derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
521 lines
21 KiB
C++
521 lines
21 KiB
C++
/* ************************************************************************
|
|
* Copyright (C) 2022-2026 Advanced Micro Devices, Inc. All rights Reserved.
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
* in the Software without restriction, including without limitation the rights
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
|
|
* copies of the Software, and to permit persons to whom the Software is
|
|
* furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in
|
|
* all copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
|
|
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
|
|
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
|
|
* THE SOFTWARE.
|
|
*
|
|
* ************************************************************************ */
|
|
|
|
#pragma once
|
|
#ifndef TESTING_SPMM_BATCHED_CSC_HPP
|
|
#define TESTING_SPMM_BATCHED_CSC_HPP
|
|
|
|
#include "display.hpp"
|
|
#include "flops.hpp"
|
|
#include "gbyte.hpp"
|
|
#include "hipsparse.hpp"
|
|
#include "hipsparse_arguments.hpp"
|
|
#include "hipsparse_graph.hpp"
|
|
#include "hipsparse_test_unique_ptr.hpp"
|
|
#include "unit.hpp"
|
|
#include "utility.hpp"
|
|
|
|
#include <hipsparse.h>
|
|
#include <string>
|
|
#include <typeinfo>
|
|
|
|
using namespace hipsparse;
|
|
using namespace hipsparse_test;
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spmm_batched_csc_bad_arg(const Arguments& argus)
|
|
{
|
|
#if(!defined(CUDART_VERSION))
|
|
int32_t m = 100;
|
|
int32_t n = 100;
|
|
int32_t k = 100;
|
|
int64_t nnz = 100;
|
|
float alpha = 0.6;
|
|
float beta = 0.2;
|
|
size_t safe_size = 100;
|
|
hipsparseOperation_t transA = HIPSPARSE_OPERATION_NON_TRANSPOSE;
|
|
hipsparseOperation_t transB = HIPSPARSE_OPERATION_NON_TRANSPOSE;
|
|
hipsparseOrder_t order = HIPSPARSE_ORDER_COL;
|
|
hipsparseIndexBase_t idxBase = HIPSPARSE_INDEX_BASE_ZERO;
|
|
hipsparseIndexType_t idxTypeI = HIPSPARSE_INDEX_64I;
|
|
hipsparseIndexType_t idxTypeJ = HIPSPARSE_INDEX_32I;
|
|
hipDataType dataType = HIP_R_32F;
|
|
|
|
#if(CUDART_VERSION >= 11003)
|
|
hipsparseSpMMAlg_t alg = HIPSPARSE_SPMM_CSR_ALG1;
|
|
#else
|
|
hipsparseSpMMAlg_t alg = HIPSPARSE_MM_ALG_DEFAULT;
|
|
#endif
|
|
|
|
hipsparseLocalHandle_t handle;
|
|
|
|
auto dptr_managed
|
|
= hipsparse_unique_ptr{device_malloc(sizeof(int64_t) * safe_size), device_free};
|
|
auto drow_managed
|
|
= hipsparse_unique_ptr{device_malloc(sizeof(int32_t) * safe_size), device_free};
|
|
auto dval_managed = hipsparse_unique_ptr{device_malloc(sizeof(float) * safe_size), device_free};
|
|
auto dB_managed = hipsparse_unique_ptr{device_malloc(sizeof(float) * safe_size), device_free};
|
|
auto dC_managed = hipsparse_unique_ptr{device_malloc(sizeof(float) * safe_size), device_free};
|
|
auto dbuf_managed = hipsparse_unique_ptr{device_malloc(sizeof(char) * safe_size), device_free};
|
|
|
|
int64_t* dptr = (int64_t*)dptr_managed.get();
|
|
int32_t* drow = (int32_t*)drow_managed.get();
|
|
float* dval = (float*)dval_managed.get();
|
|
float* dB = (float*)dB_managed.get();
|
