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
206 lines
10 KiB
C++
206 lines
10 KiB
C++
// Copyright © Advanced Micro Devices, Inc., or its affiliates.
|
|
// SPDX-License-Identifier: MIT
|
|
|
|
#pragma once
|
|
|
|
#include "RMSNormTestCase.hpp"
|
|
|
|
namespace gpu_rmsnorm_ref_test
|
|
{
|
|
|
|
using hipdnn_data_sdk::utilities::TensorLayout;
|
|
|
|
// ============================================================================
|
|
// Small test cases for quick validation
|
|
// ============================================================================
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormSmall4DTestCases()
|
|
{
|
|
return {{{2, 3, 4, 4}, {1, 3, 4, 4}, TensorLayout::NCHW},
|
|
{{32, 3, 8, 8}, {1, 1, 8, 8}, TensorLayout::NCHW},
|
|
{{1, 3, 14, 14}, {1, 1, 1, 14}, TensorLayout::NCHW},
|
|
{{16, 32, 24, 16}, {1, 32, 24, 16}, TensorLayout::NCHW},
|
|
{{16, 40, 48, 32}, {1, 1, 48, 32}, TensorLayout::NCHW},
|
|
|
|
{{2, 3, 4, 4}, {1, 3, 4, 4}, TensorLayout::NHWC},
|
|
{{32, 3, 8, 8}, {1, 1, 8, 8}, TensorLayout::NHWC},
|
|
{{1, 3, 14, 14}, {1, 1, 1, 14}, TensorLayout::NHWC},
|
|
{{16, 32, 24, 16}, {1, 32, 24, 16}, TensorLayout::NHWC},
|
|
{{16, 40, 48, 32}, {1, 1, 48, 32}, TensorLayout::NHWC}};
|
|
}
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormSmall5DTestCases()
|
|
{
|
|
return {{{2, 3, 3, 1, 1}, {1, 3, 3, 1, 1}, TensorLayout::NCDHW},
|
|
{{2, 3, 4, 2, 2}, {1, 1, 4, 2, 2}, TensorLayout::NCDHW},
|
|
{{2, 3, 4, 2, 2}, {1, 1, 1, 2, 2}, TensorLayout::NCDHW},
|
|
{{2, 3, 4, 2, 2}, {1, 1, 1, 1, 2}, TensorLayout::NCDHW},
|
|
{{4, 8, 2, 4, 4}, {1, 8, 2, 4, 4}, TensorLayout::NCDHW},
|
|
|
|
{{2, 3, 3, 1, 1}, {1, 3, 3, 1, 1}, TensorLayout::NDHWC},
|
|
{{2, 3, 4, 2, 2}, {1, 1, 4, 2, 2}, TensorLayout::NDHWC},
|
|
{{2, 3, 4, 2, 2}, {1, 1, 1, 2, 2}, TensorLayout::NDHWC},
|
|
{{2, 3, 4, 2, 2}, {1, 1, 1, 1, 2}, TensorLayout::NDHWC},
|
|
{{4, 8, 2, 4, 4}, {1, 8, 2, 4, 4}, TensorLayout::NDHWC}};
|
|
}
|
|
|
|
// ============================================================================
|
|
// Medium test cases for standard validation
|
|
// ============================================================================
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormMedium4DTestCases()
|
|
{
|
|
return {{{1, 3, 14, 14}, {1, 3, 14, 14}, TensorLayout::NCHW},
|
|
{{1, 3, 14, 14}, {1, 1, 14, 14}, TensorLayout::NCHW},
|
|
