runtime: simplify detection of preference to use AVX memmove

Reduces cmd/go by 4464 bytes on amd64.

Removes the duplicate detection of AVX support and
presence of Intel processors.

Change-Id: I4670189951a63760fae217708f68d65e94a30dc5
Reviewed-on: https://go-review.googlesource.com/41570Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Keith Randall <khr@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

src/runtime/asm_386.s

@@ -67,14 +67,16 @@ has_cpuid:
 	JNE	notintel
 	CMPL	CX, $0x6C65746E  // "ntel"
 	JNE	notintel
+	MOVB	$1, runtime·isIntel(SB)
 	MOVB	$1, runtime·lfenceBeforeRdtsc(SB)
 notintel:
 
 	// Load EAX=1 cpuid flags
 	MOVL	$1, AX
 	CPUID
-	MOVL	CX, AX // Move to global variable clobbers CX when generating PIC
-	MOVL	AX, runtime·cpuid_ecx(SB)
+	MOVL	CX, DI // Move to global variable clobbers CX when generating PIC
+	MOVL	AX, runtime·cpuid_eax(SB)
+	MOVL	DI, runtime·cpuid_ecx(SB)
 	MOVL	DX, runtime·cpuid_edx(SB)
 
 	// Check for MMX support

src/runtime/asm_amd64.s

@@ -41,12 +41,14 @@ TEXT runtime·rt0_go(SB),NOSPLIT,$0
 	JNE	notintel
 	CMPL	CX, $0x6C65746E  // "ntel"
 	JNE	notintel
+	MOVB	$1, runtime·isIntel(SB)
 	MOVB	$1, runtime·lfenceBeforeRdtsc(SB)
 notintel:
 
 	// Load EAX=1 cpuid flags
 	MOVQ	$1, AX
 	CPUID
+	MOVL	AX, runtime·cpuid_eax(SB)
 	MOVL	CX, runtime·cpuid_ecx(SB)
 	MOVL	DX, runtime·cpuid_edx(SB)
 

src/runtime/asm_amd64p32.s

@@ -32,8 +32,19 @@ TEXT runtime·rt0_go(SB),NOSPLIT,$0
 	CPUID
 	CMPQ	AX, $0
 	JE	nocpuinfo
+
+	CMPL	BX, $0x756E6547  // "Genu"
+	JNE	notintel
+	CMPL	DX, $0x49656E69  // "ineI"
+	JNE	notintel
+	CMPL	CX, $0x6C65746E  // "ntel"
+	JNE	notintel
+	MOVB	$1, runtime·isIntel(SB)
+notintel:
+
 	MOVQ	$1, AX
 	CPUID
+	MOVL	AX, runtime·cpuid_eax(SB)
 	MOVL	CX, runtime·cpuid_ecx(SB)
 	MOVL	DX, runtime·cpuid_edx(SB)
 nocpuinfo:	

src/runtime/cpuflags_amd64.go

@@ -4,72 +4,17 @@
 
 package runtime
 
-var vendorStringBytes [12]byte
-var maxInputValue uint32
-var featureFlags uint32
-var processorVersionInfo uint32
-
-var useRepMovs = true
-
-func hasFeature(feature uint32) bool {
-	return (featureFlags & feature) != 0
-}
-
-func cpuid_low(arg1, arg2 uint32) (eax, ebx, ecx, edx uint32) // implemented in cpuidlow_amd64.s
-func xgetbv_low(arg1 uint32) (eax, edx uint32)                // implemented in cpuidlow_amd64.s
+var useAVXmemmove bool
 
 func init() {
-	const cfOSXSAVE uint32 = 1 << 27
-	const cfAVX uint32 = 1 << 28
-
-	leaf0()
-	leaf1()
-
-	enabledAVX := false
-	// Let's check if OS has set CR4.OSXSAVE[bit 18]
-	// to enable XGETBV instruction.
-	if hasFeature(cfOSXSAVE) {
-		eax, _ := xgetbv_low(0)
-		// Let's check that XCR0[2:1] = ‘11b’
-		// i.e. XMM state and YMM state are enabled by OS.
-		enabledAVX = (eax & 0x6) == 0x6
-	}
+	// Let's remove stepping and reserved fields
+	processorVersionInfo := cpuid_eax & 0x0FFF3FF0
 