|
float* dC = (float*)dC_managed.get();
|
|
void* dbuf = (void*)dbuf_managed.get();
|
|
|
|
// SpMM structures
|
|
hipsparseSpMatDescr_t A;
|
|
hipsparseDnMatDescr_t B, C;
|
|
|
|
// Create SpMM structures
|
|
verify_hipsparse_status_success(
|
|
hipsparseCreateCsc(&A, m, k, nnz, dptr, drow, dval, idxTypeI, idxTypeJ, idxBase, dataType),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseCreateDnMat(&B, k, n, k, dB, dataType, order),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseCreateDnMat(&C, m, n, m, dC, dataType, order),
|
|
"success");
|
|
|
|
int batch_count_A;
|
|
int batch_count_B;
|
|
int batch_count_C;
|
|
int64_t offsets_batch_stride_A;
|
|
int64_t rows_values_batch_stride_A;
|
|
int64_t batch_stride_B;
|
|
int64_t batch_stride_C;
|
|
|
|
// C_i = A * B_i
|
|
batch_count_A = 1;
|
|
batch_count_B = 10;
|
|
batch_count_C = 5;
|
|
offsets_batch_stride_A = 0;
|
|
rows_values_batch_stride_A = 0;
|
|
batch_stride_B = k * n;
|
|
batch_stride_C = m * n;
|
|
verify_hipsparse_status_success(
|
|
hipsparseCsrSetStridedBatch(
|
|
A, batch_count_A, offsets_batch_stride_A, rows_values_batch_stride_A),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseDnMatSetStridedBatch(B, batch_count_B, batch_stride_B),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseDnMatSetStridedBatch(C, batch_count_C, batch_stride_C),
|
|
"success");
|
|
|
|
verify_hipsparse_status_invalid_value(
|
|
hipsparseSpMM(handle, transA, transB, &alpha, A, B, &beta, C, dataType, alg, dbuf),
|
|
"Error: Combination of strided batch parameters is invald");
|
|
|
|
// C_i = A_i * B
|
|
batch_count_A = 10;
|
|
batch_count_B = 1;
|
|
batch_count_C = 5;
|
|
offsets_batch_stride_A = (k + 1);
|
|
rows_values_batch_stride_A = nnz;
|
|
batch_stride_B = 0;
|
|
batch_stride_C = m * n;
|
|
verify_hipsparse_status_success(
|
|
hipsparseCsrSetStridedBatch(
|
|
A, batch_count_A, offsets_batch_stride_A, rows_values_batch_stride_A),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseDnMatSetStridedBatch(B, batch_count_B, batch_stride_B),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseDnMatSetStridedBatch(C, batch_count_C, batch_stride_C),
|
|
"success");
|
|
|
|
verify_hipsparse_status_invalid_value(
|
|
hipsparseSpMM(handle, transA, transB, &alpha, A, B, &beta, C, dataType, alg, dbuf),
|
|
"Error: Combination of strided batch parameters is invald");
|
|
|
|
// C_i = A_i * B_i
|
|
batch_count_A = 10;
|
|
batch_count_B = 10;
|
|
batch_count_C = 5;
|
|
offsets_batch_stride_A = (k + 1);
|
|
rows_values_batch_stride_A = nnz;
|
|
batch_stride_B = k * n;
|
|
batch_stride_C = m * n;
|
|
verify_hipsparse_status_success(
|
|
hipsparseCsrSetStridedBatch(
|
|
A, batch_count_A, offsets_batch_stride_A, rows_values_batch_stride_A),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseDnMatSetStridedBatch(B, batch_count_B, batch_stride_B),
|
|
"success");
|
|
verify_hipsparse_status_success(hipsparseDnMatSetStridedBatch(C, batch_count_C, batch_stride_C),
|
|
"success");
|
|
|
|
verify_hipsparse_status_invalid_value(
|
|
hipsparseSpMM(handle, transA, transB, &alpha, A, B, &beta, C, dataType, alg, dbuf),
|
|
