Merge branch 'develop' into lwpck-1815

6bbb94f4 · Rostyslav Geyyer · GitHub · 4c850c90 · a32b1bc6 · 6bbb94f4
Unverified Commit 6bbb94f4 authored Jun 26, 2024 by Rostyslav Geyyer Committed by GitHub Jun 26, 2024
20 changed files
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_v4.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_v4.hpp
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_v5.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_pipeline_xdlops_v5.hpp
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_wmma.hpp
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops.hpp
--- a/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
+++ b/include/ck/tensor_operation/gpu/block/blockwise_gemm_xdlops_skip_b_lds.hpp
--- a/include/ck/tensor_operation/gpu/device/helper.hpp
+++ b/include/ck/tensor_operation/gpu/device/helper.hpp
--- a/include/ck/tensor_operation/gpu/device/impl/codegen_device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp
+++ b/include/ck/tensor_operation/gpu/device/impl/codegen_device_grouped_conv_fwd_multiple_abd_xdl_cshuffle.hpp
--- a/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
+++ b/include/ck/tensor_operation/gpu/device/matrix_padder.hpp
--- a/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
+++ b/include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
--- a/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
+++ b/include/ck/tensor_operation/operator_transform/transform_conv_fwd_to_gemm.hpp
--- a/include/ck/utility/amd_buffer_addressing.hpp
+++ b/include/ck/utility/amd_buffer_addressing.hpp
@@ -991,7 +991,8 @@ __device__ void amd_direct_load_global_to_lds(const T* global_base_ptr,
    asm volatile("s_mov_b32 m0, %0; \n\t"
                 "buffer_load_dword %1, %2, 0 offen lds;\n\t" ::"s"(lds_ptr_sgpr),
                 "v"(global_offset_bytes),
-                 "s"(src_resource));
+                 "s"(src_resource)
+                 : "memory");
 #else
    // LDS pointer must be attributed with the LDS address space.
    __attribute__((address_space(3))) uint32_t* lds_ptr =

--- a/include/ck/utility/array.hpp
+++ b/include/ck/utility/array.hpp
--- a/include/ck_tile/core.hpp
+++ b/include/ck_tile/core.hpp
--- a/include/ck_tile/core/arch/amd_buffer_addressing.hpp
+++ b/include/ck_tile/core/arch/amd_buffer_addressing.hpp
--- a/include/ck_tile/core/numeric/null_type.hpp
+++ b/include/ck_tile/core/numeric/null_type.hpp
--- a/include/ck_tile/host.hpp
+++ b/include/ck_tile/host.hpp
--- a/include/ck_tile/host/check_err.hpp
+++ b/include/ck_tile/host/check_err.hpp
--- a/include/ck_tile/host/reference/reference_layernorm2d.hpp
+++ b/include/ck_tile/host/reference/reference_layernorm2d.hpp
--- a/include/ck_tile/host/timer.hpp
+++ b/include/ck_tile/host/timer.hpp
--- a/include/ck_tile/ops/fmha.hpp
+++ b/include/ck_tile/ops/fmha.hpp