Skip to content

Commit d3ef3a4

Browse files
vaiskvvaiskv
committed
Support for SME1 based sgemm_direct kernel for cblas_sgemm level 3 API
* Added ARMV9SME target * Added SGEMM_DIRECT kernel based on SME1
1 parent 18014b0 commit d3ef3a4

22 files changed

+595
-22
lines changed

‎Makefile.arm64‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -30,6 +30,11 @@ FCOMMON_OPT += -march=armv8-a+sve
3030
endif
3131
endif
3232

33+
ifeq ($(CORE), ARMV9SME)
34+
CCOMMON_OPT += -march=armv9-a+sve2+sme
35+
FCOMMON_OPT += -march=armv9-a+sve2
36+
endif
37+
3338
ifeq ($(CORE), CORTEXA53)
3439
CCOMMON_OPT += -march=armv8-a -mtune=cortex-a53
3540
ifneq ($(F_COMPILER), NAG)

‎Makefile.system‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -420,6 +420,7 @@ ifeq ($(ARCH), arm64)
420420
export MACOSX_DEPLOYMENT_TARGET=11.0
421421
ifeq ($(C_COMPILER), GCC)
422422
export NO_SVE = 1
423+
export NO_SME = 1
423424
endif
424425
else
425426
export MACOSX_DEPLOYMENT_TARGET=10.8
@@ -709,6 +710,9 @@ DYNAMIC_CORE += NEOVERSEN2
709710
DYNAMIC_CORE += ARMV8SVE
710711
DYNAMIC_CORE += A64FX
711712
endif
713+
ifneq ($(NO_SME), 1)
714+
DYNAMIC_CORE += ARMV9SME
715+
endif
712716
DYNAMIC_CORE += THUNDERX
713717
DYNAMIC_CORE += THUNDERX2T99
714718
DYNAMIC_CORE += TSV110
@@ -1474,6 +1478,10 @@ ifeq ($(NO_SVE), 1)
14741478
CCOMMON_OPT += -DNO_SVE
14751479
endif
14761480

1481+
ifeq ($(NO_SME), 1)
1482+
CCOMMON_OPT += -DNO_SME
1483+
endif
1484+
14771485
ifdef SMP
14781486
CCOMMON_OPT += -DSMP_SERVER
14791487

‎TargetList.txt‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -111,6 +111,7 @@ THUNDERX3T110
111111
VORTEX
112112
A64FX
113113
ARMV8SVE
114+
ARMV9SME
114115
FT2000
115116

116117
9.System Z:

‎cmake/arch.cmake‎

Lines changed: 15 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -44,9 +44,21 @@ endif ()
4444

4545
if (DYNAMIC_ARCH)
4646
if (ARM64)
47-
set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
48-
if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER 9.99)
49-
set(DYNAMIC_CORE ${DYNAMIC_CORE} NEOVERSEV1 NEOVERSEN2 ARMV8SVE A64FX)
47+
set(DYNAMIC_CORE ARMV8 CORTEXA53 CORTEXA57 THUNDERX THUNDERX2T99 TSV110 EMAG8180 NEOVERSEN1 THUNDERX3T110)
48+
if (${CMAKE_C_COMPILER_ID} STREQUAL "GNU")
49+
if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER_EQUAL 10) # SVE ACLE supported in GCC >= 10
50+
set(DYNAMIC_CORE ${DYNAMIC_CORE} NEOVERSEV1 NEOVERSEN2 ARMV8SVE A64FX)
51+
endif ()
52+
if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER_EQUAL 14) # SME ACLE supported in GCC >= 14
53+
set(DYNAMIC_CORE ${DYNAMIC_CORE} ARMV9SME)
54+
endif()
55+
elseif (${CMAKE_C_COMPILER_ID} MATCHES "Clang")
56+
if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER_EQUAL 11) # SVE ACLE supported in LLVM >= 11
57+
set(DYNAMIC_CORE ${DYNAMIC_CORE} NEOVERSEV1 NEOVERSEN2 ARMV8SVE A64FX)
58+
endif ()
59+
if (${CMAKE_C_COMPILER_VERSION} VERSION_GREATER_EQUAL 19) # SME ACLE supported in LLVM >= 19
60+
set(DYNAMIC_CORE ${DYNAMIC_CORE} ARMV9SME)
61+
endif()
5062
endif ()
5163
if (DYNAMIC_LIST)
5264
set(DYNAMIC_CORE ARMV8 ${DYNAMIC_LIST})

‎cmake/cc.cmake‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -238,6 +238,12 @@ if (${CORE} STREQUAL ARMV8SVE)
238238
endif ()
239239
endif ()
240240

241+
if (${CORE} STREQUAL ARMV9SME)
242+
if (NOT DYNAMIC_ARCH)
243+
set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv9-a+sme")
244+
endif ()
245+
endif ()
246+
241247
if (${CORE} STREQUAL CORTEXA510)
242248
if (NOT DYNAMIC_ARCH)
243249
set (CCOMMON_OPT "${CCOMMON_OPT} -march=armv8-a+sve")

‎cmake/prebuild.cmake‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1014,7 +1014,7 @@ endif ()
10141014
set(ZGEMM_UNROLL_M 4)
10151015
set(ZGEMM_UNROLL_N 4)
10161016
set(SYMV_P 16)
1017-
elseif ("${TCORE}" STREQUAL "NEOVERSEN2")
1017+
elseif ("${TCORE}" STREQUAL "NEOVERSEN2" or "${TCORE}" STREQUAL "ARMV9SME")
10181018
file(APPEND ${TARGET_CONF_TEMP}
10191019
"#define L1_CODE_SIZE\t65536\n"
10201020
"#define L1_CODE_LINESIZE\t64\n"

