[go] internal/runtime/maps,runtime/: take key by value in memHash{32,64}AES; replace strHashAes by call to memHashAES

10 views
Skip to first unread message

Arseny Samoylov (Gerrit)

unread,
May 9, 2026, 10:15:05 AM (10 days ago) May 9
to goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com

Arseny Samoylov has uploaded the change for review

Commit message

internal/runtime/maps,runtime/: take key by value in memHash{32,64}AES; replace strHashAes by call to memHashAES
Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5

Change diff

diff --git a/src/internal/runtime/maps/memhash_386.s b/src/internal/runtime/maps/memhash_386.s
index 75b1383..2a0d82f 100644
--- a/src/internal/runtime/maps/memhash_386.s
+++ b/src/internal/runtime/maps/memhash_386.s
@@ -5,19 +5,20 @@
 #include "textflag.h"
 
 // hash function using AES hardware instructions
+
+// func memHash32AES(k uint32, h uintptr) uintptr
 TEXT ·memHash32AES(SB),NOSPLIT,$0-12
-	MOVL	p+0(FP), AX	// ptr to data
-	MOVL	h+4(FP), X0	// seed
-	PINSRD	$1, (AX), X0	// data
+	MOVL	h+4(FP), X0	    // seed
+	PINSRD	$1, k+0(FP), X0	// data
 	AESENC	·aeskeysched+0(SB), X0
 	AESENC	·aeskeysched+16(SB), X0
 	AESENC	·aeskeysched+32(SB), X0
 	MOVL	X0, ret+8(FP)
 	RET
 
+// func memHash64AES(k uint64, h uintptr) uintptr
 TEXT ·memHash64AES(SB),NOSPLIT,$0-12
-	MOVL	p+0(FP), AX	// ptr to data
-	MOVQ	(AX), X0	// data
+	MOVQ	k+0(FP), X0	    // data
 	PINSRD	$2, h+4(FP), X0	// seed
 	AESENC	·aeskeysched+0(SB), X0
 	AESENC	·aeskeysched+16(SB), X0
@@ -25,24 +26,18 @@
 	MOVL	X0, ret+8(FP)
 	RET
 
+// func memHashAES(p unsafe.Pointer, h, size uintptr) uintptr
 TEXT ·memHashAES(SB),NOSPLIT,$0-16
-	MOVL	p+0(FP), AX	// ptr to data
-	MOVL	s+8(FP), BX	// size
+	// AX: data
+	// BX: size
+	// CX: hash seed
+	// DX: address to put return value
+	MOVL	p+0(FP), AX
+	MOVL	s+8(FP), BX
+	MOVL    h+4(FP), CX
 	LEAL	ret+12(FP), DX
-	JMP	·aeshashbody<>(SB)
 
-TEXT ·strHashAES(SB),NOSPLIT,$0-12
-	MOVL	p+0(FP), AX	// ptr to string object
-	MOVL	4(AX), BX	// length of string
-	MOVL	(AX), AX	// string data
-	LEAL	ret+8(FP), DX
-	JMP	·aeshashbody<>(SB)
-
-// AX: data
-// BX: length
-// DX: address to put return value
-TEXT ·aeshashbody<>(SB),NOSPLIT,$0-0
-	MOVL	h+4(FP), X0	            // 32 bits of per-table hash seed
+	MOVL	CX, X0	            // 32 bits of per-table hash seed
 	PINSRW	$4, BX, X0	            // 16 bits of length
 	PSHUFHW	$0, X0, X0	            // replace size with its low 2 bytes repeated 4 times
 	MOVO	X0, X1                      // save unscrambled seed
diff --git a/src/internal/runtime/maps/memhash_aes.go b/src/internal/runtime/maps/memhash_aes.go
index 41e90d3..8893cdd 100644
--- a/src/internal/runtime/maps/memhash_aes.go
+++ b/src/internal/runtime/maps/memhash_aes.go
@@ -19,23 +19,20 @@
 	return memHashFallback(p, h, s)
 }
 
