[linux-2.6-block.git] / drivers / gpu / drm / amd / amdgpu / amdgpu_benchmark.c

/*
 * Copyright 2009 Jerome Glisse.
 *
 * 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: Jerome Glisse
 */

#include <drm/amdgpu_drm.h>
#include "amdgpu.h"

#define AMDGPU_BENCHMARK_ITERATIONS 1024
#define AMDGPU_BENCHMARK_COMMON_MODES_N 17

static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
				    uint64_t saddr, uint64_t daddr, int n)
{
	unsigned long start_jiffies;
	unsigned long end_jiffies;
	struct dma_fence *fence = NULL;
	int i, r;

	start_jiffies = jiffies;
	for (i = 0; i < n; i++) {
		struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
		r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence,
				       false, false);
		if (r)
			goto exit_do_move;
		r = dma_fence_wait(fence, false);
		if (r)
			goto exit_do_move;
		dma_fence_put(fence);
	}
	end_jiffies = jiffies;
	r = jiffies_to_msecs(end_jiffies - start_jiffies);

exit_do_move:
	if (fence)
		dma_fence_put(fence);
	return r;
}


static void amdgpu_benchmark_log_results(int n, unsigned size,
					 unsigned int time,
					 unsigned sdomain, unsigned ddomain,
					 char *kind)
{
	unsigned int throughput = (n * (size >> 10)) / time;
	DRM_INFO("amdgpu: %s %u bo moves of %u kB from"
		 " %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n",
		 kind, n, size >> 10, sdomain, ddomain, time,
		 throughput * 8, throughput);
}

static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
				  unsigned sdomain, unsigned ddomain)
{
	struct amdgpu_bo *dobj = NULL;
	struct amdgpu_bo *sobj = NULL;
	struct amdgpu_bo_param bp;
	uint64_t saddr, daddr;
	int r, n;
	int time;

	memset(&bp, 0, sizeof(bp));
	bp.size = size;
	bp.byte_align = PAGE_SIZE;
	bp.domain = sdomain;
	bp.flags = 0;
	bp.type = ttm_bo_type_kernel;
	bp.resv = NULL;
	n = AMDGPU_BENCHMARK_ITERATIONS;
	r = amdgpu_bo_create(adev, &bp, &sobj);
	if (r) {
		goto out_cleanup;
	}
	r = amdgpu_bo_reserve(sobj, false);
	if (unlikely(r != 0))
		goto out_cleanup;
	r = amdgpu_bo_pin(sobj, sdomain);
	if (r) {
		amdgpu_bo_unreserve(sobj);
		goto out_cleanup;
	}
	r = amdgpu_ttm_alloc_gart(&sobj->tbo);
	amdgpu_bo_unreserve(sobj);
	if (r) {
		goto out_cleanup;
	}
	saddr = amdgpu_bo_gpu_offset(sobj);
	bp.domain = ddomain;
	r = amdgpu_bo_create(adev, &bp, &dobj);
	if (r) {
		goto out_cleanup;
	}
	r = amdgpu_bo_reserve(dobj, false);
	if (unlikely(r != 0))
		goto out_cleanup;
	r = amdgpu_bo_pin(dobj, ddomain);
	if (r) {
		amdgpu_bo_unreserve(sobj);
		goto out_cleanup;
	}
	r = amdgpu_ttm_alloc_gart(&dobj->tbo);
	amdgpu_bo_unreserve(dobj);
	if (r) {
		goto out_cleanup;
	}
	daddr = amdgpu_bo_gpu_offset(dobj);

	if (adev->mman.buffer_funcs) {
		time = amdgpu_benchmark_do_move(adev, size, saddr, daddr, n);
		if (time < 0)
			goto out_cleanup;
		if (time > 0)
			amdgpu_benchmark_log_results(n, size, time,
						     sdomain, ddomain, "dma");
	}

out_cleanup:
	/* Check error value now. The value can be overwritten when clean up.*/
	if (r) {
		DRM_ERROR("Error while benchmarking BO move.\n");
	}

	if (sobj) {
		r = amdgpu_bo_reserve(sobj, true);
		if (likely(r == 0)) {
			amdgpu_bo_unpin(sobj);
			amdgpu_bo_unreserve(sobj);
		}
		amdgpu_bo_unref(&sobj);
	}
	if (dobj) {
		r = amdgpu_bo_reserve(dobj, true);
		if (likely(r == 0)) {
			amdgpu_bo_unpin(dobj);
			amdgpu_bo_unreserve(dobj);
		}
		amdgpu_bo_unref(&dobj);
	}
}

