Re: [PATCH i-g-t v6 5/5] tests/intel/xe_ccs: Add compression support for Lunarlake

["Jahagirdar, Akshata" <akshata.jahagirdar@intel.com>]

[-- Attachment #1: Type: text/plain, Size: 6841 bytes --]


On 12/13/2023 1:02 AM, Zbigniew Kempczyński wrote:
> On Wed, Dec 13, 2023 at 11:55:09AM -0800, Akshata Jahagirdar wrote:
>> In XE2 IGFX platform, sysmem also participates in compression.
>> So create all blt objects in sysmem itself, and update the pat-index to reflect
>> the compression status. Since we need to align the buffer object size with page
>> size and also have the src size and dst size of CCS copy to be equal,
>> change the default width and height to 1024.
> To be honest 512 x 512 x 32bpp looks much more interesting. From my
> calculations:
>
> num_pages = 512 * 512 * 4 / 4096 -> 256
>
> 256 pages, 8B compression each gives 2048B so regardless page
> granularity this also should work. If not we need to fix the blt
> library.
>
> --
> Zbigniew

Hi, thank you for your comment.

In case of 512 x 512 x 32bpp that is the compressed blt object size.

the ccs size for this blt object = 512 * 512 * 4 / 512 = 2048

While creating the ccs bo of size 2048, it doesn't align properly with 
our page size, thats where the test fails.