-func MemHash32(p unsafe.Pointer, h uintptr) uintptr {
+func MemHash32(k uint32, h uintptr) uintptr {
 	if UseAeshash {
-		return memHash32AES(p, h)
+		return memHash32AES(k, h)
 	}
-	return memHash32Fallback(p, h)
+	return memHash32Fallback(k, h)
 }
 
-func MemHash64(p unsafe.Pointer, h uintptr) uintptr {
+func MemHash64(k uint64, h uintptr) uintptr {
 	if UseAeshash {
-		return memHash64AES(p, h)
+		return memHash64AES(k, h)
 	}
-	return memHash64Fallback(p, h)
+	return memHash64Fallback(k, h)
 }
 
-func StrHash(p unsafe.Pointer, h uintptr) uintptr {
-	if UseAeshash {
-		return strHashAES(p, h)
-	}
-	return strHashFallback(p, h)
+func StrHash(s string, h uintptr) uintptr {
+	return MemHash(unsafe.Pointer(unsafe.StringData(s)), h, uintptr(len(s)))
 }
diff --git a/src/internal/runtime/maps/memhash_aes_asm.go b/src/internal/runtime/maps/memhash_aes_asm.go
index ae3dabd..0c8a5413 100644
--- a/src/internal/runtime/maps/memhash_aes_asm.go
+++ b/src/internal/runtime/maps/memhash_aes_asm.go
@@ -18,10 +18,7 @@
 func memHashAES(p unsafe.Pointer, h, s uintptr) uintptr
 
 //go:noescape
-func memHash32AES(p unsafe.Pointer, h uintptr) uintptr
+func memHash32AES(k uint32, h uintptr) uintptr
 
 //go:noescape
-func memHash64AES(p unsafe.Pointer, h uintptr) uintptr
-
-//go:noescape
-func strHashAES(p unsafe.Pointer, h uintptr) uintptr
+func memHash64AES(k uint64, h uintptr) uintptr
diff --git a/src/internal/runtime/maps/memhash_aes_simd.go b/src/internal/runtime/maps/memhash_aes_simd.go
index 4cd71b2..bef533e 100644
--- a/src/internal/runtime/maps/memhash_aes_simd.go
+++ b/src/internal/runtime/maps/memhash_aes_simd.go
@@ -13,9 +13,9 @@
 
 const memHashUsesVAES = true
 
-func memHash32AES(p unsafe.Pointer, seed uintptr) uintptr {
+func memHash32AES(k uint32, seed uintptr) uintptr {
 	var state archsimd.Uint64x2
-	state = state.SetElem(0, uint64(seed)).SetElem(1, uint64(*(*uint32)(p)))
+	state = state.SetElem(0, uint64(seed)).SetElem(1, uint64(k))
 
 	hash := state.
 		AsUint8x16().
@@ -27,9 +27,9 @@
 	return uintptr(hash)
 }
 
-func memHash64AES(p unsafe.Pointer, seed uintptr) uintptr {
+func memHash64AES(k uint64, seed uintptr) uintptr {
 	var state archsimd.Uint64x2
-	state = state.SetElem(0, uint64(seed)).SetElem(1, *(*uint64)(p))
+	state = state.SetElem(0, uint64(seed)).SetElem(1, k)
 
 	hash := state.
 		AsUint8x16().
@@ -41,12 +41,9 @@
 	return uintptr(hash)
 }
 
-// TODO: Both strHashAES and memHashAES use aeshashbody that is quite large.
-// So there is no point in rewriting them using simd intrinsics, since they won't be inlinable.
+// TODO: memHashAES is quite large.
+// So there is no point in rewriting it using simd intrinsics, since it won't be inlinable.
 // Maybe in future we can do it for better maitanability.
 //
 //go:noescape
 func memHashAES(p unsafe.Pointer, h, s uintptr) uintptr
-
-//go:noescape
-func strHashAES(p unsafe.Pointer, h uintptr) uintptr
diff --git a/src/internal/runtime/maps/memhash_amd64.s b/src/internal/runtime/maps/memhash_amd64.s
index c649211..c922afa 100644
--- a/src/internal/runtime/maps/memhash_amd64.s
+++ b/src/internal/runtime/maps/memhash_amd64.s
@@ -7,24 +7,11 @@
 // func memHashAES(p unsafe.Pointer, h, s uintptr) uintptr
 // hash function using AES hardware instructions
 TEXT ·memHashAES<ABIInternal>(SB),NOSPLIT,$0-32
-	// AX = ptr to data
-	// BX = seed
-	// CX = size
-	JMP	·aeshashbody<>(SB)
+	// AX: data
+	// BX: hash seed
+	// CX: length
+	// At return: AX = return value
 