"Error: Combination of strided batch parameters is invald");
|
|
|
|
// Destruct
|
|
verify_hipsparse_status_success(hipsparseDestroySpMat(A), "success");
|
|
verify_hipsparse_status_success(hipsparseDestroyDnMat(B), "success");
|
|
verify_hipsparse_status_success(hipsparseDestroyDnMat(C), "success");
|
|
#endif
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spmm_batched_csc(Arguments argus)
|
|
{
|
|
#if(!defined(CUDART_VERSION))
|
|
J m = argus.M;
|
|
J n = argus.N;
|
|
J k = argus.K;
|
|
T h_alpha = argus.get_alpha<T>();
|
|
T h_beta = argus.get_beta<T>();
|
|
hipsparseOperation_t transA = argus.transA;
|
|
hipsparseOperation_t transB = argus.transB;
|
|
hipsparseOrder_t orderB = argus.orderB;
|
|
hipsparseOrder_t orderC = argus.orderC;
|
|
hipsparseIndexBase_t idx_base = argus.baseA;
|
|
|
|
J batch_count_A = 1;
|
|
J batch_count_B = 3;
|
|
J batch_count_C = 3;
|
|
|
|
#if(CUDART_VERSION >= 11003)
|
|
hipsparseSpMMAlg_t alg = HIPSPARSE_SPMM_CSR_ALG1;
|
|
#else
|
|
hipsparseSpMMAlg_t alg = HIPSPARSE_MM_ALG_DEFAULT;
|
|
#endif
|
|
|
|
std::string filename = argus.filename;
|
|
|
|
#if(defined(CUDART_VERSION))
|
|
if(orderB != orderC || orderB != HIPSPARSE_ORDER_COL)
|
|
{
|
|
return;
|
|
}
|
|
#endif
|
|
|
|
// Index and data type
|
|
hipsparseIndexType_t typeI = getIndexType<I>();
|
|
hipsparseIndexType_t typeJ = getIndexType<J>();
|
|
hipDataType typeT = getDataType<T>();
|
|
|
|
// hipSPARSE handle
|
|
hipsparseLocalHandle_t handle(argus);
|
|
|
|
// Host structures
|
|
std::vector<I> hcsc_col_ptr_temp;
|
|
std::vector<J> hcsc_row_ind_temp;
|
|
std::vector<T> hcsc_val_temp;
|
|
|
|
// Initial Data on CPU
|
|
srand(12345ULL);
|
|
|
|
I nnz_A;
|
|
CHECK_GENERATE_MATRIX_ERROR(
|
|
generate_csr_matrix(filename,
|
|
(transA == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? k : m,
|
|
(transA == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? m : k,
|
|
nnz_A,
|
|
hcsc_col_ptr_temp,
|
|
hcsc_row_ind_temp,
|
|
hcsc_val_temp,
|
|
idx_base));
|
|
|
|
// Redefine sparse matrix values
|
|
hipsparseInit<T>(hcsc_val_temp, hcsc_val_temp.size(), 1);
|
|
|
|
// Some matrix properties
|
|
J A_m = (transA == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? m : k;
|
|
J A_n = (transA == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? k : m;
|
|
J B_m = (transB == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? k : n;
|
|
J B_n = (transB == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? n : k;
|
|
J C_m = m;
|
|
J C_n = n;
|
|
|
|
int ld_multiplier_B = 1;
|
|
int ld_multiplier_C = 1;
|
|
|
|
int64_t ldb
|
|
= (orderB == HIPSPARSE_ORDER_COL)
|
|
? ((transB == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? (int64_t(ld_multiplier_B) * k)
|
|
: (int64_t(ld_multiplier_B) * n))
|
|
: ((transB == HIPSPARSE_OPERATION_NON_TRANSPOSE) ? (int64_t(ld_multiplier_B) * n)
|
|
: (int64_t(ld_multiplier_B) * k));
|
|
int64_t ldc = (orderC == HIPSPARSE_ORDER_COL) ? (int64_t(ld_multiplier_C) * m)
|
|
: (int64_t(ld_multiplier_C) * n);
|
|
|
|
ldb = std::max(int64_t(1), ldb);
|
|
ldc = std::max(int64_t(1), ldc);
|
|
|
|
int64_t nrowB = (orderB == HIPSPARSE_ORDER_COL) ? ldb : B_m;
