Merge branch 'develop' into ck-flex

6dcc40d4 · Max Podkorytov · GitHub · 2c8e04aa · 800cf897 · 6dcc40d4
Unverified Commit 6dcc40d4 authored Feb 04, 2025 by Max Podkorytov Committed by GitHub Feb 04, 2025
20 changed files
--- a/include/ck/utility/scaled_type_convert.hpp
+++ b/include/ck/utility/scaled_type_convert.hpp
--- a/include/ck/utility/type_convert.hpp
+++ b/include/ck/utility/type_convert.hpp
--- a/include/ck_tile/core.hpp
+++ b/include/ck_tile/core.hpp
@@ -27,6 +27,7 @@
 #include "ck_tile/core/numeric/float8.hpp"
 #include "ck_tile/core/numeric/half.hpp"
 #include "ck_tile/core/numeric/int8.hpp"
+#include "ck_tile/core/numeric/pk_int4.hpp"
 #include "ck_tile/core/numeric/integer.hpp"
 #include "ck_tile/core/numeric/integral_constant.hpp"
 #include "ck_tile/core/numeric/math.hpp"

--- a/include/ck_tile/core/config.hpp
+++ b/include/ck_tile/core/config.hpp
@@ -144,6 +144,10 @@
 #define CK_TILE_USE_AMD_BUFFER_ATOMIC_ADD_INTEGER 1
 #endif
+#ifndef CK_TILE_USE_PK4_LAYOUT_SHUFFLE
+#define CK_TILE_USE_PK4_LAYOUT_SHUFFLE 1
+#endif
 // buffer atomic add: floating point
 #ifndef __HIP_DEVICE_COMPILE__ // for host code
 #define CK_TILE_USE_AMD_BUFFER_ATOMIC_ADD_FLOAT 1