-	isIntelBridgeFamily := (processorVersionInfo == 0x206A0 ||
+	isIntelBridgeFamily := isIntel &&
+		(processorVersionInfo == 0x206A0 ||
 			processorVersionInfo == 0x206D0 ||
 			processorVersionInfo == 0x306A0 ||
-		processorVersionInfo == 0x306E0) &&
-		isIntel()
-
-	useRepMovs = !(hasFeature(cfAVX) && enabledAVX) || isIntelBridgeFamily
-}
-
-func leaf0() {
-	eax, ebx, ecx, edx := cpuid_low(0, 0)
-	maxInputValue = eax
-	int32ToBytes(ebx, vendorStringBytes[0:4])
-	int32ToBytes(edx, vendorStringBytes[4:8])
-	int32ToBytes(ecx, vendorStringBytes[8:12])
-}
-
-func leaf1() {
-	if maxInputValue < 1 {
-		return
-	}
-	eax, _, ecx, _ := cpuid_low(1, 0)
-	// Let's remove stepping and reserved fields
-	processorVersionInfo = eax & 0x0FFF3FF0
-	featureFlags = ecx
-}
-
-func int32ToBytes(arg uint32, buffer []byte) {
-	buffer[3] = byte(arg >> 24)
-	buffer[2] = byte(arg >> 16)
-	buffer[1] = byte(arg >> 8)
-	buffer[0] = byte(arg)
-}
+			processorVersionInfo == 0x306E0)
 
-func isIntel() bool {
-	intelSignature := [12]byte{'G', 'e', 'n', 'u', 'i', 'n', 'e', 'I', 'n', 't', 'e', 'l'}
-	return vendorStringBytes == intelSignature
+	useAVXmemmove = support_avx && !isIntelBridgeFamily
 }

src/runtime/cpuidlow_amd64.s

-// Copyright 2015 The Go Authors. All rights reserved.
-// Use of this source code is governed by a BSD-style
-// license that can be found in the LICENSE file.
-
-// func cpuid_low(arg1, arg2 uint32) (eax, ebx, ecx, edx uint32)
-TEXT ·cpuid_low(SB), 4, $0-24
-    MOVL    arg1+0(FP), AX
-    MOVL    arg2+4(FP), CX
-    CPUID
-    MOVL AX, eax+8(FP)
-    MOVL BX, ebx+12(FP)
-    MOVL CX, ecx+16(FP)
-    MOVL DX, edx+20(FP)
-    RET
-// func xgetbv_low(arg1 uint32) (eax, edx uint32)
-TEXT ·xgetbv_low(SB), 4, $0-16
-    MOVL arg1+0(FP), CX
-    // XGETBV
-    BYTE $0x0F; BYTE $0x01; BYTE $0xD0
-    MOVL AX,eax+8(FP)
-    MOVL DX,edx+12(FP)
-    RET

src/runtime/memmove_amd64.s

@@ -64,8 +64,8 @@ tail:
 	JBE	move_129through256
 	// TODO: use branch table and BSR to make this just a single dispatch
 
-	TESTB	$1, runtime·useRepMovs(SB)
-	JZ	avxUnaligned
+	TESTB	$1, runtime·useAVXmemmove(SB)
+	JNZ	avxUnaligned
 
 /*
  * check and set for backwards

src/runtime/runtime2.go

@@ -727,10 +727,12 @@ var (
 	newprocs    int32
 
 	// Information about what cpu features are available.
-	// Set on startup in asm_{x86,amd64}.s.
+	// Set on startup in asm_{386,amd64,amd64p32}.s.
+	cpuid_eax         uint32
 	cpuid_ecx         uint32
 	cpuid_edx         uint32
-	cpuid_ebx7        uint32
+	cpuid_ebx7        uint32 // not set on amd64p32
+	isIntel           bool
 	lfenceBeforeRdtsc bool
 	support_avx       bool
 	support_avx2      bool