-// func strhashAES(p unsafe.Pointer, h uintptr) uintptr
-TEXT ·strHashAES<ABIInternal>(SB),NOSPLIT,$0-24
-	// AX = ptr to string struct
-	// BX = seed
-	MOVQ	8(AX), CX	// length of string
-	MOVQ	(AX), AX	// string data
-	JMP	·aeshashbody<>(SB)
-
-// AX: data
-// BX: hash seed
-// CX: length
-// At return: AX = return value
-TEXT ·aeshashbody<>(SB),NOSPLIT,$0-0
 	// Fill an SSE register with our seeds.
 	MOVQ	BX, X0				// 64 bits of per-table hash seed
 	PINSRW	$4, CX, X0			// 16 bits of length
diff --git a/src/internal/runtime/maps/memhash_arm64.s b/src/internal/runtime/maps/memhash_arm64.s
index 3a5ae09..eb5864b 100644
--- a/src/internal/runtime/maps/memhash_arm64.s
+++ b/src/internal/runtime/maps/memhash_arm64.s
@@ -4,13 +4,13 @@
 
 #include "textflag.h"
 
-// func memHash32AES(p unsafe.Pointer, h uintptr) uintptr
+// func memHash32AES(k uint32, h uintptr) uintptr
 TEXT ·memHash32AES<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-24
 	MOVD	$·aeskeysched+0(SB), R3
 
 	VEOR	V0.B16, V0.B16, V0.B16
 	VLD1	(R3), [V2.B16]
-	VLD1	(R0), V0.S[2]
+	VMOV	R0, V0.S[2]
 	VMOV	R1, V0.D[0]
 
 	AESE	V2.B16, V0.B16
@@ -22,13 +22,13 @@
 	VMOV	V0.D[0], R0
 	RET
 
-// func memHash64AES(p unsafe.Pointer, h uintptr) uintptr
+// func memHash64AES(k uint64, h uintptr) uintptr
 TEXT ·memHash64AES<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-24
 	MOVD	$·aeskeysched+0(SB), R3
 
 	VEOR	V0.B16, V0.B16, V0.B16
 	VLD1	(R3), [V2.B16]
-	VLD1	(R0), V0.D[1]
+	VMOV	R0, V0.D[1]
 	VMOV	R1, V0.D[0]
 
 	AESE	V2.B16, V0.B16
@@ -42,18 +42,10 @@
 
 // func memHashAES(p unsafe.Pointer, h, size uintptr) uintptr
 TEXT ·memHashAES<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-32
-	B	·aeshashbody<>(SB)
-
-// func strHashAES(p unsafe.Pointer, h uintptr) uintptr
-TEXT ·strHashAES<ABIInternal>(SB),NOSPLIT|NOFRAME,$0-24
-	LDP	(R0), (R0, R2)	// string data / length
-	B	·aeshashbody<>(SB)
-
-// R0: data
-// R1: seed data
-// R2: length
-// At return, R0 = return value
-TEXT ·aeshashbody<>(SB),NOSPLIT|NOFRAME,$0
+	// R0: data
+	// R1: seed data
+	// R2: length
+	// At return, R0 = return value
 	VEOR	V30.B16, V30.B16, V30.B16
 	VMOV	R1, V30.D[0]
 	VMOV	R2, V30.D[1] // load length into seed
diff --git a/src/internal/runtime/maps/memhash_noaes.go b/src/internal/runtime/maps/memhash_noaes.go
index c9a297f..3bdf64e 100644
--- a/src/internal/runtime/maps/memhash_noaes.go
+++ b/src/internal/runtime/maps/memhash_noaes.go
@@ -14,30 +14,30 @@
 const memHashAESImplemented = false
 const memHashUsesVAES = false
 