|
|
int64_t ncolB = (orderB == HIPSPARSE_ORDER_COL) ? B_n : ldb;
|
|
int64_t nrowC = (orderC == HIPSPARSE_ORDER_COL) ? ldc : C_m;
|
|
int64_t ncolC = (orderC == HIPSPARSE_ORDER_COL) ? C_n : ldc;
|
|
|
|
int64_t nnz_B = nrowB * ncolB;
|
|
int64_t nnz_C = nrowC * ncolC;
|
|
|
|
int64_t offsets_batch_stride_A = (batch_count_A > 1) ? (A_n + 1) : 0;
|
|
int64_t rows_values_batch_stride_A = (batch_count_A > 1) ? nnz_A : 0;
|
|
int64_t batch_stride_B = (batch_count_B > 1) ? nnz_B : 0;
|
|
int64_t batch_stride_C = (batch_count_C > 1) ? nnz_C : 0;
|
|
|
|
// Allocate host memory for all batches of A matrix
|
|
std::vector<I> hcsc_col_ptr(batch_count_A * (A_n + 1));
|
|
std::vector<J> hcsc_row_ind(batch_count_A * nnz_A);
|
|
std::vector<T> hcsc_val(batch_count_A * nnz_A);
|
|
|
|
for(J i = 0; i < batch_count_A; i++)
|
|
{
|
|
for(J j = 0; j < (A_n + 1); j++)
|
|
{
|
|
hcsc_col_ptr[(A_n + 1) * i + j] = hcsc_col_ptr_temp[j];
|
|
}
|
|
|
|
for(I j = 0; j < nnz_A; j++)
|
|
{
|
|
hcsc_row_ind[nnz_A * i + j] = hcsc_row_ind_temp[j];
|
|
hcsc_val[nnz_A * i + j] = hcsc_val_temp[j];
|
|
}
|
|
}
|
|
|
|
// Allocate host memory for vectors
|
|
std::vector<T> hB(batch_count_B * nnz_B);
|
|
std::vector<T> hC_1(batch_count_C * nnz_C);
|
|
std::vector<T> hC_2(batch_count_C * nnz_C);
|
|
std::vector<T> hC_gold(batch_count_C * nnz_C);
|
|
|
|
hipsparseInit<T>(hB, batch_count_B * nnz_B, 1);
|
|
hipsparseInit<T>(hC_1, batch_count_C * nnz_C, 1);
|
|
|
|
// copy vector is easy in STL; hC_gold = hC: save a copy in hy_gold which will be output of CPU
|
|
hC_2 = hC_1;
|
|
hC_gold = hC_1;
|
|
|
|
// allocate memory on device
|
|
auto dptr_managed = hipsparse_unique_ptr{device_malloc(sizeof(I) * (A_n + 1)), device_free};
|
|
auto drow_managed = hipsparse_unique_ptr{device_malloc(sizeof(J) * nnz_A), device_free};
|
|
auto dval_managed = hipsparse_unique_ptr{device_malloc(sizeof(T) * nnz_A), device_free};
|
|
auto dB_managed
|
|
= hipsparse_unique_ptr{device_malloc(sizeof(T) * batch_count_B * nnz_B), device_free};
|
|
auto dC_1_managed
|
|
= hipsparse_unique_ptr{device_malloc(sizeof(T) * batch_count_C * nnz_C), device_free};
|
|
auto dC_2_managed
|
|
= hipsparse_unique_ptr{device_malloc(sizeof(T) * batch_count_C * nnz_C), device_free};
|
|
auto d_alpha_managed = hipsparse_unique_ptr{device_malloc(sizeof(T)), device_free};
|
|
auto d_beta_managed = hipsparse_unique_ptr{device_malloc(sizeof(T)), device_free};
|
|
|
|
I* dptr = (I*)dptr_managed.get();
|
|
J* drow = (J*)drow_managed.get();
|
|
T* dval = (T*)dval_managed.get();
|
|
T* dB = (T*)dB_managed.get();
|
|
T* dC_1 = (T*)dC_1_managed.get();
|
|
T* dC_2 = (T*)dC_2_managed.get();
|
|
T* d_alpha = (T*)d_alpha_managed.get();
|
|
T* d_beta = (T*)d_beta_managed.get();
|
|
|
|
// copy data from CPU to device
|
|
CHECK_HIP_ERROR(
|
|
hipMemcpy(dptr, hcsc_col_ptr.data(), sizeof(I) * (A_n + 1), hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(hipMemcpy(drow, hcsc_row_ind.data(), sizeof(J) * nnz_A, hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(hipMemcpy(dval, hcsc_val.data(), sizeof(T) * nnz_A, hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(
