Narrowing the scope of PR to OCP FP8 enablement only

example/35_splitK_gemm/splitK_gemm_xdl_fp16_fp8.cpp

@@ -16,7 +16,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"
 #include "ck/library/utility/literals.hpp"
 
 template <ck::index_t... Is>

example/35_splitK_gemm/splitK_gemm_xdl_fp32.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -16,7 +16,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"
 #include "ck/library/utility/literals.hpp"
 
 template <ck::index_t... Is>

example/35_splitK_gemm/splitK_gemm_xdl_int4.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -16,7 +16,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"
 #include "ck/library/utility/literals.hpp"
 
 template <ck::index_t... Is>

example/35_splitK_gemm/splitK_gemm_xdl_int8.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include <iostream>
 #include <numeric>
@@ -16,7 +16,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"
 #include "ck/library/utility/literals.hpp"
 
 template <ck::index_t... Is>

example/35_splitK_gemm/splitK_gemm_xdl_lds_direct_load_fp16.cpp

@@ -24,7 +24,6 @@
 #include "ck/library/utility/host_tensor.hpp"
 #include "ck/library/utility/host_tensor_generator.hpp"
 #include "ck/library/reference_tensor_operation/cpu/reference_gemm.hpp"
-#include "ck/library/reference_tensor_operation/gpu/reference_gemm.hpp"
 #include "ck/library/utility/literals.hpp"
 
 template <ck::index_t... Is>

example/38_grouped_conv_bwd_data_multiple_d/common.hpp

@@ -40,7 +40,7 @@ using BF8  = ck::bf8_t;
 struct ExecutionConfig final
 {
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 };
 

example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_wmma_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2023-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_wmma_cshuffle.hpp"
 #include "common.hpp"
@@ -11,7 +11,6 @@ using AccDataType      = FP32;
 using CShuffleDataType = FP16;
 using DsDataType       = ck::Tuple<>;
 using InDataType       = FP16;
-using VerifyDataType   = FP16; // is used for selection of check tolerances
 
 using OutLayout = ck::tensor_layout::convolution::GNHWK;
 using WeiLayout = ck::tensor_layout::convolution::GKYXC;

example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
 #include "common.hpp"
@@ -10,7 +10,6 @@ using AccDataType      = FP32;
 using CShuffleDataType = FP16;
 using DsDataType       = ck::Tuple<>;
 using InDataType       = FP16;
-using VerifyDataType   = FP16; // is used for selection of check tolerances
 
 using OutLayout = ck::tensor_layout::convolution::GNHWK;
 using WeiLayout = ck::tensor_layout::convolution::GKYXC;

example/38_grouped_conv_bwd_data_multiple_d/grouped_conv_bwd_data_xdl_fp16_comp_bf8_fp8.cpp

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 #include "ck/tensor_operation/gpu/device/impl/device_grouped_conv_bwd_data_multiple_d_xdl_cshuffle_v1.hpp"
 #include "common.hpp"
@@ -12,7 +12,6 @@ using DsDataType       = ck::Tuple<>;
 using InDataType       = FP16;
 using AComputeType     = BF8;
 using BComputeType     = FP8;
-using VerifyDataType   = BF8; // is used for selection of check tolerances
 
 using OutLayout = ck::tensor_layout::convolution::GNHWK;
 using WeiLayout = ck::tensor_layout::convolution::GKYXC;

example/38_grouped_conv_bwd_data_multiple_d/run_grouped_conv_bwd_data_example.inc

 // SPDX-License-Identifier: MIT
-// Copyright (c) 2018-2024, Advanced Micro Devices, Inc. All rights reserved.
-template <typename DataType>
-inline __host__ __device__ constexpr double get_rtol()
-{
-    if constexpr(std::is_same_v<DataType, float>)
-    {
-        return 1e-3;
-    }
-    else if constexpr(std::is_same_v<DataType, double>)
-    {
-        return 1e-6;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::half_t>)
-    {
-        return 1e-3;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::bhalf_t>)
-    {
-        return 5e-2;
-    }
-    else if constexpr(std::is_same_v<DataType, int32_t>)
-    {
-        return 1e-1;
-    }
-    else if constexpr(std::is_same_v<DataType, int8_t>)
-    {
-        return 1e-1;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::f8_t>)
-    {
-        return 1;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::bf8_t>)
-    {
-        return 1;
-    }
-    else
-    {
-        return 1e-3;
-    }
-}
-
-template <typename DataType>
-inline __host__ __device__ constexpr double get_atol()
-{
-    if constexpr(std::is_same_v<DataType, float>)
-    {
-        return 1e-3;
-    }
-    else if constexpr(std::is_same_v<DataType, double>)
-    {
-        return 1e-6;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::half_t>)
-    {
-        return 1e-3;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::bhalf_t>)
-    {
-        return 5e-2;
-    }
-    else if constexpr(std::is_same_v<DataType, int32_t>)
-    {
-        return 1e-1;
-    }
-    else if constexpr(std::is_same_v<DataType, int8_t>)
-    {
-        return 1e-1;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::f8_t>)
-    {
-        return 1;
-    }
-    else if constexpr(std::is_same_v<DataType, ck::bf8_t>)
-    {
-        return 1;
-    }
-    else
-    {
-        return 1e-3;
-    }
-}
+// Copyright (c) 2018-2023, Advanced Micro Devices, Inc. All rights reserved.
 