-func memHash32AES(p unsafe.Pointer, h uintptr) uintptr {
+func memHash32AES(k uint32, h uintptr) uintptr {
 	panic("memHash32AES not implemented")
 }
 
-func memHash64AES(p unsafe.Pointer, h uintptr) uintptr {
+func memHash64AES(k uint64, h uintptr) uintptr {
 	panic("memHash64AES not implemented")
 }
 
-func strHashAES(p unsafe.Pointer, h uintptr) uintptr {
-	panic("strHashAES not implemented")
+func memHashAES(p unsafe.Pointer, h, s uintptr) uintptr {
+	panic("memHashAES not implemented")
 }
 
 func MemHash(p unsafe.Pointer, h, s uintptr) uintptr {
 	return memHashFallback(p, h, s)
 }
 
-func MemHash32(p unsafe.Pointer, h uintptr) uintptr {
-	return memHash32Fallback(p, h)
+func MemHash32(k uint32, h uintptr) uintptr {
+	return memHash32Fallback(k, h)
 }
 
-func MemHash64(p unsafe.Pointer, h uintptr) uintptr {
-	return memHash64Fallback(p, h)
+func MemHash64(k uint64, h uintptr) uintptr {
+	return memHash64Fallback(k, h)
 }
 
-func StrHash(p unsafe.Pointer, h uintptr) uintptr {
-	return strHashFallback(p, h)
+func StrHash(s string, h uintptr) uintptr {
+	return memHashFallback(unsafe.Pointer(unsafe.StringData(s)), h, uintptr(len(s)))
 }
diff --git a/src/internal/runtime/maps/memhash_nosimd_amd64.s b/src/internal/runtime/maps/memhash_nosimd_amd64.s
index 3471779..ad3790c 100644
--- a/src/internal/runtime/maps/memhash_nosimd_amd64.s
+++ b/src/internal/runtime/maps/memhash_nosimd_amd64.s
@@ -6,26 +6,26 @@
 
 #include "textflag.h"
 
-// func memHash32AES(p unsafe.Pointer, h uintptr) uintptr
+// func memHash32AES(k uint32, h uintptr) uintptr
 // ABIInternal for performance.
 TEXT ·memHash32AES<ABIInternal>(SB),NOSPLIT,$0-24
 	// AX = ptr to data
 	// BX = seed
-	MOVQ	BX, X0	// X0 = seed
-	PINSRD	$2, (AX), X0	// data
+	MOVQ	BX, X0	    // X0 = seed
+	PINSRD	$2, AX, X0	// data
 	AESENC	·aeskeysched+0(SB), X0
 	AESENC	·aeskeysched+16(SB), X0
 	AESENC	·aeskeysched+32(SB), X0
 	MOVQ	X0, AX	// return X0
 	RET
 
-// func memHash64AES(p unsafe.Pointer, h uintptr) uintptr
+// func memHash64AES(k uint64, h uintptr) uintptr
 // ABIInternal for performance.
 TEXT ·memHash64AES<ABIInternal>(SB),NOSPLIT,$0-24
 	// AX = ptr to data
 	// BX = seed
-	MOVQ	BX, X0	// X0 = seed
-	PINSRQ	$1, (AX), X0	// data
+	MOVQ	BX, X0   	// X0 = seed
+	PINSRQ	$1, AX, X0	// data
 	AESENC	·aeskeysched+0(SB), X0
 	AESENC	·aeskeysched+16(SB), X0
 	AESENC	·aeskeysched+32(SB), X0
diff --git a/src/internal/runtime/maps/runtime_alg.go b/src/internal/runtime/maps/runtime_alg.go
index 9fc3225..1c80e7b 100644
--- a/src/internal/runtime/maps/runtime_alg.go
+++ b/src/internal/runtime/maps/runtime_alg.go
@@ -38,8 +38,7 @@
 		cpu.X86.HasSSSE3 && // PSHUFB
 		cpu.X86.HasSSE41 { // PINSR{D,Q}
 
