Return-Path: Received: from mga07.intel.com (mga07.intel.com [134.134.136.100]) by gabe.freedesktop.org (Postfix) with ESMTPS id 502C510F124 for ; Fri, 8 Apr 2022 12:40:54 +0000 (UTC) Date: Fri, 8 Apr 2022 14:40:49 +0200 From: Zbigniew =?utf-8?Q?Kempczy=C5=84ski?= To: Ramalingam C Message-ID: References: <20220405131407.10089-1-ramalingam.c@intel.com> <20220405141050.16037-1-ramalingam.c@intel.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20220405141050.16037-1-ramalingam.c@intel.com> Subject: Re: [igt-dev] [PATCH i-g-t v8 2/2] tests/i915/gem_lmem_swapping: Add ccs subtests List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: igt-dev , petri.latvala@intel.com Errors-To: igt-dev-bounces@lists.freedesktop.org Sender: "igt-dev" List-ID: On Tue, Apr 05, 2022 at 07:40:50PM +0530, Ramalingam C wrote: > Add subtests for covering the compressed object's eviction. > > v2: > gem_sync after the block_copy blit for init > v3: > ahnd is passed in as a param [Zbigniew] > cmd is bb [Zbigniew] > blt src and dst sizes supposed to be same [Zbigniew] > v4: > Use gem_create_from_pool [Zbigniew] > v5: > Mark only blt_ctx as non-persistence [Ashutosh] > v6: > Move the intel_allocator_open/clsoe is moved under TEST-CCS [Arjun] > blt_ctx through intel_ctx_0 [Arjun] > v7: > Get the blt engine for init and verify of ccs obj [Zbigniew] > v8: > Newline for assert msg [Petri] > > Signed-off-by: Chris Wilson > Signed-off-by: Ayaz A Siddiqui > Signed-off-by: Ramalingam C > --- > tests/i915/gem_lmem_swapping.c | 226 ++++++++++++++++++++++++++++++++- > 1 file changed, 220 insertions(+), 6 deletions(-) > > diff --git a/tests/i915/gem_lmem_swapping.c b/tests/i915/gem_lmem_swapping.c > index d682c6f2571f..b71c3b200561 100644 > --- a/tests/i915/gem_lmem_swapping.c > +++ b/tests/i915/gem_lmem_swapping.c > @@ -22,6 +22,7 @@ > #include > #include > #include "drm.h" > +#include "i915/i915_blt.h" > > IGT_TEST_DESCRIPTION("Exercise local memory swapping."); > > @@ -30,6 +31,7 @@ IGT_TEST_DESCRIPTION("Exercise local memory swapping."); > > #define PAGE_SIZE (1ULL << 12) > #define SZ_64K (16 * PAGE_SIZE) > +#define DG2_MOCS (2 << 1) > > static const char *readable_unit(uint64_t size) > { > @@ -60,6 +62,7 @@ struct params { > #define TEST_RANDOM (1 << 3) > #define TEST_ENGINES (1 << 4) > #define TEST_MULTI (1 << 5) > +#define TEST_CCS (1 << 6) > unsigned int flags; > unsigned int seed; > bool oom_test; > @@ -69,8 +72,56 @@ struct object { > uint64_t size; > uint32_t seed; > uint32_t handle; > + struct blt_copy_object *blt_obj; > }; > > +static void set_object(struct blt_copy_object *obj, > + uint32_t handle, uint64_t size, uint32_t region, > + uint8_t mocs, enum blt_tiling tiling, > + enum blt_compression compression, > + enum blt_compression_type compression_type) > +{ > + obj->handle = handle; > + obj->size = size; > + obj->region = region; > + obj->mocs = mocs; > + obj->tiling = tiling; > + obj->compression = compression; > + obj->compression_type = compression_type; > +} > + > +static void set_geom(struct blt_copy_object *obj, uint32_t pitch, > + int16_t x1, int16_t y1, int16_t x2, int16_t y2, > + uint16_t x_offset, uint16_t y_offset) > +{ > + obj->pitch = pitch; > + obj->x1 = x1; > + obj->y1 = y1; > + obj->x2 = x2; > + obj->y2 = y2; > + obj->x_offset = x_offset; > + obj->y_offset = y_offset; > +} > + > +static void set_batch(struct blt_copy_batch *batch, > + uint32_t handle, uint64_t size, uint32_t region) > +{ > + batch->handle = handle; > + batch->size = size; > + batch->region = region; > +} > + > +static void set_object_ext(struct blt_block_copy_object_ext *obj, > + uint8_t compression_format, > + uint16_t surface_width, uint16_t surface_height, > + enum blt_surface_type surface_type) > +{ > + obj->compression_format = compression_format; > + obj->surface_width = surface_width; > + obj->surface_height = surface_height; > + obj->surface_type = surface_type; > +} > + > static uint32_t create_bo(int i915, > uint64_t *size, > struct drm_i915_gem_memory_class_instance *region, > @@ -105,6 +156,51 @@ init_object(int i915, struct object *obj, unsigned long seed, unsigned int flags > munmap(buf, obj->size); > } > > +static void > +init_object_ccs(int i915, struct object *obj, struct blt_copy_object *tmp, > + unsigned long seed, const intel_ctx_t *ctx, uint32_t region, > + uint64_t ahnd) > +{ > + struct blt_block_copy_data_ext ext = {}, *pext = &ext; > + const struct intel_execution_engine2 *e; > + struct blt_copy_data blt = {}; > + struct blt_copy_batch *cmd; > + uint64_t size = 4096; > + unsigned long *buf, j; > + > + obj->seed = seed; > + for_each_ctx_engine(i915, ctx, e) { > + if (gem_engine_can_block_copy(i915, e)) > + break; > + } > + igt_assert_f(e, "Ctx dont have blt engine\n"); s/dont/don't/ > + > + cmd = calloc(1, sizeof(*cmd)); > + cmd->handle = gem_create_from_pool(i915, &size, region); > + set_batch(cmd, cmd->handle, size, region); > + > + buf = gem_mmap__device_coherent(i915, tmp->handle, 0, obj->size, PROT_WRITE); > + gem_set_domain(i915, tmp->handle, I915_GEM_DOMAIN_WC, I915_GEM_DOMAIN_WC); > + > + for (j = 0; j < obj->size / sizeof(*buf); j++) > + buf[j] = seed++; > + munmap(buf, obj->size); > + > + memset(&blt, 0, sizeof(blt)); > + blt.color_depth = CD_32bit; > + > + memcpy(&blt.src, tmp, sizeof(blt.src)); > + memcpy(&blt.dst, obj->blt_obj, sizeof(blt.dst)); > + memcpy(&blt.bb, cmd, sizeof(blt.bb)); > + > + set_object_ext(&ext.src, 0, tmp->x2, tmp->y2, SURFACE_TYPE_2D); > + set_object_ext(&ext.dst, 0, obj->blt_obj->x2, obj->blt_obj->y2, > + SURFACE_TYPE_2D); > + > + blt_block_copy(i915, ctx, e, ahnd, &blt, pext); > + free(cmd); > +} > + > static void > verify_object(int i915, const struct object *obj, unsigned int flags) > { > @@ -125,6 +221,60 @@ verify_object(int i915, const struct object *obj, unsigned int flags) > munmap(buf, obj->size); > } > > +static void > +verify_object_ccs(int i915, const struct object *obj, > + struct blt_copy_object *tmp, const intel_ctx_t *ctx, > + uint32_t region, uint64_t ahnd) > +{ > + struct blt_block_copy_data_ext ext = {}, *pext = &ext; > + const struct intel_execution_engine2 *e; > + struct blt_copy_data blt = {}; > + struct blt_copy_batch *cmd; > + uint64_t size = 4096; > + unsigned long j, val, *buf; > + > + for_each_ctx_engine(i915, ctx, e) { > + if (gem_engine_can_block_copy(i915, e)) > + break; > + } > + igt_assert_f(e, "Ctx dont have blt engine\n"); s/dont/don't/ > + > + cmd = calloc(1, sizeof(*cmd)); > + cmd->handle = gem_create_from_pool(i915, &size, region); > + set_batch(cmd, cmd->handle, size, region); > + > + memset(&blt, 0, sizeof(blt)); > + blt.color_depth = CD_32bit; > + > + memcpy(&blt.src, obj->blt_obj, sizeof(blt.src)); > + memcpy(&blt.dst, tmp, sizeof(blt.dst)); > + memcpy(&blt.bb, cmd, sizeof(blt.bb)); > + > + blt.dst.x2 = min(obj->blt_obj->x2, tmp->x2); > + blt.dst.y2 = min(obj->blt_obj->y2, tmp->y2); > + > + set_object_ext(&ext.src, 0, obj->blt_obj->x2, obj->blt_obj->y2, > + SURFACE_TYPE_2D); > + set_object_ext(&ext.dst, 0, tmp->x2, tmp->y2, SURFACE_TYPE_2D); > + blt_block_copy(i915, ctx, e, ahnd, &blt, pext); > + > + buf = gem_mmap__device_coherent(i915, tmp->handle, 0, > + obj->size, PROT_READ); > + gem_set_domain(i915, tmp->handle, I915_GEM_DOMAIN_WC, 0); > + > + for (j = 0; j < obj->size / PAGE_SIZE; j++) { > + unsigned long x = (j * PAGE_SIZE + rand() % PAGE_SIZE) / sizeof(*buf); > + > + val = obj->seed + x; > + igt_assert_f(buf[x] == val, > + "Object mismatch at offset %lu - found %lx, expected %lx, difference:%lx!\n", > + x * sizeof(*buf), buf[x], val, buf[x] ^ val); > + } > + > + munmap(buf, obj->size); > + free(cmd); > +} > + > static void move_to_lmem(int i915, > const intel_ctx_t *ctx, > struct object *list, > @@ -174,13 +324,25 @@ static void __do_evict(int i915, > region->memory_instance); > const unsigned int max_swap_in = params->count / 100 + 1; > struct object *objects, *obj, *list; > + const uint32_t bpp = 32; > + uint32_t width, height, stride; > + const intel_ctx_t *blt_ctx; > + struct blt_copy_object *tmp; > unsigned int engine = 0; > unsigned int i, l; > - uint64_t size; > + uint64_t size, ahnd; > struct timespec t = {}; > unsigned int num; > > - size = 4096; > + width = PAGE_SIZE / (bpp / 8); > + height = params->size.max / (bpp / 8) / width; > + stride = width * 4; > + > + tmp = calloc(1, sizeof(*tmp)); > + if (params->flags & TEST_CCS) { > + blt_ctx = intel_ctx_0(i915); > + __gem_context_set_persistence(i915, blt_ctx->id, false); > + } > > objects = calloc(params->count, sizeof(*objects)); > igt_assert(objects); > @@ -191,6 +353,19 @@ static void __do_evict(int i915, > srand(seed); > > /* Create the initial working set of objects. */ > + if (params->flags & TEST_CCS) { > + ahnd = intel_allocator_open_full(i915, blt_ctx->id, 0, 0, > + INTEL_ALLOCATOR_SIMPLE, > + ALLOC_STRATEGY_LOW_TO_HIGH, 0); > + > + tmp->handle = gem_create_in_memory_regions(i915, params->size.max, > + INTEL_MEMORY_REGION_ID(I915_SYSTEM_MEMORY, 0)); > + set_object(tmp, tmp->handle, params->size.max, > + INTEL_MEMORY_REGION_ID(I915_SYSTEM_MEMORY, 0), DG2_MOCS, Use intel_get_uc_mocs() from intel_mocs.h. Otherwise only DG2 is supported by it. > + T_LINEAR, COMPRESSION_DISABLED, COMPRESSION_TYPE_3D); > + set_geom(tmp, stride, 0, 0, width, height, 0, 0); > + } > + > size = 0; > for (i = 0, obj = objects; i < params->count; i++, obj++) { > if (params->flags & TEST_RANDOM) > @@ -200,6 +375,7 @@ static void __do_evict(int i915, > else > obj->size = params->size.min; > > + obj->size = ALIGN(obj->size, 4096); > size += obj->size; > if ((size >> 20) > params->mem_limit) { > params->count = i; > @@ -207,10 +383,26 @@ static void __do_evict(int i915, > } > obj->handle = create_bo(i915, &obj->size, region, params->oom_test); > > - move_to_lmem(i915, ctx, objects + i, 1, region_id, engine, > - params->oom_test); > - if (params->flags & TEST_VERIFY) > + if (params->flags & TEST_CCS) { > + width = PAGE_SIZE / (bpp / 8); > + height = obj->size / (bpp / 8) / width; > + stride = width * 4; > + > + obj->blt_obj = calloc(1, sizeof(*obj->blt_obj)); Where are those objects freed? > + set_object(obj->blt_obj, obj->handle, obj->size, region_id, > + DG2_MOCS, T_LINEAR, COMPRESSION_ENABLED, Same, use intel_get_uc_mocs(). On what kernel are you testing it? I got failure on 5.18.0-rc1+ : Starting subtest: verify-ccs Starting dynamic subtest: lmem0 Memory: system-total 31763MiB, lmem-region 4048MiB, usage-limit 31046MiB Using 1 thread(s), 6072 loop(s), 6072 objects of 1 MiB - 1 MiB, seed: 1649420849, oom: no (gem_lmem_swapping:5987) CRITICAL: Test assertion failure function verify_object_ccs, file ../tests/i915/gem_lmem_swapping.c:270: (gem_lmem_swapping:5987) CRITICAL: Failed assertion: buf[x] == val (gem_lmem_swapping:5987) CRITICAL: Object mismatch at offset 701656 - found 67b88003, expected 67b88004, difference:7! -- Zbigniew > + COMPRESSION_TYPE_3D); > + set_geom(obj->blt_obj, stride, 0, 0, width, height, 0, 0); > + init_object_ccs(i915, obj, tmp, rand(), blt_ctx, > + region_id, ahnd); > + } else if (params->flags & TEST_VERIFY) { > init_object(i915, obj, rand(), params->flags); > + move_to_lmem(i915, ctx, objects + i, 1, region_id, engine, > + params->oom_test); > + } else { > + move_to_lmem(i915, ctx, objects + i, 1, region_id, engine, > + params->oom_test); > + } > } > > igt_debug("obj size min/max=%lu %s/%lu %s, count=%u, seed: %u\n", > @@ -238,7 +430,15 @@ static void __do_evict(int i915, > if (params->flags & TEST_ENGINES) > engine = (engine + 1) % __num_engines__; > > - if (params->flags & TEST_VERIFY) { > + if (params->flags & TEST_CCS) { > + for (i = 0; i < num; i++) > + verify_object_ccs(i915, &list[i], tmp, > + blt_ctx, region_id, ahnd); > + /* Update random object - may swap it back in. */ > + i = rand() % params->count; > + init_object_ccs(i915, &objects[i], tmp, rand(), > + blt_ctx, region_id, ahnd); > + } else if (params->flags & TEST_VERIFY) { > for (i = 0; i < num; i++) > verify_object(i915, &list[i], params->flags); > > @@ -253,6 +453,12 @@ static void __do_evict(int i915, > > free(list); > free(objects); > + > + if (params->flags & TEST_CCS) { > + gem_close(i915, tmp->handle); > + intel_ctx_destroy(i915, blt_ctx); > + put_ahnd(ahnd); > + } > } > > static void fill_params(int i915, struct params *params, > @@ -354,6 +560,9 @@ static void test_evict(int i915, > const unsigned int nproc = sysconf(_SC_NPROCESSORS_ONLN) + 1; > struct params params; > > + if (flags & TEST_CCS) > + igt_require(IS_DG2(intel_get_drm_devid(i915))); > + > fill_params(i915, ¶ms, region, flags, nproc, false); > > if (flags & TEST_PARALLEL) { > @@ -526,6 +735,11 @@ igt_main_args("", long_options, help_str, opt_handler, NULL) > { "parallel-random-engines", TEST_PARALLEL | TEST_RANDOM | TEST_ENGINES }, > { "parallel-random-verify", TEST_PARALLEL | TEST_RANDOM | TEST_VERIFY }, > { "parallel-multi", TEST_PARALLEL | TEST_RANDOM | TEST_VERIFY | TEST_ENGINES | TEST_MULTI }, > + { "verify-ccs", TEST_CCS }, > + { "verify-random-ccs", TEST_CCS | TEST_RANDOM }, > + { "heavy-verify-random-ccs", TEST_CCS | TEST_RANDOM | TEST_HEAVY }, > + { "heavy-verify-multi-ccs", TEST_CCS | TEST_RANDOM | TEST_HEAVY | TEST_ENGINES | TEST_MULTI }, > + { "parallel-random-verify-ccs", TEST_PARALLEL | TEST_RANDOM | TEST_CCS }, > { } > }; > const intel_ctx_t *ctx; > -- > 2.20.1 >