BF16 replacement kernels#705
Merged
Merged
Conversation
)" This reverts commit 402d231.
amcamd
previously approved these changes
Sep 18, 2019
Contributor
amcamd
left a comment
amcamd
left a comment
There was a problem hiding this comment.
Need to change back to D matrix for testing_gemm_ex.hpp and testing_gemm_ex_strided_batched in the future
bragadeesh
approved these changes
Sep 19, 2019
zaliu
added a commit
to zaliu/rocBLAS
that referenced
this pull request
Sep 24, 2019
zaliu
added a commit
that referenced
this pull request
Sep 24, 2019
* BF16 replacement kernels (#705) * Revert "Switch to using separate D for gemm_ex benchmark calls (#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (#707) * restore UseBeta=1 logic for arcturus BF16 TN
wbgilmartin
pushed a commit
to wbgilmartin/rocBLAS
that referenced
this pull request
Oct 9, 2019
* Changed timeout from hours to minutes (ROCm#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (ROCm#701) * SLES support (ROCm#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (ROCm#695) * Fixing Timeout * BF16 replacement kernels (ROCm#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (ROCm#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (ROCm#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (ROCm#708) * Batched syr (ROCm#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (ROCm#719) * Refactor Ger and Gemv (ROCm#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (ROCm#737) * SWDEV 203994 (ROCm#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (ROCm#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (ROCm#742)
wbgilmartin
pushed a commit
to wbgilmartin/rocBLAS
that referenced
this pull request
Oct 14, 2019
* Changed timeout from hours to minutes (ROCm#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (ROCm#701) * SLES support (ROCm#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (ROCm#695) * Fixing Timeout * BF16 replacement kernels (ROCm#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (ROCm#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (ROCm#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (ROCm#708) * Batched syr (ROCm#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (ROCm#719) * Refactor Ger and Gemv (ROCm#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (ROCm#737) * SWDEV 203994 (ROCm#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (ROCm#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (ROCm#742) * Fix GEMM for half type
wbgilmartin
pushed a commit
to wbgilmartin/rocBLAS
that referenced
this pull request
Oct 16, 2019
* Changed timeout from hours to minutes (ROCm#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (ROCm#701) * SLES support (ROCm#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (ROCm#695) * Fixing Timeout * BF16 replacement kernels (ROCm#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (ROCm#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (ROCm#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (ROCm#708) * Batched syr (ROCm#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (ROCm#719) * Refactor Ger and Gemv (ROCm#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (ROCm#737) * SWDEV 203994 (ROCm#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (ROCm#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (ROCm#742) * Fix GEMM for half type * Refactoring classes to be simpler * Fix rocblas_half
wbgilmartin
pushed a commit
to wbgilmartin/rocBLAS
that referenced
this pull request
Oct 17, 2019
* Changed timeout from hours to minutes (ROCm#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (ROCm#701) * SLES support (ROCm#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (ROCm#695) * Fixing Timeout * BF16 replacement kernels (ROCm#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (ROCm#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (ROCm#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (ROCm#708) * Batched syr (ROCm#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (ROCm#719) * Refactor Ger and Gemv (ROCm#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (ROCm#737) * SWDEV 203994 (ROCm#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (ROCm#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (ROCm#742) * Changing Gemv and Ger stride type (ROCm#747) * Handle spaces and newline (ROCm#748) * Fix GEMM for half type * Tuned Shakespeare kernels (ROCm#749) * Refactoring classes to be simpler * Fix rocblas_half
wbgilmartin
pushed a commit
to wbgilmartin/rocBLAS
that referenced
this pull request
Oct 21, 2019
* Changed timeout from hours to minutes (ROCm#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (ROCm#701) * SLES support (ROCm#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (ROCm#695) * Fixing Timeout * BF16 replacement kernels (ROCm#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (ROCm#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (ROCm#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (ROCm#708) * Batched syr (ROCm#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (ROCm#719) * Refactor Ger and Gemv (ROCm#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (ROCm#737) * SWDEV 203994 (ROCm#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (ROCm#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (ROCm#742) * Changing Gemv and Ger stride type (ROCm#747) * Handle spaces and newline (ROCm#748) * Fix GEMM for half type * Tuned Shakespeare kernels (ROCm#749) * Refactoring classes to be simpler * Fix rocblas_half * Cleanup source
wbgilmartin
added a commit
that referenced
this pull request
Oct 21, 2019