{{1, 3, 14, 14}, {1, 1, 1, 14}, TensorLayout::NCHW},
|
|
{{1, 256, 1, 1}, {1, 256, 1, 1}, TensorLayout::NCHW},
|
|
{{2, 3, 1, 1}, {1, 3, 1, 1}, TensorLayout::NCHW},
|
|
{{32, 1, 14, 14}, {1, 1, 14, 14}, TensorLayout::NCHW},
|
|
{{32, 3, 1, 14}, {1, 3, 1, 14}, TensorLayout::NCHW},
|
|
{{32, 3, 14, 1}, {1, 3, 14, 1}, TensorLayout::NCHW},
|
|
{{32, 3, 14, 1}, {1, 1, 14, 1}, TensorLayout::NCHW},
|
|
{{16, 32, 192, 128}, {1, 32, 192, 128}, TensorLayout::NCHW},
|
|
{{16, 32, 192, 128}, {1, 1, 192, 128}, TensorLayout::NCHW},
|
|
{{16, 64, 225, 225}, {1, 64, 225, 225}, TensorLayout::NCHW},
|
|
{{16, 64, 225, 225}, {1, 1, 1, 225}, TensorLayout::NCHW},
|
|
{{16, 128, 56, 56}, {1, 128, 56, 56}, TensorLayout::NCHW},
|
|
{{16, 128, 56, 56}, {1, 1, 56, 56}, TensorLayout::NCHW},
|
|
|
|
{{1, 3, 14, 14}, {1, 3, 14, 14}, TensorLayout::NHWC},
|
|
{{1, 3, 14, 14}, {1, 1, 14, 14}, TensorLayout::NHWC},
|
|
{{1, 3, 14, 14}, {1, 1, 1, 14}, TensorLayout::NHWC},
|
|
{{1, 256, 1, 1}, {1, 256, 1, 1}, TensorLayout::NHWC},
|
|
{{2, 3, 1, 1}, {1, 3, 1, 1}, TensorLayout::NHWC},
|
|
{{32, 1, 14, 14}, {1, 1, 14, 14}, TensorLayout::NHWC},
|
|
{{32, 3, 1, 14}, {1, 3, 1, 14}, TensorLayout::NHWC},
|
|
{{32, 3, 14, 1}, {1, 3, 14, 1}, TensorLayout::NHWC},
|
|
{{32, 3, 14, 1}, {1, 1, 14, 1}, TensorLayout::NHWC},
|
|
{{16, 32, 192, 128}, {1, 32, 192, 128}, TensorLayout::NHWC},
|
|
{{16, 32, 192, 128}, {1, 1, 192, 128}, TensorLayout::NHWC},
|
|
{{16, 64, 225, 225}, {1, 64, 225, 225}, TensorLayout::NHWC},
|
|
{{16, 64, 225, 225}, {1, 1, 1, 225}, TensorLayout::NHWC},
|
|
{{16, 128, 56, 56}, {1, 128, 56, 56}, TensorLayout::NHWC},
|
|
{{16, 128, 56, 56}, {1, 1, 56, 56}, TensorLayout::NHWC}};
|
|
}
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormMedium5DTestCases()
|
|
{
|
|
return {{{16, 3, 8, 14, 14}, {1, 3, 8, 14, 14}, TensorLayout::NCDHW},
|
|
{{16, 32, 4, 48, 32}, {1, 32, 4, 48, 32}, TensorLayout::NCDHW},
|
|
{{16, 32, 4, 48, 32}, {1, 1, 1, 48, 32}, TensorLayout::NCDHW},
|
|
{{8, 64, 4, 28, 28}, {1, 64, 4, 28, 28}, TensorLayout::NCDHW},
|
|
{{8, 64, 4, 28, 28}, {1, 1, 4, 28, 28}, TensorLayout::NCDHW},
|
|
|
|
{{16, 3, 8, 14, 14}, {1, 3, 8, 14, 14}, TensorLayout::NDHWC},
|
|
{{16, 32, 4, 48, 32}, {1, 32, 4, 48, 32}, TensorLayout::NDHWC},
|
|
{{16, 32, 4, 48, 32}, {1, 1, 1, 48, 32}, TensorLayout::NDHWC},
|
|
{{8, 64, 4, 28, 28}, {1, 64, 4, 28, 28}, TensorLayout::NDHWC},
|
|
{{8, 64, 4, 28, 28}, {1, 1, 4, 28, 28}, TensorLayout::NDHWC}};
