Merge remote-tracking branch 'rosenrodt/gemm-standalone-bench' into wavelet_model

Jenkinsfile

@@ -233,6 +233,7 @@ def buildHipClangJob(Map conf=[:]){
         def variant = env.STAGE_NAME
 
         def retimage
+        (retimage, image) = getDockerImage(conf)
 
         gitStatusWrapper(credentialsId: "${status_wrapper_creds}", gitHubContext: "Jenkins - ${variant}", account: 'ROCmSoftwarePlatform', repo: 'composable_kernel') {
             withDockerContainer(image: image, args: dockerOpts + ' -v=/var/jenkins/:/var/jenkins') {
@@ -548,8 +549,9 @@ def process_results(Map conf=[:]){
     }
 }
 
-//launch develop branch daily at 23:00 in FULL_QA mode
-CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true''' : ""
+//launch develop branch daily at 23:00 UT in FULL_QA mode and at 19:00 UT with latest staging compiler version
+CRON_SETTINGS = BRANCH_NAME == "develop" ? '''0 23 * * * % RUN_FULL_QA=true;COMPILER_VERSION=release
+                                              0 19 * * * % BUILD_DOCKER=true;COMPILER_VERSION=amd-stg-open''' : ""
 
 pipeline {
     agent none

client_example/03_gemm_layernorm/gemm_add_add_layernorm.cpp

@@ -8,7 +8,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/device_gemm_reduce.hpp"
-#include "ck/tensor_operation/gpu/device/device_elementwise.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_elementwise.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/tensor_operation_instance/gpu/device_elementwise_instance.hpp"

client_example/05_layernorm/layernorm2d.cpp

@@ -10,7 +10,7 @@
 #include "ck/tensor_operation/gpu/device/device_normalization.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
-#include "ck/library/tensor_operation_instance/gpu/layernorm.hpp"
+#include "ck/library/tensor_operation_instance/gpu/normalization.hpp"
 
 using XDataType     = ck::half_t;
 using GammaDataType = ck::half_t;
@@ -51,14 +51,14 @@ int main(int argc, char* argv[])
     SimpleDeviceMem beta_device_buf(sizeof(BetaDataType) * N);
     SimpleDeviceMem y_device_buf(sizeof(YDataType) * xy_size);
 
-    using DeviceOp = ck::tensor_operation::device::DeviceLayernorm<XDataType,
-                                                                   GammaDataType,
-                                                                   BetaDataType,
-                                                                   AccDataType,
-                                                                   YDataType,
-                                                                   PassThrough,
-                                                                   Rank,
-                                                                   NumReduceDim>;
+    using DeviceOp = ck::tensor_operation::device::DeviceNormalization<XDataType,
+                                                                       GammaDataType,
+                                                                       BetaDataType,
+                                                                       AccDataType,
+                                                                       YDataType,
+                                                                       PassThrough,
+                                                                       Rank,
+                                                                       NumReduceDim>;
 
     // get device op instances
     const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<

client_example/07_conv2d_fwd/CMakeLists.txt

+add_executable(client_conv2d_fwd conv2d_fwd.cpp)
+target_link_libraries(client_conv2d_fwd PRIVATE composable_kernel::device_operations)

