2 * Copyright © 2009 Intel Corporation
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:
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
24 * Eric Anholt <eric@anholt.net>
29 * Roughly simulates repeatedly uploading frames of images, by uploading
30 * the data all at once with pwrite, and then blitting it to another buffer.
32 * You might think of this like a movie player, but that wouldn't be entirely
33 * accurate, since the access patterns of the memory would be different
34 * (generally, smaller source image, upscaled, an thus different memory access
35 * pattern in both texel fetch for the stretching and the destination writes).
36 * However, some things like swfdec would be doing something like this since
37 * they compute their data in host memory and upload the full sw rendered
40 * Additionally, those applications should be rendering at the screen refresh
41 * rate, while this test has no limits, and so can get itself into the
42 * working set larger than aperture size performance disaster.
44 * The current workload we have that does large drm_intel_bo_map()
45 * uploads is texture upload for OpenGL (as it frequently is doing
46 * reformatting as it uploads the user's data, making bo_subdata less
62 #include "intel_bufmgr.h"
63 #include "intel_batchbuffer.h"
65 #include "intel_chipset.h"
67 #define OBJECT_WIDTH 1280
68 #define OBJECT_HEIGHT 720
71 get_time_in_secs(void)
75 gettimeofday(&tv, NULL);
77 return (double)tv.tv_sec + tv.tv_usec / 1000000.0;
81 do_render(drm_intel_bufmgr *bufmgr, struct intel_batchbuffer *batch,
82 drm_intel_bo *dst_bo, int width, int height)
87 static uint32_t seed = 1;
89 src_bo = drm_intel_bo_alloc(bufmgr, "src", width * height * 4, 4096);
91 drm_intel_bo_map(src_bo, 1);
93 data = src_bo->virtual;
94 for (i = 0; i < width * height; i++) {
98 drm_intel_bo_unmap(src_bo);
100 /* Render the junk to the dst. */
101 BLIT_COPY_BATCH_START(batch->devid, 0);
102 OUT_BATCH((3 << 24) | /* 32 bits */
103 (0xcc << 16) | /* copy ROP */
104 (width * 4) /* dst pitch */);
105 OUT_BATCH(0); /* dst x1,y1 */
106 OUT_BATCH((height << 16) | width); /* dst x2,y2 */
107 OUT_RELOC(dst_bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, 0);
108 BLIT_RELOC_UDW(batch->devid);
109 OUT_BATCH(0); /* src x1,y1 */
110 OUT_BATCH(width * 4); /* src pitch */
111 OUT_RELOC(src_bo, I915_GEM_DOMAIN_RENDER, 0, 0);
112 BLIT_RELOC_UDW(batch->devid);
115 intel_batchbuffer_flush(batch);
117 drm_intel_bo_unreference(src_bo);
120 int main(int argc, char **argv)
123 int object_size = OBJECT_WIDTH * OBJECT_HEIGHT * 4;
124 double start_time, end_time;
125 drm_intel_bo *dst_bo;
126 drm_intel_bufmgr *bufmgr;
127 struct intel_batchbuffer *batch;
132 bufmgr = drm_intel_bufmgr_gem_init(fd, 4096);
133 drm_intel_bufmgr_gem_enable_reuse(bufmgr);
135 batch = intel_batchbuffer_alloc(bufmgr, intel_get_drm_devid(fd));
137 dst_bo = drm_intel_bo_alloc(bufmgr, "dst", object_size, 4096);
139 /* Prep loop to get us warmed up. */
140 for (i = 0; i < 60; i++) {
141 do_render(bufmgr, batch, dst_bo, OBJECT_WIDTH, OBJECT_HEIGHT);
143 drm_intel_bo_wait_rendering(dst_bo);
145 /* Do the actual timing. */
146 start_time = get_time_in_secs();
147 for (i = 0; i < 200; i++) {
148 do_render(bufmgr, batch, dst_bo, OBJECT_WIDTH, OBJECT_HEIGHT);
150 drm_intel_bo_wait_rendering(dst_bo);
151 end_time = get_time_in_secs();
153 printf("%d iterations in %.03f secs: %.01f MB/sec\n", i,
154 end_time - start_time,
155 (double)i * OBJECT_WIDTH * OBJECT_HEIGHT * 4 / 1024.0 / 1024.0 /
156 (end_time - start_time));
158 intel_batchbuffer_free(batch);
159 drm_intel_bufmgr_destroy(bufmgr);