* integration of the new tensile client * New tensile client (#1) * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * Map values to value categories currently represented as double * minor changes * merge Bill & Lee's changes * Merge Bill & Lee's changes * Fix build errors * Merge 2.10 develop into new tensile client (#2) * Changed timeout from hours to minutes (#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (#701) * SLES support (#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (#695) * Fixing Timeout * BF16 replacement kernels (#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (#708) * Batched syr (#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (#719) * Refactor Ger and Gemv (#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (#737) * SWDEV 203994 (#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (#742) * fix for changes of FreeIndices change in Tensile * New tensile client (#3) * Changed timeout from hours to minutes (#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (#701) * SLES support (#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (#695) * Fixing Timeout * BF16 replacement kernels (#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (#708) * Batched syr (#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (#719) * Refactor Ger and Gemv (#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (#737) * SWDEV 203994 (#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (#742) * Fix GEMM for half type * updates to get rocblas-test and half sizes to work * partial fix for NaN test failures * New tensile client (#4) * Changed timeout from hours to minutes (#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (#701) * SLES support (#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (#695) * Fixing Timeout * BF16 replacement kernels (#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (#708) * Batched syr (#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (#719) * Refactor Ger and Gemv (#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (#737) * SWDEV 203994 (#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (#742) * Fix GEMM for half type * Refactoring classes to be simpler * Fix rocblas_half * fix negative workgroup mapping error * fix WorkGroupMapping issue for files in asm_lite * more fixes for workgroupmapping issue * wgm issue for asm_miopen * New tensile client (#5) * Changed timeout from hours to minutes (#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (#701) * SLES support (#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (#695) * Fixing Timeout * BF16 replacement kernels (#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (#708) * Batched syr (#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (#719) * Refactor Ger and Gemv (#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (#737) * SWDEV 203994 (#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (#742) * Changing Gemv and Ger stride type (#747) * Handle spaces and newline (#748) * Fix GEMM for half type * Tuned Shakespeare kernels (#749) * Refactoring classes to be simpler * Fix rocblas_half * fix argument validation in gemm calls * fix complex strided batch implementation * fix validateArgs redefinition * New tensile client (#6) * Changed timeout from hours to minutes (#699) * set clang include directory, fix for centos build error * hot fix to restore loading of DGEMM replacement kernels (#701) * SLES support (#704) * Merging master with SLES commit * Specifying GPU architecture for ubuntu and sles (#695) * Fixing Timeout * BF16 replacement kernels (#705) * hot fix to restore loading of DGEMM replacement kernels * Revert "Switch to using separate D for gemm_ex benchmark calls (#667)" This reverts commit 402d231. * bf16 kernels for gfx908 * use bf16 UseBeta=0 replacement kernels * update tensile_tag to use bf16 UseBeta=0 replacement kernels * Restore usebeta1 logic (#707) * restore UseBeta=1 logic for arcturus BF16 TN * Supporting clang10 for SLES (#708) * Batched syr (#727) * adding syr batched and strided batched * rocblas_stride, reusable template pattern work * WIP * fixes testing and format * restore dependency * adds minimal batches & bad arg * fix bad arg testing * constify ptrs, spelling * add alpha vector support, PR feedback * more BF16 TN sizes * Refactoring * Move tensile_host.hpp from public C API to private C++ implementation * Work around missing complex; fix formatting * Remove --lib option and argument; make use of legacy handle API instead of introducing a new host handle API * gf908 BF16 TN 512x512x512 known issue * Enable SLES packaging (#719) * Refactor Ger and Gemv (#735) * Map values to value categories currently represented as double * Rot(m)(g) batched and strided_batched (#737) * SWDEV 203994 (#743) * Refactor alpha, beta logging; do not return invalid pointer errors in GEMM, GEMV when other arguments allow early exit * Fixes * update * version for master branch release * version for develop branch release * update Tensile package number * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests (#741) * Fixing SLES tests LD_LIBRARY_PATH and refactoring tests * Missing braces/spelling * spelling * Fixing packaging * New Winograd kernels added (#742) * Changing Gemv and Ger stride type (#747) * Handle spaces and newline (#748) * Fix GEMM for half type * Tuned Shakespeare kernels (#749) * Refactoring classes to be simpler * Fix rocblas_half * Cleanup source
mlse-lib-jenkins
pushed a commit
that referenced
this pull request
Jun 1, 2021
* fix until hip is revised to not require static attribute
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Planned ROCm 2.9 Feature: BF16 replacement kernels.