client_example/07_conv2d_fwd/conv2d_fwd.cpp

+// SPDX-License-Identifier: MIT
+// Copyright (c) 2018-2022, Advanced Micro Devices, Inc. All rights reserved.
+
+#include <iomanip>
+#include <iostream>
+#include <vector>
+
+#include "ck/ck.hpp"
+#include "ck/library/tensor_operation_instance/gpu/convolution_forward.hpp"
+#include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
+#include "ck/tensor_operation/gpu/device/device_conv_fwd.hpp"
+#include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
+
+using InDataType  = ck::half_t;
+using WeiDataType = ck::half_t;
+using OutDataType = ck::half_t;
+
+using InLayout    = ck::tensor_layout::convolution::NHWC;
+using WeiLayout   = ck::tensor_layout::convolution::KYXC;
+using OutLayout   = ck::tensor_layout::convolution::NHWK;
+using PassThrough = ck::tensor_operation::element_wise::PassThrough;
+
+static constexpr ck::index_t NumDimSpatial = 2;
+static constexpr ck::index_t N             = 16;
+static constexpr ck::index_t K             = 32;
+static constexpr ck::index_t C             = 3;
+static constexpr ck::index_t Y             = 3;
+static constexpr ck::index_t X             = 3;
+static constexpr ck::index_t Hi            = 224;
+static constexpr ck::index_t Wi            = 224;
+static constexpr ck::index_t Ho            = 113;
+static constexpr ck::index_t Wo            = 113;
+
+struct SimpleDeviceMem
+{
+    SimpleDeviceMem() = delete;
+
+    SimpleDeviceMem(std::size_t mem_size) : p_mem_{}
+    {
+        (void)hipMalloc(static_cast<void**>(&p_mem_), mem_size);
+    }
+
+    void* GetDeviceBuffer() { return p_mem_; }
+
+    ~SimpleDeviceMem() { (void)hipFree(p_mem_); }
+
+    void* p_mem_;
+};
+
+int main(int argc, char* argv[])
+{
+    std::vector<ck::index_t> in_spatial_lengths{Hi, Wi};
+    std::vector<ck::index_t> filter_spatial_lengths{Y, X};
+    std::vector<ck::index_t> out_spatial_lengths{Ho, Wo};
+    std::vector<ck::index_t> filter_strides{2, 2};
+    std::vector<ck::index_t> filter_dilations{1, 1};
+    std::vector<ck::index_t> input_left_pads{2, 2};
+    std::vector<ck::index_t> input_right_pads{2, 2};
+
+    SimpleDeviceMem in(sizeof(InDataType) * N * Hi * Wi * C);
+    SimpleDeviceMem wei(sizeof(WeiDataType) * K * Y * X * C);
+    SimpleDeviceMem out(sizeof(OutDataType) * N * Ho * Wo * K);
+
+    using DeviceOp = ck::tensor_operation::device::DeviceConvFwd<NumDimSpatial,
+                                                                 InLayout,
+                                                                 WeiLayout,
+                                                                 OutLayout,
+                                                                 InDataType,
+                                                                 WeiDataType,
+                                                                 OutDataType,
+                                                                 PassThrough,
+                                                                 PassThrough,
+                                                                 PassThrough>;
+    // get device op instances
+    const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
+        DeviceOp>::GetInstances();
+
+    std::cout << "found " << op_ptrs.size() << " instances" << std::endl;
+
+    std::string best_op_name;
+    int best_op_id        = -1;
+    float best_avg_time   = std::numeric_limits<float>::max();
+    float best_gb_per_sec = 0;
+    float best_tflops     = 0;
+
+    // profile device operation instances
+    std::cout << "Run all instances and do timing" << std::endl;
+
+    for(int i = 0; i < op_ptrs.size(); ++i)
+    {
+        auto& op_ptr        = op_ptrs[i];
+        auto argument_ptr   = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        K,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        filter_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+        auto invoker_ptr    = op_ptr->MakeInvokerPointer();
+        std::string op_name = op_ptr->GetTypeString();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            float avg_time = invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, true});
+
+            std::size_t flop      = 2 * N * K * C * Ho * Wo * Y * X;
+            std::size_t num_bytes = sizeof(InDataType) * N * Hi * Wi * C +
+                                    sizeof(WeiDataType) * K * Y * X * C +
+                                    sizeof(OutDataType) * N * Ho * Wo * K;
+
+            float tflops     = static_cast<float>(flop) / 1.E9 / avg_time;
+            float gb_per_sec = num_bytes / 1.E6 / avg_time;
+
+            std::cout << "Perf: " << std::setw(10) << avg_time << " ms, " << tflops << " TFlops, "
+                      << gb_per_sec << " GB/s, " << op_name << std::endl;
+
+            if(tflops > best_tflops)
+            {
+                best_op_id      = i;
+                best_op_name    = op_name;
+                best_avg_time   = avg_time;
+                best_gb_per_sec = gb_per_sec;
+                best_tflops     = tflops;
+            }
+        }
+        else
+        {
+            std::cout << op_name << " does not support this problem" << std::endl;
+        }
+    }
+
+    std::cout << "Best Perf: " << std::setw(10) << best_avg_time << " ms, " << best_tflops
+              << " TFlops, " << best_gb_per_sec << " GB/s, " << best_op_name << std::endl;
+
+    // run the best intance
+    {
+        auto& op_ptr = op_ptrs[best_op_id];
+        std::cout << "Run the best instance without timing: " << op_ptr->GetTypeString()
+                  << std::endl;
+        auto argument_ptr = op_ptr->MakeArgumentPointer(in.GetDeviceBuffer(),
+                                                        wei.GetDeviceBuffer(),
+                                                        out.GetDeviceBuffer(),
+                                                        N,
+                                                        K,
+                                                        C,
+                                                        in_spatial_lengths,
+                                                        filter_spatial_lengths,
+                                                        out_spatial_lengths,
+                                                        filter_strides,
+                                                        filter_dilations,
+                                                        input_left_pads,
+                                                        input_right_pads,
+                                                        PassThrough{},
+                                                        PassThrough{},
+                                                        PassThrough{});
+
+        auto invoker_ptr = op_ptr->MakeInvokerPointer();
+
+        if(op_ptr->IsSupportedArgument(argument_ptr.get()))
+        {
+            invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, false});
+        }
+
+        std::cout << "Done" << std::endl;
+    }
+
+    return 0;
+}
\ No newline at end of file