|
|
}
|
|
|
|
// ============================================================================
|
|
// Large test cases for comprehensive validation
|
|
// ============================================================================
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormLarge4DTestCases()
|
|
{
|
|
return {{{16, 288, 48, 32}, {1, 288, 48, 32}, TensorLayout::NCHW},
|
|
{{16, 288, 48, 32}, {1, 1, 48, 32}, TensorLayout::NCHW},
|
|
{{16, 576, 1, 30}, {1, 576, 1, 30}, TensorLayout::NCHW},
|
|
{{16, 576, 1, 30}, {1, 1, 1, 30}, TensorLayout::NCHW},
|
|
{{16, 2048, 16, 32}, {1, 2048, 16, 32}, TensorLayout::NCHW},
|
|
{{16, 2048, 16, 32}, {1, 1, 16, 32}, TensorLayout::NCHW},
|
|
{{128, 35, 48, 32}, {1, 35, 48, 32}, TensorLayout::NCHW},
|
|
{{128, 512, 24, 48}, {1, 512, 24, 48}, TensorLayout::NCHW},
|
|
{{128, 512, 24, 48}, {1, 1, 24, 48}, TensorLayout::NCHW},
|
|
|
|
{{16, 288, 48, 32}, {1, 288, 48, 32}, TensorLayout::NHWC},
|
|
{{16, 288, 48, 32}, {1, 1, 48, 32}, TensorLayout::NHWC},
|
|
{{16, 576, 1, 30}, {1, 576, 1, 30}, TensorLayout::NHWC},
|
|
{{16, 576, 1, 30}, {1, 1, 1, 30}, TensorLayout::NHWC},
|
|
{{16, 2048, 16, 32}, {1, 2048, 16, 32}, TensorLayout::NHWC},
|
|
{{16, 2048, 16, 32}, {1, 1, 16, 32}, TensorLayout::NHWC},
|
|
{{128, 35, 48, 32}, {1, 35, 48, 32}, TensorLayout::NHWC},
|
|
{{128, 512, 24, 48}, {1, 512, 24, 48}, TensorLayout::NHWC},
|
|
{{128, 512, 24, 48}, {1, 1, 24, 48}, TensorLayout::NHWC}};
|
|
}
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormLarge5DTestCases()
|
|
{
|
|
return {{{16, 128, 8, 24, 16}, {1, 128, 8, 24, 16}, TensorLayout::NCDHW},
|
|
{{16, 256, 4, 32, 32}, {1, 256, 4, 32, 32}, TensorLayout::NCDHW},
|
|
{{16, 256, 4, 32, 32}, {1, 1, 4, 32, 32}, TensorLayout::NCDHW},
|
|
{{32, 128, 8, 16, 16}, {1, 128, 8, 16, 16}, TensorLayout::NCDHW},
|
|
{{32, 128, 8, 16, 16}, {1, 1, 1, 16, 16}, TensorLayout::NCDHW},
|
|
|
|
{{16, 128, 8, 24, 16}, {1, 128, 8, 24, 16}, TensorLayout::NDHWC},
|
|
{{16, 256, 4, 32, 32}, {1, 256, 4, 32, 32}, TensorLayout::NDHWC},
|
|
{{16, 256, 4, 32, 32}, {1, 1, 4, 32, 32}, TensorLayout::NDHWC},
|
|
{{32, 128, 8, 16, 16}, {1, 128, 8, 16, 16}, TensorLayout::NDHWC},
|
|
{{32, 128, 8, 16, 16}, {1, 1, 1, 16, 16}, TensorLayout::NDHWC}};
|
|
}
|
|
|
|
// ============================================================================
|
|
// Edge-case shapes to isolate innerSize and outerSize growth.