|
|
hipMemcpy(dB, hB.data(), sizeof(T) * batch_count_B * nnz_B, hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(
|
|
hipMemcpy(dC_1, hC_1.data(), sizeof(T) * batch_count_C * nnz_C, hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(
|
|
hipMemcpy(dC_2, hC_2.data(), sizeof(T) * batch_count_C * nnz_C, hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(hipMemcpy(d_alpha, &h_alpha, sizeof(T), hipMemcpyHostToDevice));
|
|
CHECK_HIP_ERROR(hipMemcpy(d_beta, &h_beta, sizeof(T), hipMemcpyHostToDevice));
|
|
|
|
// Create matrices
|
|
hipsparseSpMatDescr_t A;
|
|
CHECK_HIPSPARSE_ERROR(
|
|
hipsparseCreateCsc(&A, A_m, A_n, nnz_A, dptr, drow, dval, typeI, typeJ, idx_base, typeT));
|
|
|
|
// Create dense matrices
|
|
hipsparseDnMatDescr_t B, C1, C2;
|
|
CHECK_HIPSPARSE_ERROR(hipsparseCreateDnMat(&B, B_m, B_n, ldb, dB, typeT, orderB));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseCreateDnMat(&C1, C_m, C_n, ldc, dC_1, typeT, orderC));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseCreateDnMat(&C2, C_m, C_n, ldc, dC_2, typeT, orderC));
|
|
|
|
CHECK_HIPSPARSE_ERROR(hipsparseCsrSetStridedBatch(
|
|
A, batch_count_A, offsets_batch_stride_A, rows_values_batch_stride_A));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDnMatSetStridedBatch(B, batch_count_B, batch_stride_B));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDnMatSetStridedBatch(C1, batch_count_C, batch_stride_C));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDnMatSetStridedBatch(C2, batch_count_C, batch_stride_C));
|
|
|
|
// Query SpMM buffer
|
|
size_t bufferSize;
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM_bufferSize(
|
|
handle, transA, transB, &h_alpha, A, B, &h_beta, C1, typeT, alg, &bufferSize));
|
|
|
|
#if(!defined(CUDART_VERSION) || CUDART_VERSION >= 11021)
|
|
//When using cusparse backend, cant pass nullptr for buffer to preprocess
|
|
if(bufferSize == 0)
|
|
{
|
|
bufferSize = 4;
|
|
}
|
|
#endif
|
|
|
|
void* buffer;
|
|
CHECK_HIP_ERROR(hipMalloc(&buffer, bufferSize));
|
|
|
|
if(argus.call_preprocess)
|
|
{
|
|
// HIPSPARSE pointer mode host
|
|
#if(!defined(CUDART_VERSION) || CUDART_VERSION >= 11021)
|
|
CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_HOST));
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM_preprocess(
|
|
handle, transA, transB, &h_alpha, A, B, &h_beta, C1, typeT, alg, buffer));
|
|
#endif
|
|
|
|
// HIPSPARSE pointer mode device
|
|
#if(!defined(CUDART_VERSION) || CUDART_VERSION >= 11021)
|
|
CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_DEVICE));
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM_preprocess(
|
|
handle, transA, transB, d_alpha, A, B, d_beta, C2, typeT, alg, buffer));
|
|
#endif
|
|
}
|
|
|
|
if(argus.unit_check)
|
|
{
|
|
CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_HOST));
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM(
|
|
handle, transA, transB, &h_alpha, A, B, &h_beta, C1, typeT, alg, buffer));
|
|
|
|
CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_DEVICE));
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM(
|
|
