Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion cmake/prebuild.cmake
Original file line number Diff line number Diff line change
Expand Up @@ -1475,7 +1475,7 @@ endif ()
"#define L2_ASSOCIATIVE 4\n")
set(SGEMM_UNROLL_M 16)
set(SGEMM_UNROLL_N 8)
set(DGEMM_UNROLL_M 8)
set(DGEMM_UNROLL_M 4)
set(DGEMM_UNROLL_N 8)
set(CGEMM_UNROLL_M 8)
set(CGEMM_UNROLL_N 8)
Expand Down
48 changes: 48 additions & 0 deletions getarch_2nd.c
Original file line number Diff line number Diff line change
Expand Up @@ -103,7 +103,55 @@ int main(int argc, char **argv) {
printf("XGEMM3M_UNROLL_N=%d\n", QGEMM_DEFAULT_UNROLL_N);
#endif

#ifdef ARCH_RISCV64
#ifdef STRMM_DEFAULT_UNROLL_M
printf("STRMM_UNROLL_M=%d\n", STRMM_DEFAULT_UNROLL_M);
#else
printf("STRMM_UNROLL_M=%d\n", SGEMM_DEFAULT_UNROLL_M);
#endif

#ifdef STRMM_DEFAULT_UNROLL_N
printf("STRMM_UNROLL_N=%d\n", STRMM_DEFAULT_UNROLL_N);
#else
printf("STRMM_UNROLL_N=%d\n", SGEMM_DEFAULT_UNROLL_N);
#endif

#ifdef DTRMM_DEFAULT_UNROLL_M
printf("DTRMM_UNROLL_M=%d\n", DTRMM_DEFAULT_UNROLL_M);
#else
printf("DTRMM_UNROLL_M=%d\n", DGEMM_DEFAULT_UNROLL_M);
#endif

#ifdef DTRMM_DEFAULT_UNROLL_N
printf("DTRMM_UNROLL_N=%d\n", DTRMM_DEFAULT_UNROLL_N);
#else
printf("DTRMM_UNROLL_N=%d\n", DGEMM_DEFAULT_UNROLL_N);
#endif

#ifdef CTRMM_DEFAULT_UNROLL_M
printf("CTRMM_UNROLL_M=%d\n", CTRMM_DEFAULT_UNROLL_M);
#else
printf("CTRMM_UNROLL_M=%d\n", CGEMM_DEFAULT_UNROLL_M);
#endif

#ifdef CTRMM_DEFAULT_UNROLL_N
printf("CTRMM_UNROLL_N=%d\n", CTRMM_DEFAULT_UNROLL_N);
#else
printf("CTRMM_UNROLL_N=%d\n", CGEMM_DEFAULT_UNROLL_N);
#endif

#ifdef ZTRMM_DEFAULT_UNROLL_M
printf("ZTRMM_UNROLL_M=%d\n", ZTRMM_DEFAULT_UNROLL_M);
#else
printf("ZTRMM_UNROLL_M=%d\n", ZGEMM_DEFAULT_UNROLL_M);
#endif

#ifdef ZTRMM_DEFAULT_UNROLL_N
printf("ZTRMM_UNROLL_N=%d\n", ZTRMM_DEFAULT_UNROLL_N);
#else
printf("ZTRMM_UNROLL_N=%d\n", ZGEMM_DEFAULT_UNROLL_N);
#endif
#endif /* ARCH_RISCV64 */
}


Expand Down
70 changes: 35 additions & 35 deletions kernel/riscv64/KERNEL.RISCV64_ZVL128B
Original file line number Diff line number Diff line change
Expand Up @@ -95,10 +95,10 @@ DGEMVTKERNEL = gemv_t_rvv.c
CGEMVTKERNEL = zgemv_t_rvv.c
ZGEMVTKERNEL = zgemv_t_rvv.c

