ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-01 20:21:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add support for RDNA1 GPUs (#3220)

Browse Source 
* Allow compilation for RDNA1 (__gfx101__)

Signed-off-by: Gavin Zhao <git@gzgz.dev>

* More RDNA1 changes

Signed-off-by: Gavin Zhao <git@gzgz.dev>

* Even more RDNA1 changes

Signed-off-by: Gavin Zhao <git@gzgz.dev>

* cmake: skip build quantization for unsupported arches

* add gfx10-1-generic support as well

* add gfx1013 and complete gfx10-1-generic

* fix clang format

* enable DL kernels on gfx101x

---------

Signed-off-by: Gavin Zhao <git@gzgz.dev>
Co-authored-by: illsilin_amdeng <Illia.Silin@amd.com>
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
This commit is contained in:
Gavin Zhao
2025-11-20 13:45:57 -05:00
committed by GitHub
parent bb155ef678
commit 07314ac543
12 changed files with 40 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_dl.hpp
View File

@@ -71,7 +71,7 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
const Block2CTileMap block_2_ctile_map)
{
#if(defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx94__) || \
defined(__gfx103__) || defined(__gfx11__) || defined(__gfx12__))
defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__) || defined(__gfx12__))
const index_t num_blocks_per_batch =
__builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);

4
include/ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_dl.hpp
View File

@@ -50,8 +50,8 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
const CGridDesc_M0_M10_M11_N0_N10_N11 e_grid_desc_m0_m10_m11_n0_n10_n11,
const Block2CTileMap block_2_ctile_map)
{
#if(defined(__gfx906__) || defined(__gfx9__) || defined(__gfx103__) || defined(__gfx11__) || \
defined(__gfx12__))
#if(defined(__gfx906__) || defined(__gfx9__) || defined(__gfx101__) || defined(__gfx103__) || \
defined(__gfx11__) || defined(__gfx12__))
constexpr index_t shared_block_size =
GridwiseGemm::GetSharedMemoryNumberOfByte() / sizeof(ABDataType);

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_weight_dl.hpp
View File

@@ -49,7 +49,7 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
#if(defined(__gfx906__) || defined(__gfx103__) || defined(__gfx90a__) || defined(__gfx908__) || \
defined(__gfx94__) || defined(__gfx11__) || defined(__gfx12__))
defined(__gfx94__) || defined(__gfx101__) || defined(__gfx11__) || defined(__gfx12__))
const index_t num_blocks_per_batch =
__builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);
const index_t g_idx = __builtin_amdgcn_readfirstlane(get_block_1d_id() / num_blocks_per_batch);

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_multiple_d_nhwc_kyxc_nhwk.hpp
View File

@@ -95,7 +95,7 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
#if(defined(__gfx906__) || defined(__gfx103__) || defined(__gfx90a__) || defined(__gfx908__) || \
defined(__gfx94__) || defined(__gfx11__) || defined(__gfx12__))
defined(__gfx94__) || defined(__gfx101__) || defined(__gfx11__) || defined(__gfx12__))
// offset base pointer for each work-group
const index_t num_blocks_per_batch =
__builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);

3
include/ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_dl_nhwc_kyxc_nhwk.hpp
View File

@@ -106,7 +106,8 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
const Block2CTileMap block_2_ctile_map,
const ComputePtrOffsetOfBatch compute_ptr_offset_of_batch)
{
#if(defined(__gfx906__) || defined(__gfx103__) || defined(__gfx11__) || defined(__gfx12__))
#if(defined(__gfx906__) || defined(__gfx101__) || defined(__gfx103__) || defined(__gfx11__) || \
defined(__gfx12__))
// offset base pointer for each work-group
const index_t num_blocks_per_batch =
__builtin_amdgcn_readfirstlane(get_grid_size() / batch_count);

4
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_dl.hpp
View File

@@ -39,8 +39,8 @@ __launch_bounds__(CK_MAX_THREAD_PER_BLOCK, CK_MIN_BLOCK_PER_CU)
const BElementwiseOperation b_element_op,
const CDEElementwiseOperation cde_element_op)
{
#if(defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx103__) || \
defined(__gfx11__) || defined(__gfx94__) || defined(__gfx12__))
#if(defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(__gfx101__) || \
defined(__gfx103__) || defined(__gfx11__) || defined(__gfx94__) || defined(__gfx12__))
__shared__ char p_shared[GridwiseGemm::GetSharedMemoryNumberOfByte()];
const index_t block_id = get_block_1d_id();