Merge branch 'develop' into conv_layout

Dockerfile

 FROM ubuntu:20.04
 
-ARG ROCMVERSION=5.3
+ARG ROCMVERSION=5.4.3
 ARG compiler_version="release"
 ARG compiler_commit=""
 
@@ -8,23 +8,27 @@ RUN set -xe
 
 ARG DEB_ROCM_REPO=http://repo.radeon.com/rocm/apt/.apt_$ROCMVERSION/
 RUN useradd -rm -d /home/jenkins -s /bin/bash -u 1004 jenkins
-RUN useradd -rm -d /home/manitera -s /bin/bash -u 1002 manitera
 # Add rocm repository
 RUN apt-get update
-RUN apt-get install -y wget gnupg
-RUN wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add -
+RUN apt-get install -y wget gnupg curl
+RUN --mount=type=ssh if [ "$ROCMVERSION" != "5.5"]; then \
+	wget -qO - http://repo.radeon.com/rocm/rocm.gpg.key | apt-key add - ; \
+    else sh -c "wget http://artifactory-cdn.amd.com/artifactory/list/amdgpu-deb/amd-nonfree-radeon_20.04-1_all.deb" && \
+         apt update && apt-get install -y ./amd-nonfree-radeon_20.04-1_all.deb && \
+         sh -c 'echo deb [arch=amd64 trusted=yes] http://compute-artifactory.amd.com/artifactory/list/rocm-release-archive-20.04-deb/ 5.5 rel-50 > /etc/apt/sources.list.d/rocm-build.list' && \
+         amdgpu-repo --amdgpu-build=1558725 && DEBIAN_FRONTEND=noninteractive amdgpu-install -y --usecase=rocm ; \
+    fi
 RUN sh -c "echo deb [arch=amd64] $DEB_ROCM_REPO ubuntu main > /etc/apt/sources.list.d/rocm.list"
 RUN wget --no-check-certificate -qO - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | apt-key add -
 RUN sh -c "echo deb http://mirrors.kernel.org/ubuntu focal main universe | tee -a /etc/apt/sources.list"
+RUN curl -fsSL https://repo.radeon.com/rocm/rocm.gpg.key | gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm-keyring.gpg
 
 # Install dependencies
 RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-unauthenticated \
     apt-utils \
     build-essential \
     ccache \
-    cmake-data \
     cmake \
-    curl \
     git \
     hip-rocclr \
     jq \
@@ -45,6 +49,7 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     rocm-device-libs \
     rocm-cmake \
     vim \
+    nano \
     zlib1g-dev \
     openssh-server \
     clang-format-10 \
@@ -52,6 +57,17 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --allow-
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
+#Install latest version of cmake
+RUN apt purge --auto-remove -y cmake
+RUN apt update
+RUN apt install -y software-properties-common lsb-release
+RUN apt clean all
+RUN wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | gpg --dearmor - | tee /etc/apt/trusted.gpg.d/kitware.gpg >/dev/null
+RUN apt-add-repository "deb https://apt.kitware.com/ubuntu/ $(lsb_release -cs) main"
+RUN apt install -y kitware-archive-keyring
+RUN rm /etc/apt/trusted.gpg.d/kitware.gpg
+RUN apt install -y cmake
+
 # Setup ubsan environment to printstacktrace
 RUN ln -s /usr/bin/llvm-symbolizer-3.8 /usr/local/bin/llvm-symbolizer
 ENV UBSAN_OPTIONS=print_stacktrace=1
@@ -87,12 +103,7 @@ ENV compiler_commit=$compiler_commit
 RUN sh -c "echo compiler version = '$compiler_version'"
 RUN sh -c "echo compiler commit = '$compiler_commit'"
 
-RUN --mount=type=ssh if [ "$compiler_version" = "amd-stg-open" ]; then \
-        sed -i '/$HIP_CLANG_TARGET = chomp($HIP_CLANG_TARGET);/c\    chomp($HIP_CLANG_TARGET);' /opt/rocm/hip/bin/hipcc.pl && \
-        sed -i '/$HIP_CLANG_TARGET = chomp($HIP_CLANG_TARGET);/c\    chomp($HIP_CLANG_TARGET);' /opt/rocm/bin/hipcc.pl; \
-    fi
-
-RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_commit" = "" ]; then \
+RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_version" !=~ ^"rc" ] && [ "$compiler_commit" = "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && mkdir build && cd build && \
         cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \
@@ -100,7 +111,7 @@ RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_com
     else echo "using the release compiler"; \
     fi
 
-RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_commit" != "" ]; then \
+RUN --mount=type=ssh if [ "$compiler_version" != "release" ] && [ "$compiler_version" !=~ ^"rc" ] && [ "$compiler_commit" != "" ]; then \
         git clone -b "$compiler_version" https://github.com/RadeonOpenCompute/llvm-project.git && \
         cd llvm-project && git checkout "$compiler_commit" && echo "checking out commit $compiler_commit" && mkdir build && cd build && \
         cmake -DCMAKE_INSTALL_PREFIX=/opt/rocm/llvm -DCMAKE_BUILD_TYPE=Release -DLLVM_ENABLE_ASSERTIONS=1 -DLLVM_TARGETS_TO_BUILD="AMDGPU;X86" -DLLVM_ENABLE_PROJECTS="clang;lld;compiler-rt" ../llvm && \

Jenkinsfile

@@ -19,12 +19,23 @@ def runShell(String command){
 
 def getDockerImageName(){
     def img
-    if (params.COMPILER_COMMIT == ""){
-        img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}"
+    if (params.ROCMVERSION != "5.5"){
+       if (params.COMPILER_COMMIT == ""){
+           img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}"
+       }
+       else{
+           def commit = "${params.COMPILER_COMMIT}"[0..6]
+           img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}_${commit}"
+       }
     }
     else{
-        def commit = "${params.COMPILER_COMMIT}"[0..6]
-        img = "${env.CK_DOCKERHUB}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}_${commit}"
+       if (params.COMPILER_COMMIT == ""){
+           img = "${env.CK_DOCKERHUB_PRIVATE}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}"
+       }
+       else{
+           def commit = "${params.COMPILER_COMMIT}"[0..6]
+           img = "${env.CK_DOCKERHUB_PRIVATE}:ck_ub20.04_rocm${params.ROCMVERSION}_${params.COMPILER_VERSION}_${commit}"
+       }
     }
     return img
 }
@@ -49,11 +60,11 @@ def build_compiler(){
         compiler = '/opt/rocm/bin/hipcc'
     }
     else{
-        if (params.COMPILER_VERSION == "release"){
-            compiler = "/opt/rocm/llvm/bin/clang++"
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
+            compiler = "/llvm-project/build/bin/clang++"
         }
         else{
-            compiler = "/llvm-project/build/bin/clang++"
+            compiler = "/opt/rocm/llvm/bin/clang++"
         }        
     }
     return compiler
@@ -232,7 +243,7 @@ def buildHipClangJob(Map conf=[:]){
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
         }
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-        if (params.COMPILER_VERSION != "release"){
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
 
@@ -287,7 +298,7 @@ def runCKProfiler(Map conf=[:]){
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
         }
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-        if (params.COMPILER_VERSION != "release"){
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
 
@@ -420,7 +431,7 @@ def Build_CK(Map conf=[:]){
             dockerOpts = dockerOpts + " --env HSA_XNACK=1 "
         }
         def dockerArgs = "--build-arg PREFIX=${prefixpath} --build-arg compiler_version='${params.COMPILER_VERSION}' --build-arg compiler_commit='${params.COMPILER_COMMIT}' --build-arg ROCMVERSION='${params.ROCMVERSION}' "
-        if (params.COMPILER_VERSION != "release"){
+        if (params.COMPILER_VERSION == "amd-stg-open" || params.COMPILER_COMMIT != ""){
             dockerOpts = dockerOpts + " --env HIP_CLANG_PATH='/llvm-project/build/bin' "
         }
 

client_example/18_groupnorm/groupnorm_swish.cpp

@@ -13,8 +13,8 @@
 #include "ck/library/tensor_operation_instance/gpu/normalization_swish.hpp"
 
 using XDataType       = ck::half_t;
-using GammaDataType   = ck::half_t;
-using BetaDataType    = ck::half_t;
+using GammaDataType   = float;
+using BetaDataType    = float;
 using YDataType       = ck::half_t;
 using ComputeDataType = float;
 using Swish           = ck::tensor_operation::element_wise::Swish;

include/ck/tensor_operation/gpu/grid/gridwise_gemm_wmma.hpp

@@ -266,8 +266,8 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_wmma
         // TODO: also check validity of all components (blockwise-copy, threadwise-copy, etc)
         constexpr long_index_t TwoGB = (long_index_t{1} << 31);
 
-        if(!(a_grid_desc_k0_m_k1.GetElementSpaceSize() * sizeof(ADataType) <= TwoGB &&
-             b_grid_desc_k0_n_k1.GetElementSpaceSize() * sizeof(BDataType) <= TwoGB))
+        if(!(a_grid_desc_k0_m_k1.GetElementSpaceSize() * sizeof(FloatA) <= TwoGB &&
+             b_grid_desc_k0_n_k1.GetElementSpaceSize() * sizeof(FloatB) <= TwoGB))
         {
             return false;
         }

