diff --git a/src/internal/bytealg/compare_amd64.s b/src/internal/bytealg/compare_amd64.s
index 4ccaca5e87..fdd015f560 100644
--- a/src/internal/bytealg/compare_amd64.s
+++ b/src/internal/bytealg/compare_amd64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "go_asm.h"
+#include "asm_amd64.h"
 #include "textflag.h"
 
 TEXT ·Compare<ABIInternal>(SB),NOSPLIT,$0-56
@@ -44,9 +45,13 @@ TEXT cmpbody<>(SB),NOSPLIT,$0-0
 
 	CMPQ	R8, $63
 	JBE	loop
+#ifndef hasAVX2
 	CMPB	internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
 	JEQ     big_loop_avx2
 	JMP	big_loop
+#else
+	JMP	big_loop_avx2
+#endif
 loop:
 	CMPQ	R8, $16
 	JBE	_0through16
@@ -155,6 +160,7 @@ allsame:
 	RET
 
 	// this works for >= 64 bytes of data.
+#ifndef hasAVX2
 big_loop:
 	MOVOU	(SI), X0
 	MOVOU	(DI), X1
@@ -190,6 +196,7 @@ big_loop:
 	CMPQ	R8, $64
 	JBE	loop
 	JMP	big_loop
+#endif
 
 	// Compare 64-bytes per loop iteration.
 	// Loop is unrolled and uses AVX2.
diff --git a/src/internal/bytealg/count_amd64.s b/src/internal/bytealg/count_amd64.s
index fa864c4c76..efb17f84b7 100644
--- a/src/internal/bytealg/count_amd64.s
+++ b/src/internal/bytealg/count_amd64.s
@@ -3,12 +3,15 @@
 // license that can be found in the LICENSE file.
 
 #include "go_asm.h"
+#include "asm_amd64.h"
 #include "textflag.h"
 
 TEXT ·Count(SB),NOSPLIT,$0-40
+#ifndef hasPOPCNT
 	CMPB	internal∕cpu·X86+const_offsetX86HasPOPCNT(SB), $1
 	JEQ	2(PC)
 	JMP	·countGeneric(SB)
+#endif
 	MOVQ	b_base+0(FP), SI
 	MOVQ	b_len+8(FP), BX
 	MOVB	c+24(FP), AL
@@ -16,9 +19,11 @@ TEXT ·Count(SB),NOSPLIT,$0-40
 	JMP	countbody<>(SB)
 
 TEXT ·CountString(SB),NOSPLIT,$0-32
+#ifndef hasPOPCNT
 	CMPB	internal∕cpu·X86+const_offsetX86HasPOPCNT(SB), $1
 	JEQ	2(PC)
 	JMP	·countGenericString(SB)
+#endif
 	MOVQ	s_base+0(FP), SI
 	MOVQ	s_len+8(FP), BX
 	MOVB	c+16(FP), AL
@@ -151,8 +156,10 @@ endofpage:
 	RET
 
 avx2:
+#ifndef hasAVX2
 	CMPB   internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
 	JNE sse
+#endif
 	MOVD AX, X0
 	LEAQ -32(SI)(BX*1), R11
 	VPBROADCASTB  X0, Y1
diff --git a/src/internal/bytealg/equal_amd64.s b/src/internal/bytealg/equal_amd64.s
index dd46e2e0fd..d178a33779 100644
--- a/src/internal/bytealg/equal_amd64.s
+++ b/src/internal/bytealg/equal_amd64.s
@@ -3,6 +3,7 @@
 // license that can be found in the LICENSE file.
 
 #include "go_asm.h"
+#include "asm_amd64.h"
 #include "textflag.h"
 
 // memequal(a, b unsafe.Pointer, size uintptr) bool
@@ -46,6 +47,7 @@ TEXT memeqbody<>(SB),NOSPLIT,$0-0
 	JB	small
 	CMPQ	BX, $64
 	JB	bigloop
+#ifndef hasAVX2
 	CMPB	internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
 	JE	hugeloop_avx2
 
@@ -76,6 +78,7 @@ hugeloop:
 	JEQ	hugeloop
 	XORQ	AX, AX	// return 0
 	RET
+#endif
 
