The LLVM and GCC teams agreed to rename the __builtin_mma_assemble_pair and
__builtin_mma_disassemble_pair built-ins to __builtin_vsx_assemble_pair and
__builtin_vsx_disassemble_pair respectively. This patch is to make
corresponding changes in dgemm kernel. Also made changes in
inputs to those builtins to avoid some potential typecasting issues.
Reference gcc commit id:77ef995c1fbcab76a2a69b9f4700bcfd005d8e62
This patch aligns the stores to 32 byte boundary for scopy and dcopy
before entering into vector pair loop. For ccopy, changed the store
instructions to stxv to improve performance of unaligned cases.
This patch aligns the stores to 32 byte boundary for saxpy and daxpy
before entering into vector pair loop. Fox caxpy, changed the store
instructions to stxv to improve performance of unaligned cases.
This patch makes use of new POWER10 vector pair instructions for
loads and stores. Also reorganized all variants of copy functions
to make use of same kernel.
As the new MMA instructions need the inputs in 4x2 order for bfloat16,
changing the format in copy/packing code. This avoids permute instructions
in the gemm kernel inner loop.
For the first iteration, it is better to use xvf*ger instead of xvf*gerpp
builtins which helps to avoid setting accumulators to zero. This helps
to reduce few instructions.
Making use of new vector pair POWER10 instructions in dgemv_n and dgemv_t.
Also adding a new block 4x128 to make use of Matrix-Multiply Assist (MMA)
feature introduced in POWER ISA v3.1. Tested on simulator and there
are no new test failures.
As gcc defaults to -malign-power, removing that option. Also
adding -fno-integrated-as to use GNU assembler for powerpc
assembly optimization files. Fixed other compilation errors
reported in dgemv_t.c file.
There is a recent compiler change in __builtin_mma_disassemble_acc() which
affects the order of storing result in POWER10. Also removing new LDFLAG
-mno-power10-stub as it is handled by linker automatically.
This patch introduces new optimized version of SHGEMM kernel
using power10 Matrix-Multiply Assist (MMA) feature introduced in
POWER ISA v3.1. This patch makes use of new POWER10 compute instructions
for matrix multiplication operation.
Tested on simulator and there are no new test failures.
This patch introduces new optimized version of ZGEMM kernel
using power10 Matrix-Multiply Assist (MMA) feature introduced in
POWER ISA v3.1. This patch makes use of new POWER10 compute instructions
for matrix multiplication operation.
Tested on simulator and there are no new test failures.
Cycles count reduced by 30-50% compared to POWER9 version depending on
M/N/K sizes.
This patch introduces new optimized version of SGEMM, CGEMM and DGEMM
using power10 Matrix-Multiply Assist (MMA) feature introduced in
POWER ISA v3.1. This patch makes use of new POWER10 compute instructions
for matrix multiplication operation.
Tested on simulator and there are no new test failures.
Cycles count reduced by 30-50% compared to POWER9 version depending on
M/N/K sizes.
MMA GCC patch for reference:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=8ee2640bfdc62f835ec9740278f948034bc7d9f1
Rajalakshmi Srinivasaraghavan
Martin Kroeker
Gordon Fossum
EGuesnet