void amdgpu_benchmark(struct amdgpu_device *adev, int test_number)
{
	int i;
	static const int common_modes[AMDGPU_BENCHMARK_COMMON_MODES_N] = {
		640 * 480 * 4,
		720 * 480 * 4,
		800 * 600 * 4,
		848 * 480 * 4,
		1024 * 768 * 4,
		1152 * 768 * 4,
		1280 * 720 * 4,
		1280 * 800 * 4,
		1280 * 854 * 4,
		1280 * 960 * 4,
		1280 * 1024 * 4,
		1440 * 900 * 4,
		1400 * 1050 * 4,
		1680 * 1050 * 4,
		1600 * 1200 * 4,
		1920 * 1080 * 4,
		1920 * 1200 * 4
	};

	switch (test_number) {
	case 1:
		/* simple test, VRAM to GTT and GTT to VRAM */
		amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_GTT,
				      AMDGPU_GEM_DOMAIN_VRAM);
		amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
				      AMDGPU_GEM_DOMAIN_GTT);
		break;
	case 2:
		/* simple test, VRAM to VRAM */
		amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
				      AMDGPU_GEM_DOMAIN_VRAM);
		break;
	case 3:
		/* GTT to VRAM, buffer size sweep, powers of 2 */
		for (i = 1; i <= 16384; i <<= 1)
			amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
					      AMDGPU_GEM_DOMAIN_GTT,
					      AMDGPU_GEM_DOMAIN_VRAM);
		break;
	case 4:
		/* VRAM to GTT, buffer size sweep, powers of 2 */
		for (i = 1; i <= 16384; i <<= 1)
			amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
					      AMDGPU_GEM_DOMAIN_VRAM,
					      AMDGPU_GEM_DOMAIN_GTT);
		break;
	case 5:
		/* VRAM to VRAM, buffer size sweep, powers of 2 */
		for (i = 1; i <= 16384; i <<= 1)
			amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
					      AMDGPU_GEM_DOMAIN_VRAM,
					      AMDGPU_GEM_DOMAIN_VRAM);
		break;
	case 6:
		/* GTT to VRAM, buffer size sweep, common modes */
		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
			amdgpu_benchmark_move(adev, common_modes[i],
					      AMDGPU_GEM_DOMAIN_GTT,
					      AMDGPU_GEM_DOMAIN_VRAM);
		break;
	case 7:
		/* VRAM to GTT, buffer size sweep, common modes */
		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
			amdgpu_benchmark_move(adev, common_modes[i],
					      AMDGPU_GEM_DOMAIN_VRAM,
					      AMDGPU_GEM_DOMAIN_GTT);
		break;
	case 8:
		/* VRAM to VRAM, buffer size sweep, common modes */
		for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
			amdgpu_benchmark_move(adev, common_modes[i],
					      AMDGPU_GEM_DOMAIN_VRAM,
					      AMDGPU_GEM_DOMAIN_VRAM);
		break;