SGEMMKERNEL = sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_zvl128b.c
SGEMMKERNEL = sgemm_kernel_zvl128b.c
ifneq ($(filter $(SGEMM_UNROLL_N),4 8 16),)
SGEMMONCOPY = gemm_ncopy_$(SGEMM_UNROLL_N)_rvv.c
SGEMMOTCOPY = gemm_tcopy_$(SGEMM_UNROLL_N)_rvv.c
SGEMMOTCOPY = gemm_tcopy_$(SGEMM_UNROLL_N)_rvv_max.c
else
SGEMMONCOPY = ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
SGEMMOTCOPY = ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
Expand All @@ -109,7 +109,7 @@ SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
ifneq ($(filter $(SGEMM_UNROLL_M),4 8 16),)
SGEMMINCOPY = gemm_ncopy_$(SGEMM_UNROLL_M)_rvv.c
SGEMMITCOPY = gemm_tcopy_$(SGEMM_UNROLL_M)_rvv.c
SGEMMITCOPY = gemm_tcopy_$(SGEMM_UNROLL_M)_rvv_max.c
else
SGEMMINCOPY = ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
SGEMMITCOPY = ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
Expand All @@ -118,10 +118,10 @@ SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

DGEMMKERNEL = dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N)_zvl128b.c
DGEMMKERNEL = dgemm_kernel_zvl128b.c
DGEMMONCOPY = ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
ifneq ($(filter $(DGEMM_UNROLL_N),4 8 16),)
DGEMMOTCOPY = gemm_tcopy_$(DGEMM_UNROLL_N)_rvv.c
DGEMMOTCOPY = gemm_tcopy_$(DGEMM_UNROLL_N)_rvv_max.c
else
DGEMMOTCOPY = ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
endif
Expand All @@ -131,63 +131,63 @@ DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
DGEMMINCOPY = ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
ifneq ($(filter $(DGEMM_UNROLL_M),4 8 16),)
DGEMMITCOPY = gemm_tcopy_$(DGEMM_UNROLL_M)_rvv.c
DGEMMITCOPY = gemm_tcopy_$(DGEMM_UNROLL_M)_rvv_max.c
else
DGEMMITCOPY = ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
endif
DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

CGEMMKERNEL = cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N)_zvl128b.c
CGEMMKERNEL = cgemm_kernel_zvl128b.c
CGEMMONCOPY = ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
CGEMMOTCOPY = ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
CGEMMOTCOPY = zgemm_tcopy_$(CGEMM_UNROLL_N)_rvv_max.c
CGEMMONCOPYOBJ = cgemm_oncopy$(TSUFFIX).$(SUFFIX)
CGEMMOTCOPYOBJ = cgemm_otcopy$(TSUFFIX).$(SUFFIX)

ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
CGEMMINCOPY = ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
CGEMMITCOPY = ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
CGEMMITCOPY = zgemm_tcopy_$(CGEMM_UNROLL_M)_rvv_max.c
CGEMMINCOPYOBJ = cgemm_incopy$(TSUFFIX).$(SUFFIX)
CGEMMITCOPYOBJ = cgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

ZGEMMKERNEL = zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N)_zvl128b.c
ZGEMMKERNEL = zgemm_kernel_zvl128b.c
ZGEMMONCOPY = ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
ZGEMMOTCOPY = ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
ZGEMMOTCOPY = zgemm_tcopy_$(ZGEMM_UNROLL_N)_rvv_max.c
ZGEMMONCOPYOBJ = zgemm_oncopy$(TSUFFIX).$(SUFFIX)
ZGEMMOTCOPYOBJ = zgemm_otcopy$(TSUFFIX).$(SUFFIX)

ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
ZGEMMINCOPY = ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
ZGEMMITCOPY = ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
ZGEMMITCOPY = zgemm_tcopy_$(ZGEMM_UNROLL_M)_rvv_max.c
ZGEMMINCOPYOBJ = zgemm_incopy$(TSUFFIX).$(SUFFIX)
ZGEMMITCOPYOBJ = zgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