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_bwd_weight.hpp

@@ -265,7 +265,7 @@ struct GridwiseGemm_bk0mk1_bk0nk1_mn_xdlops_bwd_weight
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
     // FloatABAdjusted -> FloatAB throughout this file
-#if defined(__gfx90a__)
+#if CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
     using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
 #else
     using FloatABAdjusted = FloatAB;

include/ck/tensor_operation/gpu/grid/gridwise_gemm_xdlops_v2r3.hpp

@@ -135,7 +135,7 @@ struct GridwiseGemm_k0mk1_k0nk1_mn_xdlops_v2r3
     // we convert fp16->fp32->bf16 and execute bf16 mfma instruction
     // when mfma if fixed, remove this section and update
     // FloatABAdjusted -> FloatAB throughout this file
-#if defined(__gfx90a__)
+#if CK_WORKAROUND_DENORM_FIX && defined(__gfx90a__)
     using FloatABAdjusted = conditional_t<is_same_v<FloatAB, ck::half_t>, ck::bhalf_t, FloatAB>;
 #else
     using FloatABAdjusted = FloatAB;

library/include/ck/library/tensor_operation_instance/gpu/normalization_swish.hpp

@@ -25,6 +25,10 @@ void add_device_normalization_rank_5_3_swish_f16_instances(
 void add_device_normalization_rank_5_3_swish_f32_instances(
     std::vector<std::unique_ptr<DeviceNormalization<F32, F32, F32, F32, F32, Swish, 5, 3>>>&);
 
+// [x, gamma, beta, y] = [f16, f32, f32, f16]
+void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&);
+
 template <typename XDataType,
           typename GammaDataType,
           typename BetaDataType,
@@ -70,6 +74,14 @@ struct DeviceOperationInstanceFactory<
                 add_device_normalization_rank_5_3_swish_f32_instances(op_ptrs);
             }
         }
+        else if constexpr(is_same_v<XDataType, F16> && is_same_v<GammaDataType, F32> &&
+                          is_same_v<BetaDataType, F32> && is_same_v<YDataType, F16>)
+        {
+            if constexpr(Rank == 5 && NumReduceDim == 3)
+            {
+                add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(op_ptrs);
+            }
+        }
 
         return op_ptrs;
     }

library/src/tensor_operation_instance/gpu/normalization/CMakeLists.txt

@@ -7,4 +7,5 @@ add_instance_library(device_normalization_instance
     device_groupnorm_f32_instance.cpp
     device_groupnorm_swish_f16_instance.cpp
     device_groupnorm_swish_f32_instance.cpp
+    device_groupnorm_swish_f16_f32_f32_f16_instance.cpp
 )

library/src/tensor_operation_instance/gpu/normalization/device_groupnorm_swish_f16_f32_f32_f16_instance.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "normalization_instance_common.hpp"
+
+namespace ck {
+namespace tensor_operation {
+namespace device {
+namespace instance {
+
+using Swish = ck::tensor_operation::element_wise::Swish;
+
+void add_device_normalization_rank_5_3_swish_f16_f32_f32_f16_instances(
+    std::vector<std::unique_ptr<DeviceNormalization<F16, F32, F32, F32, F16, Swish, 5, 3>>>&
+        instances)
+{
+    add_device_operation_instances(instances,
+                                   device_normalization_f16_f32_f32_f16_instances<Swish, 5, 3>{});
+}
+
+} // namespace instance
+} // namespace device
+} // namespace tensor_operation
+} // namespace ck

library/src/tensor_operation_instance/gpu/normalization/normalization_instance_common.hpp

@@ -69,6 +69,32 @@ using device_normalization_f32_instances = std::tuple<
     // clang-format on
     >;
 
+template <typename OutElementwise, index_t Rank, index_t Reduce>
+using device_normalization_f16_f32_f32_f16_instances = std::tuple<
+    // clang-format off
+        // XDataType, GammaDataType, BetaDataType, ComputeDataType, YDataType, Rank, NumReduceDim, BlockSize, MThreadClusterSize, KThreadClusterSize, MThreadSliceSize, KThreadSliceSize, XYSrcVectorDim, XSrcVectorSize, GammaSrcVectorSize, BetaSrcVectorSize, YDstVectorSize>
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 1, 1, 1, 1, 1, 1, 1, 1>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 1, 1, 1, 1, 1, 1, 1, 1>, // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 2, 1, 2, 1, 2, 1, 2, 2>,   // irregular size
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 128, 1, 128, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 2, 16, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 256, 1, 256, 1, 32, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 1, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 512, 1, 512, 2, 8, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 4, 1, 4, 1, 4, 1, 4, 4>,
+        DeviceNormalizationImpl<F16, F32, F32, F32, F16, OutElementwise, Rank, Reduce, 1024, 1, 1024, 1, 8, 1, 4, 1, 4, 1, 4, 4>
+    // clang-format on
+    >;
+
 } // namespace instance
 } // namespace device
 } // namespace tensor_operation