example/01_gemm/gemm_dl_fp16.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 
 using ADataType   = ck::half_t;
 using BDataType   = ck::half_t;

example/01_gemm/gemm_dl_fp32.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 
 using ADataType   = float;
 using BDataType   = float;

example/01_gemm/gemm_dl_int4.cpp

@@ -7,7 +7,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 
 using ADataType       = ck::int4_t;
 using BDataType       = ck::int4_t;

example/01_gemm/gemm_dl_int8.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_dl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_dl.hpp"
 
 using ADataType   = int8_t;
 using BDataType   = int8_t;

example/01_gemm/gemm_xdl_bf16.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 
 using ADataType        = ck::bhalf_t;
 using BDataType        = ck::bhalf_t;

example/01_gemm/gemm_xdl_fp64.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
 
 using ADataType   = double;
 using BDataType   = double;

example/01_gemm/gemm_xdl_int4.cpp

@@ -7,7 +7,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 
 using ADataType        = ck::int4_t;
 using BDataType        = ck::int4_t;

example/01_gemm/gemm_xdl_int8.cpp

@@ -3,7 +3,7 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_cshuffle.hpp"
 
 using ADataType        = int8_t;
 using BDataType        = int8_t;

example/01_gemm/gemm_xdl_skip_b_lds_fp16.cpp

@@ -3,8 +3,8 @@
 
 #include "common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_xdl_skip_b_lds.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_xdl_skip_b_lds.hpp"
 
 using F16 = ck::half_t;
 using F32 = float;

example/02_gemm_bilinear/gemm_bilinear_xdl_fp16.cpp

@@ -8,7 +8,7 @@
 
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/device_memory.hpp"

example/03_gemm_bias_relu/gemm_bias_relu_xdl_fp16.cpp

@@ -9,7 +9,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 
 #include "ck/library/utility/device_memory.hpp"

example/04_gemm_add_add_fastgelu/common.hpp

@@ -12,7 +12,7 @@
 #include "ck/ck.hpp"
 #include "ck/tensor_operation/gpu/device/tensor_layout.hpp"
 #include "ck/tensor_operation/gpu/device/gemm_specialization.hpp"
-#include "ck/tensor_operation/gpu/device/device_gemm_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_gemm_multiple_d_xdl_cshuffle.hpp"
 #include "ck/tensor_operation/gpu/element/element_wise_operation.hpp"
 #include "ck/utility/data_type.hpp"
 

example/09_convnd_fwd/convnd_fwd_xdl_bf16.cpp

@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 

example/09_convnd_fwd/convnd_fwd_xdl_fp16.cpp

@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"
 

example/09_convnd_fwd/convnd_fwd_xdl_fp32.cpp

@@ -3,7 +3,7 @@
 
 #include "convnd_fwd_common.hpp"
 
-#include "ck/tensor_operation/gpu/device/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
+#include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_fwd_multiple_d_xdl_cshuffle.hpp"
 
 #include "ck/library/utility/convolution_host_tensor_descriptor_helper.hpp"