STRMMKERNEL = strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_zvl128b.c
STRMMUNCOPY_M = ../generic/trmm_uncopy_$(SGEMM_UNROLL_M).c
STRMMLNCOPY_M = ../generic/trmm_lncopy_$(SGEMM_UNROLL_M).c
STRMMUTCOPY_M = ../generic/trmm_utcopy_$(SGEMM_UNROLL_M).c
STRMMLTCOPY_M = ../generic/trmm_ltcopy_$(SGEMM_UNROLL_M).c

DTRMMKERNEL = dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N)_zvl128b.c
DTRMMUNCOPY_M = ../generic/trmm_uncopy_$(DGEMM_UNROLL_M).c
DTRMMLNCOPY_M = ../generic/trmm_lncopy_$(DGEMM_UNROLL_M).c
DTRMMUTCOPY_M = ../generic/trmm_utcopy_$(DGEMM_UNROLL_M).c
DTRMMLTCOPY_M = ../generic/trmm_ltcopy_$(DGEMM_UNROLL_M).c

CTRMMKERNEL = ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N)_zvl128b.c
CTRMMUNCOPY_M = ../generic/ztrmm_uncopy_$(CGEMM_UNROLL_M).c
CTRMMLNCOPY_M = ../generic/ztrmm_lncopy_$(CGEMM_UNROLL_M).c
CTRMMUTCOPY_M = ../generic/ztrmm_utcopy_$(CGEMM_UNROLL_M).c
CTRMMLTCOPY_M = ../generic/ztrmm_ltcopy_$(CGEMM_UNROLL_M).c

ZTRMMKERNEL = ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N)_zvl128b.c
ZTRMMUNCOPY_M = ../generic/ztrmm_uncopy_$(ZGEMM_UNROLL_M).c
ZTRMMLNCOPY_M = ../generic/ztrmm_lncopy_$(ZGEMM_UNROLL_M).c
ZTRMMUTCOPY_M = ../generic/ztrmm_utcopy_$(ZGEMM_UNROLL_M).c
ZTRMMLTCOPY_M = ../generic/ztrmm_ltcopy_$(ZGEMM_UNROLL_M).c
STRMMKERNEL = strmm_kernel_$(STRMM_UNROLL_M)x$(STRMM_UNROLL_N)_zvl128b.c
STRMMUNCOPY_M = ../generic/trmm_uncopy_$(STRMM_UNROLL_M).c
STRMMLNCOPY_M = ../generic/trmm_lncopy_$(STRMM_UNROLL_M).c
STRMMUTCOPY_M = ../generic/trmm_utcopy_$(STRMM_UNROLL_M).c
STRMMLTCOPY_M = ../generic/trmm_ltcopy_$(STRMM_UNROLL_M).c

DTRMMKERNEL = dtrmm_kernel_$(DTRMM_UNROLL_M)x$(DTRMM_UNROLL_N)_zvl128b.c
DTRMMUNCOPY_M = ../generic/trmm_uncopy_$(DTRMM_UNROLL_M).c
DTRMMLNCOPY_M = ../generic/trmm_lncopy_$(DTRMM_UNROLL_M).c
DTRMMUTCOPY_M = ../generic/trmm_utcopy_$(DTRMM_UNROLL_M).c
DTRMMLTCOPY_M = ../generic/trmm_ltcopy_$(DTRMM_UNROLL_M).c

CTRMMKERNEL = ctrmm_kernel_$(CTRMM_UNROLL_M)x$(CTRMM_UNROLL_N)_zvl128b.c
CTRMMUNCOPY_M = ../generic/ztrmm_uncopy_$(CTRMM_UNROLL_M).c
CTRMMLNCOPY_M = ../generic/ztrmm_lncopy_$(CTRMM_UNROLL_M).c
CTRMMUTCOPY_M = ../generic/ztrmm_utcopy_$(CTRMM_UNROLL_M).c
CTRMMLTCOPY_M = ../generic/ztrmm_ltcopy_$(CTRMM_UNROLL_M).c