|
|
// Only used with the DISABLED_TestGpuRMSNormFwdRefEdgeCaseValidation fixture.
|
|
// ============================================================================
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormSkinnyModerateTestCases()
|
|
{
|
|
// Matches the large scale dims used in getRMSnormLarge4DTestCases
|
|
constexpr int64_t N = 2048;
|
|
|
|
return {// 1 thread block, all work in one grid-stride loop
|
|
{{1, 1, 1, N}, {1, 1, 1, N}, TensorLayout::NCHW},
|
|
// Large thread blocks with a single active thread (degenerate reduction)
|
|
{{N, 1, 1, 1}, {1, 1, 1, 1}, TensorLayout::NCHW},
|
|
// Large thread blocks with two active threads (partial loop + two-thread reduction)
|
|
{{N, 1, 1, 2}, {1, 1, 1, 2}, TensorLayout::NCHW}};
|
|
}
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormSkinnyInt32ScaleTestCases(int64_t outerBound)
|
|
{
|
|
// NOTE: INNER_BOUND in these cases should be INT32_MAX, but is reduced here due to
|
|
// slow CPU fill/reference functions. Revisit once rocRAND-based GPU fill and
|
|
// golden references for large tensors are available.
|
|
constexpr int64_t INNER_BOUND = 100000000; // 100 million elements
|
|
|
|
// Cases are same as getRMSnormSkinnyModerateTestCases, but with innerSize and outerSize set to the int32_t boundary values
|
|
return {{{1, 1, 1, INNER_BOUND}, {1, 1, 1, INNER_BOUND}, TensorLayout::NCHW},
|
|
{{outerBound, 1, 1, 1}, {1, 1, 1, 1}, TensorLayout::NCHW},
|
|
{{outerBound, 1, 1, 2}, {1, 1, 1, 2}, TensorLayout::NCHW}};
|
|
}
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormPowerOfTwoTestCases()
|
|
{
|
|
// Outer size is fixed to small value to isolate innerSize's effect
|
|
constexpr int64_t OUTER = 2;
|
|
|
|
const std::vector<int64_t> innerSizes
|
|
= {1, 2, 3, 7, 8, 9, 15, 16, 17, 63, 64, 65,
|
|
127, 128, 129, 255, 256, 257, 511, 512, 513, 1023, 1024, 1025};
|
|
|
|
std::vector<RMSNormTestCase> cases;
|
|
cases.reserve(innerSizes.size());
|
|
for(int64_t inner : innerSizes)
|
|
{
|
|
cases.push_back({{OUTER, 1, 1, inner}, {1, 1, 1, inner}, TensorLayout::NCHW});
|
|
}
|
|
return cases;
|
|
}
|
|
|
|
inline std::vector<RMSNormTestCase> getRMSnormInnerSizeInt32BoundaryTestCases()
|
|
{
|
|
// Unguarded innerSize boundary test to verify int32 truncation is handled correctly
|
|
// NOTE: INNER_BOUND in these cases should be INT32_MAX, but is reduced here due to
|
|
// slow CPU fill/reference functions. Revisit once rocRAND-based GPU fill and
|
|
// golden references for large tensors are available.
|
|
constexpr int64_t INNER_BOUND = 100000000; // 100 million elements
|
|
|
|
return {{{1, 1, 1, INNER_BOUND - 1}, {1, 1, 1, INNER_BOUND - 1}, TensorLayout::NCHW},
|
|
{{1, 1, 1, INNER_BOUND}, {1, 1, 1, INNER_BOUND}, TensorLayout::NCHW},
|
|
{{1, 1, 1, INNER_BOUND + 1}, {1, 1, 1, INNER_BOUND + 1}, TensorLayout::NCHW}};
|
|
}
|
|
|
|
} // namespace gpu_rmsnorm_ref_test
|