 bool run_conv_bwd_data(const ExecutionConfig& config,
                        const ck::utils::conv::ConvParam& conv_params,
@@ -108,7 +27,7 @@ bool run_conv_bwd_data(const ExecutionConfig& config,
         wei.GenerateTensorValue(GeneratorTensor_2<WeiDataType>{-5, 5});
         break;
     default:
-        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{-0.5, 0.5});
+        out.GenerateTensorValue(GeneratorTensor_3<OutDataType>{0.0, 1.0});
         wei.GenerateTensorValue(GeneratorTensor_3<WeiDataType>{-0.5, 0.5});
     }
 
@@ -219,11 +138,7 @@ bool run_conv_bwd_data(const ExecutionConfig& config,
 
         in_device_buf.FromDevice(in_device.mData.data());
 
-        return ck::utils::check_err(in_device.mData,
-                                    in_host.mData,
-                                    "Error: Incorrect results!",
-                                    get_rtol<VerifyDataType>(),
-                                    get_atol<VerifyDataType>());
+        return ck::utils::check_err(in_device.mData, in_host.mData);
     }
 
     return true;

example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_xdl_fixed_nk_bias_bf16_i8.cpp

@@ -194,9 +194,9 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
             b1_tensors[i].GenerateTensorValue(GeneratorTensor_3<B1DataType>{-0.5, 0.5});
             break;
         default:
-            a0_tensors[i].GenerateTensorValue(GeneratorTensor_1<A0DataType>{1});
-            b0_tensors[i].GenerateTensorValue(GeneratorTensor_1<B0DataType>{1});
-            b1_tensors[i].GenerateTensorValue(GeneratorTensor_1<B1DataType>{1});
+            a0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<A0DataType, 0>{});
+            b0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<B0DataType, 1>{});
+            b1_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<B1DataType, 1>{});
         }
 
         d0_tensors[i].GenerateTensorValue(GeneratorTensor_3<D0DataType>{-0.5, 0.5});

example/59_grouped_gemm_multi_ABD/grouped_gemm_multi_abd_xdl_fixed_nk_bias_fp16.cpp

@@ -184,9 +184,9 @@ bool run_grouped_gemm(const ProblemSize& problem_size, const ExecutionConfig& co
             b_tensors[i].GenerateTensorValue(GeneratorTensor_3<B0DataType>{-0.5, 0.5});
             break;
         default:
-            a0_tensors[i].GenerateTensorValue(GeneratorTensor_1<A0DataType>{1});
-            a1_tensors[i].GenerateTensorValue(GeneratorTensor_1<A1DataType>{1});
-            b_tensors[i].GenerateTensorValue(GeneratorTensor_1<B0DataType>{-1});
+            a0_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<A0DataType, 0>{});
+            a1_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<A1DataType, 0>{});
+            b_tensors[i].GenerateTensorValue(GeneratorTensor_Sequential<B0DataType, 1>{});
         }
 
         d0_tensors[i].GenerateTensorValue(GeneratorTensor_3<D0DataType>{-0.5, 0.5});

example/62_convnd_activ/convinvscale/run_convnd_fwd_convinvscale_example.inc

@@ -8,7 +8,7 @@ bool run_convnd_fwd_example(int argc, char* argv[])
     print_helper_msg();
 
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     ck::utils::conv::ConvParam conv_param{

example/62_convnd_activ/convscale/run_convnd_fwd_convscale_example.inc

@@ -8,7 +8,7 @@ bool run_convnd_fwd_example(int argc, char* argv[])
     print_helper_msg();
 
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     ck::utils::conv::ConvParam conv_param{

example/62_convnd_activ/convscale_add/run_convnd_fwd_convscale_add_example.inc

@@ -8,7 +8,7 @@ bool run_convnd_fwd_example(int argc, char* argv[])
     print_helper_msg();
 
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     ck::utils::conv::ConvParam conv_param{

example/62_convnd_activ/convscale_reduce/run_convnd_fwd_example.inc

@@ -8,7 +8,7 @@ bool run_convnd_fwd_example(int argc, char* argv[])
     print_helper_msg();
 
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     ck::utils::conv::ConvParam conv_param{

example/62_convnd_activ/convscale_relu/run_convnd_fwd_convscale_relu_example.inc

@@ -8,7 +8,7 @@ bool run_convnd_fwd_example(int argc, char* argv[])
     print_helper_msg();
 
