fix: use sse instead of sse4 (#305)

* fix: use sse instead of sse4 * fix: use dispatch * fix: remove lzero Co-authored-by: liuqiang <liuqiang.06@bytedance.com>
bytedance · Sep 26, 2022 · 56e81a6 · 56e81a6
1 parent 5e54c02
commit 56e81a6
Show file tree

Hide file tree

Showing 20 changed files with 8,748 additions and 11,357 deletions.
diff --git a/Makefile b/Makefile
@@ -14,22 +14,22 @@
 # limitations under the License.
 #
 
-ARCH			:= avx avx2 sse4
+ARCH			:= avx avx2 sse
 TMP_DIR			:= output
 OUT_DIR			:= internal/native
 SRC_FILE		:= native/native.c
 
 CPU_avx			:= amd64
 CPU_avx2		:= amd64
-CPU_sse4		:= amd64
+CPU_sse  		:= amd64
 
 TMPL_avx		:= fastint_amd64_test fastfloat_amd64_test native_amd64_test native_export_amd64
 TMPL_avx2		:= fastint_amd64_test fastfloat_amd64_test native_amd64_test native_export_amd64
-TMPL_sse4		:= fastint_amd64_test fastfloat_amd64_test native_amd64_test native_export_amd64
+TMPL_sse 		:= fastint_amd64_test fastfloat_amd64_test native_amd64_test native_export_amd64
 
-CFLAGS_avx		:= -msse4 -mavx -mno-avx2 -DUSE_AVX=1 -DUSE_AVX2=0
-CFLAGS_avx2		:= -msse4 -mavx -mavx2    -DUSE_AVX=1 -DUSE_AVX2=1 
-CFLAGS_sse4		:= -msse4 -mno-avx -mno-avx2
+CFLAGS_avx		:= -msse -mno-sse4 -mavx -mno-avx2 -DUSE_AVX=1 -DUSE_AVX2=0
+CFLAGS_avx2		:= -msse -mno-sse4 -mavx -mavx2    -DUSE_AVX=1 -DUSE_AVX2=1 
+CFLAGS_sse		:= -msse -mno-sse4 -mno-avx -mno-avx2
 
 CC_amd64		:= clang
 ASM2ASM_amd64	:= tools/asm2asm/asm2asm.py
@@ -100,8 +100,8 @@ endef
 all: ${ARCH}
 
 clean:
-	rm -vfr ${TMP_DIR}/{sse4,avx,avx2}
-	rm -vfr ${OUT_DIR}/{sse4,avx,avx2}
+	rm -vfr ${TMP_DIR}/{sse,avx,avx2}
+	rm -vfr ${OUT_DIR}/{sse,avx,avx2}
 
 $(foreach 								\
 	arch,								\

diff --git a/internal/cpu/features.go b/internal/cpu/features.go
@@ -26,7 +26,7 @@ import (
 var (
     HasAVX  = cpuid.CPU.Has(cpuid.AVX)
     HasAVX2 = cpuid.CPU.Has(cpuid.AVX2)
-    HasSSE4 = cpuid.CPU.Has(cpuid.SSE4)
+    HasSSE = cpuid.CPU.Has(cpuid.SSE)
 )
 
 func init() {

diff --git a/internal/native/avx/native_amd64.go b/internal/native/avx/native_amd64.go
diff --git a/internal/native/avx/native_amd64.s b/internal/native/avx/native_amd64.s
diff --git a/internal/native/avx/native_subr_amd64.go b/internal/native/avx/native_subr_amd64.go
diff --git a/internal/native/avx2/native_amd64.go b/internal/native/avx2/native_amd64.go