‎cmake/system.cmake‎

Lines changed: 15 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -310,6 +310,9 @@ if (${TARGET} STREQUAL NEOVERSEV1)
310310
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv8.2-a+sve")
311311
endif()
312312
endif()
313+
if (${TARGET} STREQUAL ARMV9SME)
314+
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -march=armv9-a+sme -O3")
315+
endif()
313316
if (${TARGET} STREQUAL A64FX)
314317
if (${CMAKE_C_COMPILER_ID} STREQUAL "PGI" AND NOT NO_SVE)
315318
set (KERNEL_DEFINITIONS "${KERNEL_DEFINITIONS} -Msve-intrinsics -march=armv8.2-a+sve -mtune=a64fx")
@@ -382,6 +385,8 @@ if (NEED_PIC)
382385
if (NOT NOFORTRAN)
383386
if (${F_COMPILER} STREQUAL "SUN")
384387
set(FCOMMON_OPT "${FCOMMON_OPT} -pic")
388+
elseif (${F_COMPILER} STREQUAL "NAGFOR")
389+
set(FCOMMON_OPT "${FCOMMON_OPT} -PIC")
385390
else ()
386391
set(FCOMMON_OPT "${FCOMMON_OPT} -fPIC")
387392
endif ()
@@ -640,17 +645,17 @@ if (${CMAKE_SYSTEM_NAME} STREQUAL "Windows")
640645
endif ()
641646

642647
if (CMAKE_Fortran_COMPILER)
643-
if ("${F_COMPILER}" STREQUAL "NAG" OR "${F_COMPILER}" STREQUAL "CRAY" OR CMAKE_Fortran_COMPILER_ID MATCHES "LLVMFlang.*")
644-
set(FILTER_FLAGS "-msse3;-mssse3;-msse4.1;-mavx;-mavx2,-mskylake-avx512")
645-
if (CMAKE_Fortran_COMPILER_ID MATCHES "LLVMFlang.*")
646-
message(STATUS "removing fortran flags")
647-
set(FILTER_FLAGS "${FILTER_FLAGS};-m32;-m64")
648+
if ("${F_COMPILER}" STREQUAL "NAGFOR" OR "${F_COMPILER}" STREQUAL "CRAY" OR CMAKE_Fortran_COMPILER_ID MATCHES "LLVMFlang.*")
649+
set(FILTER_FLAGS "-msse3;-mssse3;-msse4.1;-mavx;-mavx2,-mskylake-avx512")
650+
if (CMAKE_Fortran_COMPILER_ID MATCHES "LLVMFlang.*")
651+
message(STATUS "removing fortran flags")
652+
set(FILTER_FLAGS "${FILTER_FLAGS};-m32;-m64")
653+
endif ()
654+
foreach (FILTER_FLAG ${FILTER_FLAGS})
655+
string(REPLACE ${FILTER_FLAG} "" LAPACK_FFLAGS ${LAPACK_FFLAGS})
656+
string(REPLACE ${FILTER_FLAG} "" LAPACK_FPFLAGS ${LAPACK_FPFLAGS})
657+
endforeach ()
648658
endif ()
649-
foreach (FILTER_FLAG ${FILTER_FLAGS})
650-
string(REPLACE ${FILTER_FLAG} "" LAPACK_FFLAGS ${LAPACK_FFLAGS})
651-
string(REPLACE ${FILTER_FLAG} "" LAPACK_FPFLAGS ${LAPACK_FPFLAGS})
652-
endforeach ()
653-
endif ()
654659
endif ()
655660

656661
if ("${F_COMPILER}" STREQUAL "GFORTRAN")

‎common.h‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -696,6 +696,7 @@ void gotoblas_profile_init(void);
696696
void gotoblas_profile_quit(void);
697697

698698
int support_avx512(void);
699+
int support_sme1(void);
699700

700701
#ifdef USE_OPENMP
701702

‎common_arm64.h‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -175,7 +175,7 @@ static inline int blas_quickdivide(blasint x, blasint y){
175175
#define HUGE_PAGESIZE ( 4 << 20)
176176

177177
#ifndef BUFFERSIZE
178-
#if defined(NEOVERSEN1) || defined(NEOVERSEN2) || defined(NEOVERSEV1) || defined(A64FX) || defined(ARMV8SVE)
178+
#if defined(NEOVERSEN1) || defined(NEOVERSEN2) || defined(NEOVERSEV1) || defined(A64FX) || defined(ARMV8SVE) || defined(ARMV9SME)
179179
#define BUFFER_SIZE (32 << 22)
180180
#else
181181
#define BUFFER_SIZE (32 << 20)

‎common_param.h‎

Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -221,6 +221,12 @@ BLASLONG (*ismin_k) (BLASLONG, float *, BLASLONG);
221221
void (*sgemm_direct) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG , float *, BLASLONG , float * , BLASLONG);
222222
int (*sgemm_direct_performant) (BLASLONG M, BLASLONG N, BLASLONG K);
223223
#endif
224+
#ifdef ARCH_ARM64
225+
#ifdef HAVE_SME
226+
void (*sgemm_direct) (BLASLONG, BLASLONG, BLASLONG, float *, BLASLONG , float *, BLASLONG , float * , BLASLONG);
227+
#endif
228+
#endif
229+
224230

225231
int (*sgemm_kernel )(BLASLONG, BLASLONG, BLASLONG, float, float *, float *, float *, BLASLONG);
226232
int (*sgemm_beta )(BLASLONG, BLASLONG, BLASLONG, float, float *, BLASLONG, float *, BLASLONG, float *, BLASLONG);

0 commit comments

Comments
 (0)