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     ck::utils::conv::ConvParam conv_param{

example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8.cpp

@@ -86,7 +86,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_Xdl_CShu
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
-    int init_method      = 2;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     // GEMM shape

example/65_gemm_multiply_multiply/gemm_multiply_multiply_xdl_fp8_ab_scale.cpp

@@ -78,7 +78,7 @@ using DeviceOpInstance = ck::tensor_operation::device::DeviceGemmMultiD_ABScale_
 int main(int argc, char* argv[])
 {
     bool do_verification = true;
-    int init_method      = 5;
+    int init_method      = 1;
     bool time_kernel     = false;
 
     // GEMM shape
@@ -186,20 +186,6 @@ int main(int argc, char* argv[])
         b0_k_n.GenerateTensorValue(GeneratorTensor_1<B0DataType>{});
         a1_m_k.GenerateTensorValue(GeneratorTensor_3<A1DataType>{0, 1.0});
         b1_k_n.GenerateTensorValue(GeneratorTensor_3<B1DataType>{0, 1.0});
-        break;
-    case 6:
-        a0_m_k.GenerateTensorValue(GeneratorTensor_PI<A0DataType>{});
-        b0_k_n.GenerateTensorValue(GeneratorTensor_1<B0DataType>{0.5f});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{0.5});
-        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{4});
-
-        break;
-    case 7:
-        a0_m_k.GenerateTensorValue(GeneratorTensor_PI_A<A0DataType>{});
-        b0_k_n.GenerateTensorValue(GeneratorTensor_PI_B<B0DataType>{});
-        a1_m_k.GenerateTensorValue(GeneratorTensor_1<A1DataType>{2});
-        b1_k_n.GenerateTensorValue(GeneratorTensor_1<B1DataType>{0.5});
-
         break;
     default:
         a0_m_k.GenerateTensorValue(GeneratorTensor_3<A0DataType>{-0.5, 0.5});
@@ -254,28 +240,23 @@ int main(int argc, char* argv[])
             "not support this GEMM problem");
     }
 
-    std::cout << "Compute GEMM on device...  \n";
     float ave_time = invoker.Run(argument, StreamConfig{nullptr, time_kernel, 20, 50});
-    std::cout << "DONE!" << std::endl;
-    if(time_kernel)
-    {
-        std::size_t flop = std::size_t(2) * M * N * K;
-        std::size_t num_btype =
-            sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + sizeof(EDataType) * M * N;
 
-        float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
+    std::size_t flop = std::size_t(2) * M * N * K;
+    std::size_t num_btype =
+        sizeof(A0DataType) * M * K + sizeof(B0DataType) * K * N + sizeof(EDataType) * M * N;
 
-        float gb_per_sec = num_btype / 1.E6 / ave_time;
+    float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
 
-        std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec
-                  << " GB/s" << std::endl;
-    }
+    float gb_per_sec = num_btype / 1.E6 / ave_time;
+
+    std::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s"
+              << std::endl;
 
     e_device_buf.FromDevice(e_m_n_device_result.mData.data());
 
     if(do_verification)
     {
-        std::cout << "Running verification on CPU." << std::endl;
         Tensor<AccDataType> c_m_n({M, N});
         Tensor<float> a_m_k({M, K});
         Tensor<float> b_k_n({K, N});
@@ -325,28 +306,10 @@ int main(int argc, char* argv[])
 
         e_device_buf.FromDevice(e_m_n_device_result.mData.data());
 
-        if(init_method == 6 || init_method == 7)
-        {
-            std::cout << std::fixed << std::setprecision(16);
-
-            float d = ck::type_convert<float>(e_m_n_device_result(0, 10));
-            float h = ck::type_convert<float>(e_m_n_host_result(10, 0));
-            std::cout << "device result: " << d << std::endl;
-            std::cout << "host result: " << h << std::endl;
-            std::cout << "expected result: " << M_PI << std::endl;
-            std::cout << "device - host: " << std::abs(d - h) << std::endl;
-            std::cout << "device - expected: " << std::abs(d - M_PI) << std::endl;
-            std::cout << "atol: " << 5e-2 << std::endl;
-        }
-
-        if(ck::utils::check_err(
-               e_m_n_device_result, e_m_n_host_result, "Error: Incorrect results!", 5e-2, 5e-2))
-        {
-            std::cout << "Verification on CPU: PASS" << std::endl;
-            return 0;
-        }
-        else
-            return 1;
+        return ck::utils::check_err(
+                   e_m_n_device_result, e_m_n_host_result, "Error: Incorrect results!", 5e-2, 5e-2)
+                   ? 0
+                   : 1;
     }
 
     return 0;

include/ck/utility/amd_buffer_addressing.hpp

@@ -847,7 +847,7 @@ amd_buffer_load_invalid_element_return_zero(const T* p_src_wave,
 
     vector_t tmp{amd_buffer_load_impl<scalar_t, vector_size, coherence>(
         src_wave_buffer_resource, src_thread_addr_offset, 0)};
-    return src_thread_element_valid ? tmp : vector_t{0};
+    return src_thread_element_valid ? tmp : vector_t(0);
 #endif
 }