	default:
		DRM_ERROR("Unknown benchmark\n");
	}
}
Commit	Line	Data
d38ceaf9 AD	1	/*
	2	* Copyright 2009 Jerome Glisse.
	3	*
	4	* Permission is hereby granted, free of charge, to any person obtaining a
	5	* copy of this software and associated documentation files (the "Software"),
	6	* to deal in the Software without restriction, including without limitation
	7	* the rights to use, copy, modify, merge, publish, distribute, sublicense,
	8	* and/or sell copies of the Software, and to permit persons to whom the
	9	* Software is furnished to do so, subject to the following conditions:
	10	*
	11	* The above copyright notice and this permission notice shall be included in
	12	* all copies or substantial portions of the Software.
	13	*
	14	* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
	15	* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
	16	* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
	17	* THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
	18	* OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
	19	* ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
	20	* OTHER DEALINGS IN THE SOFTWARE.
	21	*
	22	* Authors: Jerome Glisse
	23	*/
fdf2f6c5	24
d38ceaf9 AD	25	#include <drm/amdgpu_drm.h>
	26	#include "amdgpu.h"
	27
	28	#define AMDGPU_BENCHMARK_ITERATIONS 1024
	29	#define AMDGPU_BENCHMARK_COMMON_MODES_N 17
	30
	31	static int amdgpu_benchmark_do_move(struct amdgpu_device *adev, unsigned size,
	32	uint64_t saddr, uint64_t daddr, int n)
	33	{
	34	unsigned long start_jiffies;
	35	unsigned long end_jiffies;
f54d1867	36	struct dma_fence *fence = NULL;
d38ceaf9 AD	37	int i, r;
	38
	39	start_jiffies = jiffies;
	40	for (i = 0; i < n; i++) {
	41	struct amdgpu_ring *ring = adev->mman.buffer_funcs_ring;
e24db985	42	r = amdgpu_copy_buffer(ring, saddr, daddr, size, NULL, &fence,
fc9c8f54	43	false, false);
d38ceaf9 AD	44	if (r)
d38ceaf9 AD	45	goto exit_do_move;
f54d1867	46	r = dma_fence_wait(fence, false);
d38ceaf9 AD	47	if (r)
d38ceaf9 AD	48	goto exit_do_move;
f54d1867	49	dma_fence_put(fence);
d38ceaf9 AD	50	}
	51	end_jiffies = jiffies;
	52	r = jiffies_to_msecs(end_jiffies - start_jiffies);
	53
	54	exit_do_move:
	55	if (fence)
f54d1867	56	dma_fence_put(fence);
d38ceaf9 AD	57	return r;
	58	}
	59
	60
	61	static void amdgpu_benchmark_log_results(int n, unsigned size,
	62	unsigned int time,
	63	unsigned sdomain, unsigned ddomain,
	64	char *kind)
	65	{
	66	unsigned int throughput = (n * (size >> 10)) / time;
	67	DRM_INFO("amdgpu: %s %u bo moves of %u kB from"
	68	" %d to %d in %u ms, throughput: %u Mb/s or %u MB/s\n",
	69	kind, n, size >> 10, sdomain, ddomain, time,
	70	throughput * 8, throughput);
	71	}
	72
	73	static void amdgpu_benchmark_move(struct amdgpu_device *adev, unsigned size,
	74	unsigned sdomain, unsigned ddomain)
	75	{
	76	struct amdgpu_bo *dobj = NULL;
	77	struct amdgpu_bo *sobj = NULL;
3216c6b7	78	struct amdgpu_bo_param bp;
d38ceaf9 AD	79	uint64_t saddr, daddr;
	80	int r, n;
	81	int time;
	82
3216c6b7 CZ	83	memset(&bp, 0, sizeof(bp));
	84	bp.size = size;
	85	bp.byte_align = PAGE_SIZE;
	86	bp.domain = sdomain;
	87	bp.flags = 0;
	88	bp.type = ttm_bo_type_kernel;
	89	bp.resv = NULL;
d38ceaf9	90	n = AMDGPU_BENCHMARK_ITERATIONS;
3216c6b7	91	r = amdgpu_bo_create(adev, &bp, &sobj);
d38ceaf9 AD	92	if (r) {
	93	goto out_cleanup;
	94	}
	95	r = amdgpu_bo_reserve(sobj, false);
	96	if (unlikely(r != 0))
	97	goto out_cleanup;
7b7c6c81	98	r = amdgpu_bo_pin(sobj, sdomain);
bb812f1e JZ	99	if (r) {
	100	amdgpu_bo_unreserve(sobj);
	101	goto out_cleanup;
	102	}
	103	r = amdgpu_ttm_alloc_gart(&sobj->tbo);
d38ceaf9 AD	104	amdgpu_bo_unreserve(sobj);
	105	if (r) {
	106	goto out_cleanup;
	107	}
bb812f1e	108	saddr = amdgpu_bo_gpu_offset(sobj);
3216c6b7 CZ	109	bp.domain = ddomain;
3216c6b7 CZ	110	r = amdgpu_bo_create(adev, &bp, &dobj);
d38ceaf9 AD	111	if (r) {
	112	goto out_cleanup;