--- a/include/ck_tile/core/numeric/pk_int4.hpp
+++ b/include/ck_tile/core/numeric/pk_int4.hpp
--- a/include/ck_tile/core/numeric/vector_type.hpp
+++ b/include/ck_tile/core/numeric/vector_type.hpp
 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2025, Advanced Micro Devices, Inc. All rights reserved.
 #pragma once
@@ -200,4 +200,21 @@ using bf8x32_t = bf8_t __attribute((ext_vector_type(32)));
 using bf8x64_t = bf8_t __attribute((ext_vector_type(64)));
 #endif
+CK_TILE_HOST fp16x2_t pk_add_f16(const fp16x2_t& x, const fp16x2_t& y)
+{
+    fp16x2_t vector_res;
+    vector_res.x = x.x + y.x;
+    vector_res.y = x.y + y.y;
+    return vector_res;
+}
+CK_TILE_DEVICE fp16x2_t pk_add_f16(const fp16x2_t& x, const fp16x2_t& y)
+{
+    fp16x2_t c;
+    asm volatile("v_pk_add_f16 %0, %1, %2" : "=v"(c) : "v"(x), "v"(y));
+    return c;
+}
 } // namespace ck_tile
--- a/include/ck_tile/ops/flatmm/block/uk/flatmm_sn_uk_gfx9_32x128x512_1x4x1_16x16x16.inc
+++ b/include/ck_tile/ops/flatmm/block/uk/flatmm_sn_uk_gfx9_32x128x512_1x4x1_16x16x16.inc
@@ -824,4 +824,4 @@
 #undef _UK_PK_CVT_
 #undef _UK_ATOMIC_ADD_
 #undef CK_TILE_FLATMM_UK_MFMA
-// clang-format on
+    // clang-format on
--- a/include/ck_tile/ops/flatmm/block/uk/flatmm_sn_uk_gfx9_32x128x512_1x4x1_16x16x16_itl.inc
+++ b/include/ck_tile/ops/flatmm/block/uk/flatmm_sn_uk_gfx9_32x128x512_1x4x1_16x16x16_itl.inc
@@ -722,4 +722,4 @@
 #undef _UK_PK_CVT_
 #undef _UK_ATOMIC_ADD_
 #undef CK_TILE_FLATMM_UK_MFMA
-// clang-format on
+    // clang-format on
--- a/include/ck_tile/ops/flatmm/block/uk/flatmm_uk_gfx9_32x512x128_1x1x1_16x16x16.inc
+++ b/include/ck_tile/ops/flatmm/block/uk/flatmm_uk_gfx9_32x512x128_1x1x1_16x16x16.inc
@@ -771,4 +771,4 @@
 #undef _UK_MFMA_
 #undef CK_TILE_FLATMM_UK_2B
 #undef CK_TILE_FLATMM_UK_MFMA
-// clang-format on
+    // clang-format on
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_two_stage_xdl_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_bwd_weight/device_grouped_conv_bwd_weight_two_stage_xdl_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_comp_instance.hpp
--- a/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp
+++ b/library/include/ck/library/tensor_operation_instance/gpu/grouped_conv_fwd/device_grouped_conv_fwd_xdl_merged_groups_instance.hpp
--- a/library/src/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/tensor_operation_instance/gpu/CMakeLists.txt
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instance.cpp
@@ -27,12 +27,15 @@ using PassThrough = ck::tensor_operation::element_wise::PassThrough;
 // Compilation parameters for a[k, m] * b[k, n] = c[m, n]
 using device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple<
-    // clang-format off
+// clang-format off
        //##########|        AData| BData| CData| AccData| ALayout| BLayout| CLayout|           A|           B|           C| Block|  MPer|  NPer| K0Per| K1| MPer| NPer| MXdl| NXdl|  ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockTransfer| ABlockLds|  BBlockTransfer| BBlockTransfer| BBlockTransfer| BlockTransfer| BBlockTransfer| BBlockTransfer| BBlockLds| CThreadTransfer| CThreadTransfer| NumGemmK|          LoopScheduler|                    Pipeline|
        //##########|         Type|  Type|  Type|    Type|        |        |        | Elementwise| Elementwise| Elementwise|  Size| Block| Block| Block|   |  XDL|  XDL|  Per|  Per|   ThreadCluster|  ThreadCluster| SrcAccessOrder|   SrcVectorDim|      SrcScalar|      DstScalar| AddExtraM|   ThreadCluster|  ThreadCluster| SrcAccessOrder|  SrcVectorDim|      SrcScalar|      DstScalar| AddExtraN| SrcDstVectorDim|       DstScalar| Prefetch|                       |                            |
        //##########|             |      |      |        |        |        |        |   Operation|   Operation|   Operation|      |      |      |      |   |     |     | Wave| Wave| Lengths_K0_M_K1|   ArrangeOrder|               |               |      PerVector|   PerVector_K1|          | Lengths_K0_N_K1|   ArrangeOrder|               |              |      PerVector|   PerVector_K1|          |                |       PerVector| Stage   |                       |                            |
        //##########|             |      |      |        |        |        |        |            |            |            |      |      |      |      |   |     |     |     |     |                |               |               |               |               |               |          |                |               |               |              |               |               |          |                |                |         |                       |                            |
        // pipeline v1, 1 wave
+#if defined(CK_USE_AMD_MFMA_GFX950)
+        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   128,     4, 16,   32,   32,    2,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v1>
+#else
        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   256,   128,     4,  8,   32,   32,    4,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v1>,
        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,   256,     4,  8,   32,   32,    2,    4,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v1>,
        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   128,   128,   128,     4,  8,   32,   32,    4,    2,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              4,              8,      true,     S<4, 32, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              4,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v1>,
@@ -65,6 +68,7 @@ using device_batched_gemm_xdl_f16_f16_f16_gkm_gkn_gmn_instances = std::tuple<
        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,   128,    64,     4,  8,   32,   32,    2,    1,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              2,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              1,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v2>,
        DeviceBatchedGemmXdl<  F16,   F16,   F16,     F32,     Col,      Row,    Row, PassThrough, PassThrough, PassThrough,   256,    64,   128,     4,  8,   32,   32,    1,    2,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,              1,              1,              8,      true,     S<4, 64, 1>,     S<0, 2, 1>,     S<0, 2, 1>,             1,              2,              8,      true,               7,               1,        1, LoopScheduler::Default,        PipelineVersion::v2>
 #endif
+#endif // defined(CK_USE_AMD_MFMA_GFX950)
    // clang-format on
    >;

--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gkm_gnk_gmn_instance.cpp
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gkn_gmn_instance.cpp
--- a/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm/device_batched_gemm_xdl_f16_f16_f16_gmk_gnk_gmn_instance.cpp
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm/device_batched_gemm_softmax_gemm_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_bf16_bf16_bf16_bf16_gmk_gnk_gno_gmo_instance.cpp
--- a/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp
+++ b/library/src/tensor_operation_instance/gpu/batched_gemm_softmax_gemm_permute/device_batched_gemm_bias_softmax_gemm_permute_xdl_cshuffle_f16_f16_f16_f16_gmk_gnk_gno_gmo_instance.cpp