[PATCHv2 1/2] drm/amdkfd: Populate cache info for GFX 9.4.3
Felix Kuehling
felix.kuehling at amd.com
Fri Oct 27 20:06:18 UTC 2023
On 2023-10-27 15:04, Mukul Joshi wrote:
> GFX 9.4.3 uses a new version of the GC info table which
> contains the cache info. This patch adds a new function
> to populate the cache info from IP discovery for GFX 9.4.3.
>
> Signed-off-by: Mukul Joshi <mukul.joshi at amd.com>
> ---
> v1->v2:
> - Separate out the original patch into 2 patches.
>
> drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 66 ++++++++++++++++++++++++++-
> 1 file changed, 65 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
> index 0e792a8496d6..cd8e459201f1 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
> @@ -1404,6 +1404,66 @@ static int kfd_fill_gpu_cache_info_from_gfx_config(struct kfd_dev *kdev,
> return i;
> }
>
> +static int kfd_fill_gpu_cache_info_from_gfx_config_v2(struct kfd_dev *kdev,
> + struct kfd_gpu_cache_info *pcache_info)
> +{
> + struct amdgpu_device *adev = kdev->adev;
> + int i = 0;
> +
> + /* TCP L1 Cache per CU */
> + if (adev->gfx.config.gc_tcp_size_per_cu) {
> + pcache_info[i].cache_size = adev->gfx.config.gc_tcp_size_per_cu;
> + pcache_info[i].cache_level = 1;
> + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED |
> + CRAT_CACHE_FLAGS_DATA_CACHE |
> + CRAT_CACHE_FLAGS_SIMD_CACHE);
> + pcache_info[i].num_cu_shared = 1;
> + i++;
> + }
> + /* Scalar L1 Instruction Cache per SQC */
> + if (adev->gfx.config.gc_l1_instruction_cache_size_per_sqc) {
> + pcache_info[i].cache_size =
> + adev->gfx.config.gc_l1_instruction_cache_size_per_sqc;
> + pcache_info[i].cache_level = 1;
> + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED |
> + CRAT_CACHE_FLAGS_INST_CACHE |
> + CRAT_CACHE_FLAGS_SIMD_CACHE);
> + pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_cu_per_sqc;
> + i++;
> + }
> + /* Scalar L1 Data Cache per SQC */
> + if (adev->gfx.config.gc_l1_data_cache_size_per_sqc) {
> + pcache_info[i].cache_size = adev->gfx.config.gc_l1_data_cache_size_per_sqc;
> + pcache_info[i].cache_level = 1;
> + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED |
> + CRAT_CACHE_FLAGS_DATA_CACHE |
> + CRAT_CACHE_FLAGS_SIMD_CACHE);
> + pcache_info[i].num_cu_shared = adev->gfx.config.gc_num_cu_per_sqc;
> + i++;
> + }
> + /* L2 Data Cache per GPU (Total Tex Cache) */
> + if (adev->gfx.config.gc_tcc_size) {
> + pcache_info[i].cache_size = adev->gfx.config.gc_tcc_size;
> + pcache_info[i].cache_level = 2;
> + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED |
> + CRAT_CACHE_FLAGS_DATA_CACHE |
> + CRAT_CACHE_FLAGS_SIMD_CACHE);
> + pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh;
> + i++;
> + }
> + /* L3 Data Cache per GPU */
> + if (adev->gmc.mall_size) {
> + pcache_info[i].cache_size = adev->gmc.mall_size / 1024;
Is this /1024 a unit conversion? What are the units for L1/L2 caches
compared to L3 caches?
When we report the sizes in the topology, they should be in the same
units for all cache levels, I believe. Given that L3 is likely the
largest, I'm a bit suspicious of this conversion.
Other than that, the series is
Reviewed-by: Felix Kuehling <Felix.Kuehling at amd.com>
> + pcache_info[i].cache_level = 3;
> + pcache_info[i].flags = (CRAT_CACHE_FLAGS_ENABLED |
> + CRAT_CACHE_FLAGS_DATA_CACHE |
> + CRAT_CACHE_FLAGS_SIMD_CACHE);
> + pcache_info[i].num_cu_shared = adev->gfx.config.max_cu_per_sh;
> + i++;
> + }
> + return i;
> +}
> +
> int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pcache_info)
> {
> int num_of_cache_types = 0;
> @@ -1461,10 +1521,14 @@ int kfd_get_gpu_cache_info(struct kfd_node *kdev, struct kfd_gpu_cache_info **pc
> num_of_cache_types = ARRAY_SIZE(vega20_cache_info);
> break;
> case IP_VERSION(9, 4, 2):
> - case IP_VERSION(9, 4, 3):
> *pcache_info = aldebaran_cache_info;
> num_of_cache_types = ARRAY_SIZE(aldebaran_cache_info);
> break;
> + case IP_VERSION(9, 4, 3):
> + num_of_cache_types =
> + kfd_fill_gpu_cache_info_from_gfx_config_v2(kdev->kfd,
> + *pcache_info);
> + break;
> case IP_VERSION(9, 1, 0):
> case IP_VERSION(9, 2, 2):
> *pcache_info = raven_cache_info;
More information about the amd-gfx
mailing list