Mentions légales du service

Skip to content
Snippets Groups Projects

Introduce half-precision conversion and gemm kernels for GPUs

Merged Introduce half-precision conversion and gemm kernels for GPUs
faverge/chameleon:gpus/half_kernels into master
Merged Mathieu Faverge requested to merge faverge/chameleon:gpus/half_kernels into master
Compare
and
22 files
+ 967
− 49
Compare changes
  • Side-by-side
  • Inline
Files
22
cmake_modules/local_subs.py
+ 9
− 2
@@ -6,12 +6,12 @@
@copyright 2019-2023 Bordeaux INP, CNRS (LaBRI UMR 5800), Inria,
Univ. Bordeaux. All rights reserved.
@version 1.2.0
@version 1.3.0
@author Mathieu Faverge
@author Florent Pruvost
@author Nathalie Furmento
@author Alycia Lisito
@date 2022-02-22
@date 2023-07-04
"""
_extra_blas = [
@@ -39,6 +39,10 @@ _extra_blas = [
('', 'slatm1', 'dlatm1', 'slatm1', 'dlatm1' ),
('', 'sgenm2', 'dgenm2', 'cgenm2', 'zgenm2' ),
('', 'slag2c_fake', 'dlag2z_fake', 'slag2c', 'dlag2z' ),
('', 'slag2h', 'dlag2h', 'slag2h', 'dlag2h' ),
('', 'hlag2s', 'hlag2d', 'hlag2s', 'hlag2d' ),
('', 'slag2h', 'dlag2h', 'clag2x', 'zlag2x' ),
('', 'hlag2s', 'hlag2d', 'xlag2c', 'xlag2z' ),
('', 'sgepdf', 'dgepdf', 'cgepdf', 'zgepdf' ),
('', 'scesca', 'dcesca', 'ccesca', 'zcesca' ),
('', 'sgesum', 'dgesum', 'cgesum', 'zgesum' ),
@@ -73,6 +77,9 @@ subs = {
('int', 'float', 'double', 'CHAMELEON_Complex32_t', r'\bCHAMELEON_Complex64_t'),
('ChamPattern', 'ChamRealFloat', 'ChamRealDouble', 'ChamComplexFloat', r'\bChamComplexDouble' ),
('ChamPattern', 'ChamRealFloat', 'ChamRealDouble', 'ChamRealFloat', r'\bChamRealDouble' ),
('ChamPattern', 'ChamRealFloat', 'ChamRealFloat', 'ChamComplexFloat', r'\bChamComplexFloat' ),
('ChamPattern', 'ChamRealFloat', 'ChamRealFloat', 'ChamRealFloat', r'\bChamRealFloat' ),
('ChamPattern', 'ChamRealHalf', 'ChamRealHalf', 'ChamComplexHalf', r'\bChamComplexHalf' ),
('int', 'float', 'double', 'complex32', 'complex64' ),
('Int', 'Float', 'Double', 'Complex32', 'Complex64' ),
('Int', 'HMAT_SIMPLE_PRECISION','HMAT_DOUBLE_PRECISION','HMAT_SIMPLE_COMPLEX', 'HMAT_DOUBLE_COMPLEX'),