ZTRMMKERNEL = ztrmm_kernel_$(ZTRMM_UNROLL_M)x$(ZTRMM_UNROLL_N)_zvl128b.c
ZTRMMUNCOPY_M = ../generic/ztrmm_uncopy_$(ZTRMM_UNROLL_M).c
ZTRMMLNCOPY_M = ../generic/ztrmm_lncopy_$(ZTRMM_UNROLL_M).c
ZTRMMUTCOPY_M = ../generic/ztrmm_utcopy_$(ZTRMM_UNROLL_M).c
ZTRMMLTCOPY_M = ../generic/ztrmm_ltcopy_$(ZTRMM_UNROLL_M).c

STRSMKERNEL_LN = ../generic/trsm_kernel_LN.c
STRSMKERNEL_LT = ../generic/trsm_kernel_LT.c
Expand Down
33 changes: 17 additions & 16 deletions kernel/riscv64/KERNEL.RISCV64_ZVL256B
Original file line number Diff line number Diff line change
Expand Up @@ -90,15 +90,15 @@ DGEMVTKERNEL = gemv_t_vector.c
CGEMVTKERNEL = zgemv_t_vector.c
ZGEMVTKERNEL = zgemv_t_vector.c

STRMMKERNEL = strmm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_zvl256b.c
DTRMMKERNEL = dtrmm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N)_zvl256b.c
CTRMMKERNEL = ctrmm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N)_zvl256b.c
ZTRMMKERNEL = ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N)_zvl256b.c
STRMMKERNEL = strmm_kernel_$(STRMM_UNROLL_M)x$(STRMM_UNROLL_N)_zvl256b.c
DTRMMKERNEL = dtrmm_kernel_$(DTRMM_UNROLL_M)x$(DTRMM_UNROLL_N)_zvl256b.c
CTRMMKERNEL = ctrmm_kernel_$(CTRMM_UNROLL_M)x$(CTRMM_UNROLL_N)_zvl256b.c
ZTRMMKERNEL = ztrmm_kernel_$(ZTRMM_UNROLL_M)x$(ZTRMM_UNROLL_N)_zvl256b.c

SGEMMKERNEL = sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N)_zvl256b.c
SGEMMKERNEL = sgemm_kernel_zvl256b.c
ifneq ($(filter $(SGEMM_UNROLL_N),4 8 16),)
SGEMMONCOPY = gemm_ncopy_$(SGEMM_UNROLL_N)_rvv.c
SGEMMOTCOPY = gemm_tcopy_$(SGEMM_UNROLL_N)_rvv.c
SGEMMOTCOPY = gemm_tcopy_$(SGEMM_UNROLL_N)_rvv_max.c
else
SGEMMONCOPY = ../generic/gemm_ncopy_$(SGEMM_UNROLL_N).c
SGEMMOTCOPY = ../generic/gemm_tcopy_$(SGEMM_UNROLL_N).c
Expand All @@ -108,7 +108,7 @@ SGEMMOTCOPYOBJ = sgemm_otcopy$(TSUFFIX).$(SUFFIX)
ifneq ($(SGEMM_UNROLL_M), $(SGEMM_UNROLL_N))
ifneq ($(filter $(SGEMM_UNROLL_M),4 8 16),)
SGEMMINCOPY = gemm_ncopy_$(SGEMM_UNROLL_M)_rvv.c
SGEMMITCOPY = gemm_tcopy_$(SGEMM_UNROLL_M)_rvv.c
SGEMMITCOPY = gemm_tcopy_$(SGEMM_UNROLL_M)_rvv_max.c
else
SGEMMINCOPY = ../generic/gemm_ncopy_$(SGEMM_UNROLL_M).c
SGEMMITCOPY = ../generic/gemm_tcopy_$(SGEMM_UNROLL_M).c
Expand All @@ -117,48 +117,49 @@ SGEMMINCOPYOBJ = sgemm_incopy$(TSUFFIX).$(SUFFIX)
SGEMMITCOPYOBJ = sgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

