radeonsi: display cs blit count for AMD_DEBUG=testdma

[mesa.git] / src / gallium / drivers / radeonsi / si_test_dma.c

diff --git a/src/gallium/drivers/radeonsi/si_test_dma.c b/src/gallium/drivers/radeonsi/si_test_dma.c
index 6c3cde49727b3015ca8e07014012e90ea07983b0..b3ab2750f25a313cfa3f8f1fe83d0699c77233f2 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_test_dma.c
+++ b/src/gallium/drivers/radeonsi/si_test_dma.c
@@ -192,11 +192,10 @@ void si_test_dma(struct si_screen *sscreen)
        struct pipe_context *ctx = screen->context_create(screen, NULL, 0);
        struct si_context *sctx = (struct si_context*)ctx;
        uint64_t max_alloc_size;
-       unsigned i, iterations, num_partial_copies, max_levels, max_tex_side;
+       unsigned i, iterations, num_partial_copies, max_tex_side;
        unsigned num_pass = 0, num_fail = 0;
 
-       max_levels = screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_LEVELS);
-       max_tex_side = 1 << (max_levels - 1);
+       max_tex_side = screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_2D_SIZE);
 
        /* Max 128 MB allowed for both textures. */
        max_alloc_size = 128 * 1024 * 1024;
@@ -220,11 +219,11 @@ void si_test_dma(struct si_screen *sscreen)
         */
        for (i = 0; i < iterations; i++) {
                struct pipe_resource tsrc = {}, tdst = {}, *src, *dst;
-               struct r600_texture *rdst;
-               struct r600_texture *rsrc;
+               struct si_texture *sdst;
+               struct si_texture *ssrc;
                struct cpu_texture src_cpu, dst_cpu;
                unsigned bpp, max_width, max_height, max_depth, j, num;
-               unsigned gfx_blits = 0, dma_blits = 0, max_tex_side_gen;
+               unsigned gfx_blits = 0, dma_blits = 0, cs_blits = 0, max_tex_side_gen;
                unsigned max_tex_layers;
                bool pass;
                bool do_partial_copies = rand() & 1;
@@ -290,24 +289,26 @@ void si_test_dma(struct si_screen *sscreen)
                dst = screen->resource_create(screen, &tdst);
                assert(src);
                assert(dst);
-               rdst = (struct r600_texture*)dst;
-               rsrc = (struct r600_texture*)src;
+               sdst = (struct si_texture*)dst;
+               ssrc = (struct si_texture*)src;
                alloc_cpu_texture(&src_cpu, &tsrc, bpp);
                alloc_cpu_texture(&dst_cpu, &tdst, bpp);
 
                printf("%4u: dst = (%5u x %5u x %u, %s), "
                       " src = (%5u x %5u x %u, %s), bpp = %2u, ",
                       i, tdst.width0, tdst.height0, tdst.array_size,
-                      array_mode_to_string(sscreen, &rdst->surface),
+                      array_mode_to_string(sscreen, &sdst->surface),
                       tsrc.width0, tsrc.height0, tsrc.array_size,
-                      array_mode_to_string(sscreen, &rsrc->surface), bpp);
+                      array_mode_to_string(sscreen, &ssrc->surface), bpp);
                fflush(stdout);
 
                /* set src pixels */
                set_random_pixels(ctx, src, &src_cpu);
 
                /* clear dst pixels */
-               si_clear_buffer(sctx, dst, 0, rdst->surface.surf_size, 0, true);
+               uint32_t zero = 0;
+               si_clear_buffer(sctx, dst, 0, sdst->surface.surf_size, &zero, 4,
+                               SI_COHERENCY_SHADER, false);
                memset(dst_cpu.ptr, 0, dst_cpu.layer_stride * tdst.array_size);
 
                /* preparation */
@@ -322,6 +323,7 @@ void si_test_dma(struct si_screen *sscreen)
                        struct pipe_box box;
                        unsigned old_num_draw_calls = sctx->num_draw_calls;
                        unsigned old_num_dma_calls = sctx->num_dma_calls;
+                       unsigned old_num_cs_calls = sctx->num_compute_calls;
 
                        if (!do_partial_copies) {
                                /* copy whole src to dst */
@@ -337,8 +339,8 @@ void si_test_dma(struct si_screen *sscreen)
                                dstz = rand() % (tdst.array_size - depth + 1);
 
                                /* special code path to hit the tiled partial copies */
-                               if (!rsrc->surface.is_linear &&
-                                   !rdst->surface.is_linear &&
+                               if (!ssrc->surface.is_linear &&
+                                   !sdst->surface.is_linear &&
                                    rand() & 1) {
                                        if (max_width < 8 || max_height < 8)
                                                continue;
@@ -365,8 +367,8 @@ void si_test_dma(struct si_screen *sscreen)
                                }
 
                                /* special code path to hit out-of-bounds reads in L2T */
-                               if (rsrc->surface.is_linear &&
-                                   !rdst->surface.is_linear &&
+                               if (ssrc->surface.is_linear &&
+                                   !sdst->surface.is_linear &&
                                    rand() % 4 == 0) {
                                        srcx = 0;
                                        srcy = 0;
@@ -381,6 +383,7 @@ void si_test_dma(struct si_screen *sscreen)
                        /* See which engine was used. */
                        gfx_blits += sctx->num_draw_calls > old_num_draw_calls;
                        dma_blits += sctx->num_dma_calls > old_num_dma_calls;
+                       cs_blits  += sctx->num_compute_calls > old_num_cs_calls;
 
                        /* CPU copy */
                        util_copy_box(dst_cpu.ptr, tdst.format, dst_cpu.stride,
@@ -397,8 +400,8 @@ void si_test_dma(struct si_screen *sscreen)
                else
                        num_fail++;
 
-               printf("BLITs: GFX = %2u, DMA = %2u, %s [%u/%u]\n",
-                      gfx_blits, dma_blits, pass ? "pass" : "fail",
+               printf("BLITs: GFX = %2u, DMA = %2u, CS = %2u, %s [%u/%u]\n",
+                      gfx_blits, dma_blits, cs_blits, pass ? "pass" : "fail",
                       num_pass, num_pass+num_fail);
 
                /* cleanup */