	113	}
	114	r = amdgpu_bo_reserve(dobj, false);
	115	if (unlikely(r != 0))
	116	goto out_cleanup;
7b7c6c81	117	r = amdgpu_bo_pin(dobj, ddomain);
bb812f1e JZ	118	if (r) {
	119	amdgpu_bo_unreserve(sobj);
	120	goto out_cleanup;
	121	}
	122	r = amdgpu_ttm_alloc_gart(&dobj->tbo);
d38ceaf9 AD	123	amdgpu_bo_unreserve(dobj);
	124	if (r) {
	125	goto out_cleanup;
	126	}
bb812f1e	127	daddr = amdgpu_bo_gpu_offset(dobj);
d38ceaf9 AD	128
	129	if (adev->mman.buffer_funcs) {
	130	time = amdgpu_benchmark_do_move(adev, size, saddr, daddr, n);
	131	if (time < 0)
	132	goto out_cleanup;
	133	if (time > 0)
	134	amdgpu_benchmark_log_results(n, size, time,
	135	sdomain, ddomain, "dma");
	136	}
	137
	138	out_cleanup:
d159f26c AX	139	/* Check error value now. The value can be overwritten when clean up.*/
	140	if (r) {
	141	DRM_ERROR("Error while benchmarking BO move.\n");
	142	}
	143
d38ceaf9	144	if (sobj) {
4a9ed100	145	r = amdgpu_bo_reserve(sobj, true);
d38ceaf9 AD	146	if (likely(r == 0)) {
	147	amdgpu_bo_unpin(sobj);
	148	amdgpu_bo_unreserve(sobj);
	149	}
	150	amdgpu_bo_unref(&sobj);
	151	}
	152	if (dobj) {
4a9ed100	153	r = amdgpu_bo_reserve(dobj, true);
d38ceaf9 AD	154	if (likely(r == 0)) {
	155	amdgpu_bo_unpin(dobj);
	156	amdgpu_bo_unreserve(dobj);
	157	}
	158	amdgpu_bo_unref(&dobj);
	159	}
d38ceaf9 AD	160	}
	161
	162	void amdgpu_benchmark(struct amdgpu_device *adev, int test_number)
	163	{
	164	int i;
aeba709a	165	static const int common_modes[AMDGPU_BENCHMARK_COMMON_MODES_N] = {
d38ceaf9 AD	166	640 * 480 * 4,
	167	720 * 480 * 4,
	168	800 * 600 * 4,
	169	848 * 480 * 4,
	170	1024 * 768 * 4,
	171	1152 * 768 * 4,
	172	1280 * 720 * 4,
	173	1280 * 800 * 4,
	174	1280 * 854 * 4,
	175	1280 * 960 * 4,
	176	1280 * 1024 * 4,
	177	1440 * 900 * 4,
	178	1400 * 1050 * 4,
	179	1680 * 1050 * 4,
	180	1600 * 1200 * 4,
	181	1920 * 1080 * 4,
	182	1920 * 1200 * 4
	183	};
	184
	185	switch (test_number) {
	186	case 1:
	187	/* simple test, VRAM to GTT and GTT to VRAM */
	188	amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_GTT,
	189	AMDGPU_GEM_DOMAIN_VRAM);
	190	amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
	191	AMDGPU_GEM_DOMAIN_GTT);
	192	break;
	193	case 2:
	194	/* simple test, VRAM to VRAM */
	195	amdgpu_benchmark_move(adev, 1024*1024, AMDGPU_GEM_DOMAIN_VRAM,
	196	AMDGPU_GEM_DOMAIN_VRAM);
	197	break;
	198	case 3:
	199	/* GTT to VRAM, buffer size sweep, powers of 2 */
	200	for (i = 1; i <= 16384; i <<= 1)
	201	amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
	202	AMDGPU_GEM_DOMAIN_GTT,
	203	AMDGPU_GEM_DOMAIN_VRAM);
	204	break;
	205	case 4:
	206	/* VRAM to GTT, buffer size sweep, powers of 2 */
	207	for (i = 1; i <= 16384; i <<= 1)
	208	amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
	209	AMDGPU_GEM_DOMAIN_VRAM,
	210	AMDGPU_GEM_DOMAIN_GTT);
	211	break;
	212	case 5:
	213	/* VRAM to VRAM, buffer size sweep, powers of 2 */
	214	for (i = 1; i <= 16384; i <<= 1)
	215	amdgpu_benchmark_move(adev, i * AMDGPU_GPU_PAGE_SIZE,
	216	AMDGPU_GEM_DOMAIN_VRAM,
	217	AMDGPU_GEM_DOMAIN_VRAM);
	218	break;
	219	case 6:
	220	/* GTT to VRAM, buffer size sweep, common modes */
	221	for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
	222	amdgpu_benchmark_move(adev, common_modes[i],
	223	AMDGPU_GEM_DOMAIN_GTT,
	224	AMDGPU_GEM_DOMAIN_VRAM);
	225	break;
	226	case 7:
	227	/* VRAM to GTT, buffer size sweep, common modes */
	228	for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
	229	amdgpu_benchmark_move(adev, common_modes[i],
230	AMDGPU_GEM_DOMAIN_VRAM,
231	AMDGPU_GEM_DOMAIN_GTT);
232	break;
233	case 8:
234	/* VRAM to VRAM, buffer size sweep, common modes */
235	for (i = 0; i < AMDGPU_BENCHMARK_COMMON_MODES_N; i++)
236	amdgpu_benchmark_move(adev, common_modes[i],
237	AMDGPU_GEM_DOMAIN_VRAM,
238	AMDGPU_GEM_DOMAIN_VRAM);
239	break;
240
241	default:
242	DRM_ERROR("Unknown benchmark\n");
243	}
244	}