handle, transA, transB, d_alpha, A, B, d_beta, C2, typeT, alg, buffer));
|
|
|
|
// copy output from device to CPU
|
|
CHECK_HIP_ERROR(
|
|
hipMemcpy(hC_1.data(), dC_1, sizeof(T) * batch_count_C * nnz_C, hipMemcpyDeviceToHost));
|
|
CHECK_HIP_ERROR(
|
|
hipMemcpy(hC_2.data(), dC_2, sizeof(T) * batch_count_C * nnz_C, hipMemcpyDeviceToHost));
|
|
|
|
// CPU
|
|
host_cscmm_batched(A_m,
|
|
n,
|
|
A_n,
|
|
batch_count_A,
|
|
(I)offsets_batch_stride_A,
|
|
(I)rows_values_batch_stride_A,
|
|
transA,
|
|
transB,
|
|
h_alpha,
|
|
hcsc_col_ptr.data(),
|
|
hcsc_row_ind.data(),
|
|
hcsc_val.data(),
|
|
hB.data(),
|
|
(J)ldb,
|
|
batch_count_B,
|
|
(I)batch_stride_B,
|
|
orderB,
|
|
h_beta,
|
|
hC_gold.data(),
|
|
(J)ldc,
|
|
batch_count_C,
|
|
(I)batch_stride_C,
|
|
orderC,
|
|
idx_base);
|
|
|
|
unit_check_near(1, batch_count_C * nnz_C, 1, hC_gold.data(), hC_1.data());
|
|
unit_check_near(1, batch_count_C * nnz_C, 1, hC_gold.data(), hC_2.data());
|
|
}
|
|
|
|
if(argus.timing)
|
|
{
|
|
int number_cold_calls = 2;
|
|
int number_hot_calls = argus.iters;
|
|
|
|
CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_HOST));
|
|
|
|
// Warm up
|
|
for(int iter = 0; iter < number_cold_calls; ++iter)
|
|
{
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM(
|
|
handle, transA, transB, &h_alpha, A, B, &h_beta, C1, typeT, alg, buffer));
|
|
}
|
|
|
|
double gpu_time_used = get_time_us();
|
|
|
|
// Performance run
|
|
for(int iter = 0; iter < number_hot_calls; ++iter)
|
|
{
|
|
CHECK_HIPSPARSE_ERROR(testing::hipsparseSpMM(
|
|
handle, transA, transB, &h_alpha, A, B, &h_beta, C1, typeT, alg, buffer));
|
|
}
|
|
|
|
gpu_time_used = (get_time_us() - gpu_time_used) / number_hot_calls;
|
|
|
|
double gflop_count
|
|
= batch_count_C
|
|
* spmm_gflop_count(n, nnz_A, (I)C_m * (I)C_n, h_beta != make_DataType<T>(0));
|
|
double gpu_gflops = get_gpu_gflops(gpu_time_used, gflop_count);
|
|
|
|
double gbyte_count = cscmm_batched_gbyte_count<T>(A_n,
|
|
nnz_A,
|
|
(I)B_m * (I)B_n,
|
|
(I)C_m * (I)C_n,
|
|
batch_count_A,
|
|
batch_count_B,
|
|
batch_count_C,
|
|
h_beta != make_DataType<T>(0));
|
|
double gpu_gbyte = get_gpu_gbyte(gpu_time_used, gbyte_count);
|
|
|
|
display_timing_info(display_key_t::M,
|
|
m,
|
|
display_key_t::N,
|
|
n,
|
|
display_key_t::K,
|
|
k,
|
|
display_key_t::nnzA,
|
|
nnz_A,
|
|
display_key_t::batch_countA,
|
|
batch_count_A,
|
|
display_key_t::batch_countB,
|
|
batch_count_B,
|
|
display_key_t::batch_countC,
|
|
batch_count_C,
|
|
display_key_t::alpha,
|
|
h_alpha,
|
|
display_key_t::beta,
|
|
h_beta,
|
|
display_key_t::algorithm,
|
|
hipsparse_spmmalg2string(alg),
|
|
display_key_t::gflops,
|
|
gpu_gflops,
|
|
display_key_t::bandwidth,
|
|
gpu_gbyte,
|
|
display_key_t::time_ms,
|
|
get_gpu_time_msec(gpu_time_used));
|
|
}
|
|
|
|
CHECK_HIP_ERROR(hipFree(buffer));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDestroySpMat(A));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDestroyDnMat(B));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDestroyDnMat(C1));
|
|
CHECK_HIPSPARSE_ERROR(hipsparseDestroyDnMat(C2));
|
|
|
|
#endif
|
|
}
|
|
|
|
#endif // TESTING_SPMM_BATCHED_CSC_HPP
|