Merge remote-tracking branch 'origin/develop' into mozga-amd/universal_gemm_weight

2a768425 · Mateusz Ozga · f358055f · 1c45ca35 · 2a768425 · 2a768425
Commit 2a768425 authored Dec 20, 2024 by Mateusz Ozga
18 changed files
--- a/include/ck_tile/ops/fmha/pipeline/tile_fmha_shape.hpp
+++ b/include/ck_tile/ops/fmha/pipeline/tile_fmha_shape.hpp
@@ -43,8 +43,6 @@ struct TileFmhaShape
    static constexpr index_t NumWarps = max(NumGemm0Warps, NumGemm1Warps);
-    static_assert(std::is_same_v<Gemm0WarpTile, Gemm1WarpTile>);
    static constexpr index_t kM0 = BlockTile::at(number<0>{}); // tile size along q seqlen
    static constexpr index_t kN0 = BlockTile::at(number<1>{}); // tile size along k seqlen
    static constexpr index_t kK0 = BlockTile::at(number<2>{}); // tile size along qk gemm unroll

--- a/include/ck_tile/ops/gemm/block/block_gemm_areg_bsmem_creg_one_warp_v1.hpp
+++ b/include/ck_tile/ops/gemm/block/block_gemm_areg_bsmem_creg_one_warp_v1.hpp
--- a/include/ck_tile/ops/gemm/block/block_gemm_areg_bsmem_creg_v2.hpp
+++ b/include/ck_tile/ops/gemm/block/block_gemm_areg_bsmem_creg_v2.hpp
--- a/include/ck_tile/ops/gemm/kernel/batched_gemm_kernel.hpp
+++ b/include/ck_tile/ops/gemm/kernel/batched_gemm_kernel.hpp
--- a/include/ck_tile/ops/gemm/kernel/gemm_kernel.hpp
+++ b/include/ck_tile/ops/gemm/kernel/gemm_kernel.hpp
--- a/include/ck_tile/ops/gemm/warp/warp_gemm.hpp
+++ b/include/ck_tile/ops/gemm/warp/warp_gemm.hpp
--- a/include/ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma.hpp
+++ b/include/ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma.hpp
--- a/include/ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma_impl.hpp
+++ b/include/ck_tile/ops/gemm/warp/warp_gemm_attribute_mfma_impl.hpp
--- a/include/ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp
+++ b/include/ck_tile/ops/gemm/warp/warp_gemm_dispatcher.hpp
--- a/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_kn_mn.hpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_kn_mn.hpp
--- a/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp
+++ b/library/src/tensor_operation_instance/gpu/gemm_universal_streamk/device_gemm_xdl_universal_streamk_f16_f16_f16/device_gemm_xdl_universal_streamk_f16_f16_f16_mk_nk_mn.hpp
--- a/profiler/include/profiler/profile_gemm_universal_impl.hpp
+++ b/profiler/include/profiler/profile_gemm_universal_impl.hpp
--- a/profiler/include/profiler/profile_grouped_gemm_impl.hpp
+++ b/profiler/include/profiler/profile_grouped_gemm_impl.hpp
@@ -77,7 +77,7 @@ bool profile_grouped_gemm_impl(int do_verification,
    std::vector<Tensor<CDataType>> c_m_n_host_results;
    std::vector<Tensor<CDataType>> c_m_n_device_results;
-    ComputeDataType max_abs_in_val = 0.f;
+    double max_abs_in_val = 0.f;
    for(std::size_t i = 0; i < group_count; i++)
    {
        a_m_k.push_back(

--- a/test/ck_tile/batched_gemm/test_batched_gemm_util.hpp
+++ b/test/ck_tile/batched_gemm/test_batched_gemm_util.hpp
--- a/test/ck_tile/gemm/CMakeLists.txt
+++ b/test/ck_tile/gemm/CMakeLists.txt
--- a/test/ck_tile/gemm/test_gemm_mem_pipeline.cpp
+++ b/test/ck_tile/gemm/test_gemm_mem_pipeline.cpp
--- a/test/ck_tile/gemm/test_gemm_mem_pipeline_ut_cases.inc
+++ b/test/ck_tile/gemm/test_gemm_mem_pipeline_ut_cases.inc
--- a/test/ck_tile/gemm/test_gemm_mem_pipeline_util.hpp
+++ b/test/ck_tile/gemm/test_gemm_mem_pipeline_util.hpp