Best,

Akshata

>> Signed-off-by: Akshata Jahagirdar<akshata.jahagirdar@intel.com>
>> ---
>>   tests/intel/xe_ccs.c | 45 ++++++++++++++++++++++++++------------------
>>   1 file changed, 27 insertions(+), 18 deletions(-)
>>
>> diff --git a/tests/intel/xe_ccs.c b/tests/intel/xe_ccs.c
>> index ac0805017..a780140fd 100644
>> --- a/tests/intel/xe_ccs.c
>> +++ b/tests/intel/xe_ccs.c
>> @@ -63,8 +63,8 @@ static struct param {
>>   	.write_png = false,
>>   	.print_bb = false,
>>   	.print_surface_info = false,
>> -	.width = 512,
>> -	.height = 512,
>> +	.width = 1024,
>> +	.height = 1024,
>>   };
>>   
>>   struct test_config {
>> @@ -99,17 +99,23 @@ static void surf_copy(int xe,
>>   	uint32_t *ccscopy;
>>   	uint8_t uc_mocs = intel_get_uc_mocs_index(xe);
>>   	uint32_t sysmem = system_memory(xe);
>> +	uint8_t comp_pat_index = DEFAULT_PAT_INDEX;
>> +	uint16_t cpu_caching = __xe_default_cpu_caching(xe, sysmem, 0);
>>   	int result;
>>   
>>   	igt_assert(mid->compression);
>> +	if (AT_LEAST_GEN(intel_get_drm_devid(xe), 20) && mid->compression) {
>> +		comp_pat_index  = intel_get_pat_idx_uc_comp(xe);
>> +		cpu_caching = DRM_XE_GEM_CPU_CACHING_WC;
>> +	}
>>   	ccscopy = (uint32_t *) malloc(ccssize);
>> -	ccs = xe_bo_create(xe, 0, ccssize, sysmem, 0);
>> -	ccs2 = xe_bo_create(xe, 0, ccssize, sysmem, 0);
>> +	ccs = xe_bo_create_caching(xe, 0, ccssize, sysmem, 0, cpu_caching);
>> +	ccs2 = xe_bo_create_caching(xe, 0, ccssize, sysmem, 0, cpu_caching);
>>   
>>   	blt_ctrl_surf_copy_init(xe, &surf);
>>   	surf.print_bb = param.print_bb;
>>   	blt_set_ctrl_surf_object(&surf.src, mid->handle, mid->region, mid->size,
>> -				 uc_mocs, DEFAULT_PAT_INDEX, BLT_INDIRECT_ACCESS);
>> +				 uc_mocs, comp_pat_index, BLT_INDIRECT_ACCESS);
>>   	blt_set_ctrl_surf_object(&surf.dst, ccs, sysmem, ccssize, uc_mocs,
>>   				 DEFAULT_PAT_INDEX, DIRECT_ACCESS);
>>   	bb_size = xe_get_default_alignment(xe);
>> @@ -157,7 +163,7 @@ static void surf_copy(int xe,
>>   	blt_set_ctrl_surf_object(&surf.src, ccs, sysmem, ccssize,
>>   				 uc_mocs, DEFAULT_PAT_INDEX, DIRECT_ACCESS);
>>   	blt_set_ctrl_surf_object(&surf.dst, mid->handle, mid->region, mid->size,
>> -				 uc_mocs, DEFAULT_PAT_INDEX, INDIRECT_ACCESS);
>> +				 uc_mocs, comp_pat_index, INDIRECT_ACCESS);
>>   	blt_ctrl_surf_copy(xe, ctx, NULL, ahnd, &surf);
>>   	intel_ctx_xe_sync(ctx, true);
>>   
>> @@ -234,10 +240,10 @@ static int blt_block_copy3(int xe,
>>   	igt_assert_f(blt3, "block-copy3 requires data to do blit\n");
>>   
>>   	alignment = xe_get_default_alignment(xe);
>> -	get_offset(ahnd, blt3->src.handle, blt3->src.size, alignment);
>> -	get_offset(ahnd, blt3->mid.handle, blt3->mid.size, alignment);
>> -	get_offset(ahnd, blt3->dst.handle, blt3->dst.size, alignment);
>> -	get_offset(ahnd, blt3->final.handle, blt3->final.size, alignment);
>> +	get_offset_pat_index(ahnd, blt3->src.handle, blt3->src.size, alignment, blt3->src.pat_index);
>> +	get_offset_pat_index(ahnd, blt3->mid.handle, blt3->mid.size, alignment, blt3->mid.pat_index);
>> +	get_offset_pat_index(ahnd, blt3->dst.handle, blt3->dst.size, alignment, blt3->dst.pat_index);
>> +	get_offset_pat_index(ahnd, blt3->final.handle, blt3->final.size, alignment, blt3->final.pat_index);
>>   	bb_offset = get_offset(ahnd, blt3->bb.handle, blt3->bb.size, alignment);
>>   
>>   	/* First blit src -> mid */
>> @@ -291,8 +297,9 @@ static void block_copy(int xe,
>>   	uint64_t bb_size = xe_get_default_alignment(xe);
>>   	uint64_t ahnd = intel_allocator_open(xe, ctx->vm, INTEL_ALLOCATOR_RELOC);
>>   	uint32_t run_id = mid_tiling;
>> -	uint32_t mid_region = region2, bb;
>> -	uint32_t width = param.width, height = param.height;
>> +	uint32_t mid_region = (AT_LEAST_GEN(intel_get_drm_devid(xe), 20) &
>> +							!xe_has_vram(xe)) ? region1 : region2;
>> +	uint32_t width = param.width, height = param.height, bb;
>>   	enum blt_compression mid_compression = config->compression;
>>   	int mid_compression_format = param.compression_format;
>>   	enum blt_compression_type comp_type = COMPRESSION_TYPE_3D;
>> @@ -413,8 +420,9 @@ static void block_multicopy(int xe,
>>   	uint64_t bb_size = xe_get_default_alignment(xe);
>>   	uint64_t ahnd = intel_allocator_open(xe, ctx->vm, INTEL_ALLOCATOR_RELOC);
>>   	uint32_t run_id = mid_tiling;
>> -	uint32_t mid_region = region2, bb;
>> -	uint32_t width = param.width, height = param.height;
>> +	uint32_t mid_region = (AT_LEAST_GEN(intel_get_drm_devid(xe), 20) &
>> +							!xe_has_vram(xe)) ? region1 : region2;
>> +	uint32_t width = param.width, height = param.height, bb;
>>   	enum blt_compression mid_compression = config->compression;
>>   	int mid_compression_format = param.compression_format;
>>   	enum blt_compression_type comp_type = COMPRESSION_TYPE_3D;
>> @@ -539,8 +547,9 @@ static void block_copy_test(int xe,
>>   			region1 = igt_collection_get_value(regions, 0);
>>   			region2 = igt_collection_get_value(regions, 1);
>>   
>> -			/* Compressed surface must be in device memory */
>> -			if (config->compression && !XE_IS_VRAM_MEMORY_REGION(xe, region2))
>> +			/* if not XE2, then Compressed surface must be in device memory */
>> +			if (config->compression && !(AT_LEAST_GEN((intel_get_drm_devid(xe)), 20)) &&
>> +									!XE_IS_VRAM_MEMORY_REGION(xe, region2))
>>   				continue;
>>   
>>   			regtxt = xe_memregion_dynamic_subtest_name(xe, regions);
>> @@ -621,8 +630,8 @@ const char *help_str =
>>   	"  -p\tWrite PNG\n"
>>   	"  -s\tPrint surface info\n"
>>   	"  -t\tTiling format (0 - linear, 1 - XMAJOR, 2 - YMAJOR, 3 - TILE4, 4 - TILE64)\n"
>> -	"  -W\tWidth (default 512)\n"
>> -	"  -H\tHeight (default 512)"
>> +	"  -W\tWidth (default 1024)\n"
>> +	"  -H\tHeight (default 1024)"
>>   	;
>>   
>>   igt_main_args("bf:pst:W:H:", NULL, help_str, opt_handler, NULL)
>> -- 
>> 2.34.1
>>

[-- Attachment #2: Type: text/html, Size: 7785 bytes --]