_fixpt_mul(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2);
-
-
-/*
- * @brief
- * result = arg1 * arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_mul_int(struct spl_fixed31_32 arg1, int arg2)
-{
-       return spl_fixpt_mul(arg1, spl_fixpt_from_int(arg2));
-}
-
-/*
- * @brief
- * result = square(arg) := arg * arg
- */
-struct spl_fixed31_32 spl_fixpt_sqr(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * result = arg1 / arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_div_int(struct spl_fixed31_32 arg1, long long arg2)
-{
-       return spl_fixpt_from_fraction(arg1.value, spl_fixpt_from_int((int)arg2).value);
-}
-
-/*
- * @brief
- * result = arg1 / arg2
- */
-static inline struct spl_fixed31_32 spl_fixpt_div(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       return spl_fixpt_from_fraction(arg1.value, arg2.value);
-}
-
-/*
- * @brief
- * Reciprocal function
- */
-
-/*
- * @brief
- * result = reciprocal(arg) := 1 / arg
- *
- * @note
- * No special actions taken in case argument is zero.
- */
-struct spl_fixed31_32 spl_fixpt_recip(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * Trigonometric functions
- */
-
-/*
- * @brief
- * result = sinc(arg) := sin(arg) / arg
- *
- * @note
- * Argument specified in radians,
- * internally it's normalized to [-2pi...2pi] range.
- */
-struct spl_fixed31_32 spl_fixpt_sinc(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * result = sin(arg)
- *
- * @note
- * Argument specified in radians,
- * internally it's normalized to [-2pi...2pi] range.
- */
-struct spl_fixed31_32 spl_fixpt_sin(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * result = cos(arg)
- *
- * @note
- * Argument specified in radians
- * and should be in [-2pi...2pi] range -
- * passing arguments outside that range
- * will cause incorrect result!
- */
-struct spl_fixed31_32 spl_fixpt_cos(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * Transcendent functions
- */
-
-/*
- * @brief
- * result = exp(arg)
- *
- * @note
- * Currently, function is verified for abs(arg) <= 1.
- */
-struct spl_fixed31_32 spl_fixpt_exp(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * result = log(arg)
- *
- * @note
- * Currently, abs(arg) should be less than 1.
- * No normalization is done.
- * Currently, no special actions taken
- * in case of invalid argument(s). Take care!
- */
-struct spl_fixed31_32 spl_fixpt_log(struct spl_fixed31_32 arg);
-
-/*
- * @brief
- * Power function
- */
-
-/*
- * @brief
- * result = pow(arg1, arg2)
- *
- * @note
- * Currently, abs(arg1) should be less than 1. Take care!
- */
-static inline struct spl_fixed31_32 spl_fixpt_pow(struct spl_fixed31_32 arg1, struct spl_fixed31_32 arg2)
-{
-       if (arg1.value == 0)
-               return arg2.value == 0 ? spl_fixpt_one : spl_fixpt_zero;
-
-       return spl_fixpt_exp(
-               spl_fixpt_mul(
-                       spl_fixpt_log(arg1),
-                       arg2));
-}
-
-/*
- * @brief
- * Rounding functions
- */
-
-/*
- * @brief
- * result = floor(arg) := greatest integer lower than or equal to arg
- */
-static inline int spl_fixpt_floor(struct spl_fixed31_32 arg)
-{
-       unsigned long long arg_value = arg.value > 0 ? arg.value : -arg.value;
-
-       if (arg.value >= 0)
-               return (int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART);
-       else
-               return -(int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART);
-}
-
-/*
- * @brief
- * result = round(arg) := integer nearest to arg
- */
-static inline int spl_fixpt_round(struct spl_fixed31_32 arg)
-{
-       unsigned long long arg_value = arg.value > 0 ? arg.value : -arg.value;
-
-       const long long summand = spl_fixpt_half.value;
-
-       ASSERT(LLONG_MAX - (long long)arg_value >= summand);
-
-       arg_value += summand;
-
-       if (arg.value >= 0)
-               return (int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART);
-       else
-               return -(int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART);
-}
-
-/*
- * @brief
- * result = ceil(arg) := lowest integer greater than or equal to arg
- */
-static inline int spl_fixpt_ceil(struct spl_fixed31_32 arg)
-{
-       unsigned long long arg_value = arg.value > 0 ? arg.value : -arg.value;
-
-       const long long summand = spl_fixpt_one.value -
-               spl_fixpt_epsilon.value;
-
-       ASSERT(LLONG_MAX - (long long)arg_value >= summand);
-
-       arg_value += summand;
-
-       if (arg.value >= 0)
-               return (int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART);
-       else
-               return -(int)(arg_value >> FIXED31_32_BITS_PER_FRACTIONAL_PART);
-}
-
-/* the following two function are used in scaler hw programming to convert fixed
- * point value to format 2 bits from integer part and 19 bits from fractional
- * part. The same applies for u0d19, 0 bits from integer part and 19 bits from
- * fractional
- */
-
-unsigned int spl_fixpt_u4d19(struct spl_fixed31_32 arg);
-
-unsigned int spl_fixpt_u3d19(struct spl_fixed31_32 arg);
-
-unsigned int spl_fixpt_u2d19(struct spl_fixed31_32 arg);
-
-unsigned int spl_fixpt_u0d19(struct spl_fixed31_32 arg);
-
-unsigned int spl_fixpt_clamp_u0d14(struct spl_fixed31_32 arg);
-
-unsigned int spl_fixpt_clamp_u0d10(struct spl_fixed31_32 arg);
-
-int spl_fixpt_s4d19(struct spl_fixed31_32 arg);
-
-static inline struct spl_fixed31_32 spl_fixpt_truncate(struct spl_fixed31_32 arg, unsigned int frac_bits)
-{
-       bool negative = arg.value < 0;
-
-       if (frac_bits >= FIXED31_32_BITS_PER_FRACTIONAL_PART) {
-               ASSERT(frac_bits == FIXED31_32_BITS_PER_FRACTIONAL_PART);
-               return arg;
-       }
-
-       if (negative)
-               arg.value = -arg.value;
-       arg.value &= (~0ULL) << (FIXED31_32_BITS_PER_FRACTIONAL_PART - frac_bits);
-       if (negative)
-               arg.value = -arg.value;
-       return arg;
-}
-
-struct spl_fixed31_32 spl_fixpt_from_ux_dy(unsigned int value, unsigned int integer_bits, unsigned int fractional_bits);
-struct spl_fixed31_32 spl_fixpt_from_int_dy(unsigned int int_value,
-               unsigned int frac_value,
-               unsigned int integer_bits,
-               unsigned int fractional_bits);
-
-#endif