 	// 64 bytes at a time using ymm registers
 hugeloop_avx2:
diff --git a/src/internal/bytealg/index_amd64.s b/src/internal/bytealg/index_amd64.s
index 6193b57239..04314917b8 100644
--- a/src/internal/bytealg/index_amd64.s
+++ b/src/internal/bytealg/index_amd64.s
@@ -233,8 +233,10 @@ success_avx2:
 	VZEROUPPER
 	JMP success
 sse42:
+#ifndef hasSSE42
 	CMPB internal∕cpu·X86+const_offsetX86HasSSE42(SB), $1
 	JNE no_sse42
+#endif
 	CMPQ AX, $12
 	// PCMPESTRI is slower than normal compare,
 	// so using it makes sense only if we advance 4+ bytes per compare
diff --git a/src/internal/bytealg/indexbyte_amd64.s b/src/internal/bytealg/indexbyte_amd64.s
index f78093c539..1ca70e39e2 100644
--- a/src/internal/bytealg/indexbyte_amd64.s
+++ b/src/internal/bytealg/indexbyte_amd64.s
@@ -115,8 +115,10 @@ endofpage:
 	RET
 
 avx2:
+#ifndef hasAVX2
 	CMPB   internal∕cpu·X86+const_offsetX86HasAVX2(SB), $1
 	JNE sse
+#endif
 	MOVD AX, X0
 	LEAQ -32(SI)(BX*1), R11
 	VPBROADCASTB  X0, Y1
diff --git a/src/runtime/asm_amd64.h b/src/runtime/asm_amd64.h
index 49e0ee2323..f7a8896db6 100644
--- a/src/runtime/asm_amd64.h
+++ b/src/runtime/asm_amd64.h
@@ -5,10 +5,21 @@
 // Define features that are guaranteed to be supported by setting the AMD64 variable.
 // If a feature is supported, there's no need to check it at runtime every time.
 
+#ifdef GOAMD64_v2
+#define hasPOPCNT
+#define hasSSE42
+#endif
+
 #ifdef GOAMD64_v3
+#define hasAVX
 #define hasAVX2
+#define hasPOPCNT
+#define hasSSE42
 #endif
 
 #ifdef GOAMD64_v4
+#define hasAVX
 #define hasAVX2
+#define hasPOPCNT
+#define hasSSE42
 #endif
diff --git a/src/runtime/mkpreempt.go b/src/runtime/mkpreempt.go
index 28befcbd0d..61d2d0247e 100644
--- a/src/runtime/mkpreempt.go
+++ b/src/runtime/mkpreempt.go
@@ -126,6 +126,9 @@ func header(arch string) {
 		fmt.Fprintf(out, "//go:build %s || %sle\n\n", base, base)
 	}
 	fmt.Fprintf(out, "#include \"go_asm.h\"\n")
+	if arch == "amd64" {
+		fmt.Fprintf(out, "#include \"asm_amd64.h\"\n")
+	}
 	fmt.Fprintf(out, "#include \"textflag.h\"\n\n")
 	fmt.Fprintf(out, "TEXT ·asyncPreempt(SB),NOSPLIT|NOFRAME,$0-0\n")
 }
@@ -267,8 +270,10 @@ func genAMD64() {
 	// Clear the upper bits to get to a clean state. See issue #37174.
 	// It is safe here as Go code don't use the upper bits of Y registers.
 	p("#ifdef GOOS_darwin")
+	p("#ifndef hasAVX")
 	p("CMPB internal∕cpu·X86+const_offsetX86HasAVX(SB), $0")
 	p("JE 2(PC)")
+	p("#endif")
 	p("VZEROUPPER")
 	p("#endif")
 
diff --git a/src/runtime/preempt_amd64.s b/src/runtime/preempt_amd64.s
index 31f7c8b66f..94a84fb74c 100644
--- a/src/runtime/preempt_amd64.s
+++ b/src/runtime/preempt_amd64.s
@@ -1,6 +1,7 @@
 // Code generated by mkpreempt.go; DO NOT EDIT.
 
 #include "go_asm.h"
+#include "asm_amd64.h"
 #include "textflag.h"
 
 TEXT ·asyncPreempt(SB),NOSPLIT|NOFRAME,$0-0
@@ -27,8 +28,10 @@ TEXT ·asyncPreempt(SB),NOSPLIT|NOFRAME,$0-0
 	MOVQ R14, 96(SP)
 	MOVQ R15, 104(SP)
 	#ifdef GOOS_darwin
+	#ifndef hasAVX
 	CMPB internal∕cpu·X86+const_offsetX86HasAVX(SB), $0
 	JE 2(PC)
+	#endif
 	VZEROUPPER
 	#endif
 	MOVUPS X0, 112(SP)