DGEMMKERNEL = dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N)_zvl256b.c
DGEMMKERNEL = dgemm_kernel_zvl256b.c
DGEMMONCOPY = ../generic/gemm_ncopy_$(DGEMM_UNROLL_N).c
ifneq ($(filter $(DGEMM_UNROLL_N),4 8 16),)
DGEMMOTCOPY = gemm_tcopy_$(DGEMM_UNROLL_N)_rvv.c
DGEMMOTCOPY = gemm_tcopy_$(DGEMM_UNROLL_N)_rvv_max.c
else
DGEMMOTCOPY = ../generic/gemm_tcopy_$(DGEMM_UNROLL_N).c
endif

DGEMMONCOPYOBJ = dgemm_oncopy$(TSUFFIX).$(SUFFIX)
DGEMMOTCOPYOBJ = dgemm_otcopy$(TSUFFIX).$(SUFFIX)
ifneq ($(DGEMM_UNROLL_M), $(DGEMM_UNROLL_N))
DGEMMINCOPY = ../generic/gemm_ncopy_$(DGEMM_UNROLL_M).c
ifneq ($(filter $(DGEMM_UNROLL_M),4 8 16),)
DGEMMITCOPY = gemm_tcopy_$(DGEMM_UNROLL_M)_rvv.c
DGEMMITCOPY = gemm_tcopy_$(DGEMM_UNROLL_M)_rvv_max.c
else
DGEMMITCOPY = ../generic/gemm_tcopy_$(DGEMM_UNROLL_M).c
endif
DGEMMINCOPYOBJ = dgemm_incopy$(TSUFFIX).$(SUFFIX)
DGEMMITCOPYOBJ = dgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

CGEMMKERNEL = cgemm_kernel_$(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N)_zvl256b.c
CGEMMKERNEL = cgemm_kernel_zvl256b.c
CGEMMONCOPY = ../generic/zgemm_ncopy_$(CGEMM_UNROLL_N).c
CGEMMOTCOPY = ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
CGEMMOTCOPY = zgemm_tcopy_$(CGEMM_UNROLL_N)_rvv_max.c
CGEMMONCOPYOBJ = cgemm_oncopy$(TSUFFIX).$(SUFFIX)
CGEMMOTCOPYOBJ = cgemm_otcopy$(TSUFFIX).$(SUFFIX)

ifneq ($(CGEMM_UNROLL_M), $(CGEMM_UNROLL_N))
CGEMMINCOPY = ../generic/zgemm_ncopy_$(CGEMM_UNROLL_M).c
CGEMMITCOPY = ../generic/zgemm_tcopy_$(CGEMM_UNROLL_M).c
CGEMMITCOPY = zgemm_tcopy_$(CGEMM_UNROLL_M)_rvv_max.c
CGEMMINCOPYOBJ = cgemm_incopy$(TSUFFIX).$(SUFFIX)
CGEMMITCOPYOBJ = cgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif

ZGEMMKERNEL = zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N)_zvl256b.c
ZGEMMKERNEL = zgemm_kernel_zvl256b.c
ZGEMMONCOPY = ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_N).c
ZGEMMOTCOPY = ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_N).c
ZGEMMOTCOPY = zgemm_tcopy_$(ZGEMM_UNROLL_N)_rvv_max.c
ZGEMMONCOPYOBJ = zgemm_oncopy$(TSUFFIX).$(SUFFIX)
ZGEMMOTCOPYOBJ = zgemm_otcopy$(TSUFFIX).$(SUFFIX)

ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
ZGEMMINCOPY = ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
ZGEMMITCOPY = ../generic/zgemm_tcopy_$(ZGEMM_UNROLL_M).c
ZGEMMITCOPY = zgemm_tcopy_$(ZGEMM_UNROLL_M)_rvv_max.c
ZGEMMINCOPYOBJ = zgemm_incopy$(TSUFFIX).$(SUFFIX)
ZGEMMITCOPYOBJ = zgemm_itcopy$(TSUFFIX).$(SUFFIX)
endif
Expand Down
Loading