-		// In aeshashbody (that is used by memhash & strhash)
-		// we have global variables that should be properly aligned.
+		// In memHashAES we have global variables that should be properly aligned.
 		//
 		// See #12415
 		if !checkMasksAndShiftsAlignment() {
@@ -71,15 +70,6 @@
 	}
 }
 
-func strHashFallback(a unsafe.Pointer, h uintptr) uintptr {
-	type stringStruct struct {
-		str unsafe.Pointer
-		len int
-	}
-	x := (*stringStruct)(a)
-	return memHashFallback(x.str, h, uintptr(x.len))
-}
-
 //go:nosplit
 func add(p unsafe.Pointer, x uintptr) unsafe.Pointer {
 	return unsafe.Pointer(uintptr(p) + x)
diff --git a/src/internal/runtime/maps/runtime_fast32.go b/src/internal/runtime/maps/runtime_fast32.go
index dce1676..d80489e 100644
--- a/src/internal/runtime/maps/runtime_fast32.go
+++ b/src/internal/runtime/maps/runtime_fast32.go
@@ -63,18 +63,6 @@
 		return unsafe.Pointer(&zeroVal[0]), false
 	}
 
-	// Don't pass address of the key directly to the hashing function.
-	// Hashing functions are implemented in Go assembly and cannot be inlined,
-	// so compiler doesn't optimize redundant address taking/dereference.
-	//
-	// Taking &key makes compiler treat key as address-taken, which forces it to spill on the stack
-	// and reload it in the loop.
-	// This is suboptimal for performance.
-	//
-	// Note: Even when we pass k (local copy of key), the compiler still spills the key to the stack.
-	// However, from compiler's perspective, key is no longer address-taken and
-	// filled back in register before the loop.
-	k := key
 	var hash uintptr
 	// Explicitly inline MemHash32.
 	// MemHash32 cost is higher than the threshold for inlining.
@@ -84,9 +72,9 @@
 	// Note: memHashAESImplemented is compile time constant. We use it to remove runtime UseAeshash check
 	// for architectures where we don't have AES hashing implementations.
 	if memHashAESImplemented && UseAeshash {
-		hash = memHash32AES(unsafe.Pointer(&k), m.seed)
+		hash = memHash32AES(key, m.seed)
 	} else {
-		hash = memHash32Fallback(unsafe.Pointer(&k), m.seed)
+		hash = memHash32Fallback(key, m.seed)
 	}
 
 	// Select table.
@@ -209,15 +197,12 @@
 		fatal("concurrent map writes")
 	}
 
-	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
 	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
 	if memHashAESImplemented && UseAeshash {
-		hash = memHash32AES(unsafe.Pointer(&k), m.seed)
+		hash = memHash32AES(key, m.seed)
 	} else {
-		hash = memHash32Fallback(unsafe.Pointer(&k), m.seed)
+		hash = memHash32Fallback(key, m.seed)
 	}
 
 	// Set writing after calling Hasher, since Hasher may panic, in which
@@ -361,15 +346,12 @@
 		fatal("concurrent map writes")
 	}
 
-	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
 	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
 	if memHashAESImplemented && UseAeshash {
-		hash = memHash32AES(unsafe.Pointer(&k), m.seed)
+		hash = memHash32AES(*(*uint32)((unsafe.Pointer)(&key)), m.seed)
 	} else {
-		hash = memHash32Fallback(unsafe.Pointer(&k), m.seed)
+		hash = memHash32Fallback(*(*uint32)((unsafe.Pointer)(&key)), m.seed)
 	}
 
 	// Set writing after calling Hasher, since Hasher may panic, in which
diff --git a/src/internal/runtime/maps/runtime_fast64.go b/src/internal/runtime/maps/runtime_fast64.go
index 82b8e98..8a2c53d 100644
--- a/src/internal/runtime/maps/runtime_fast64.go
+++ b/src/internal/runtime/maps/runtime_fast64.go
@@ -63,15 +63,12 @@
 		return unsafe.Pointer(&zeroVal[0]), false
 	}
 
-	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
 	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
 	if memHashAESImplemented && UseAeshash {
-		hash = memHash64AES(unsafe.Pointer(&k), m.seed)
+		hash = memHash64AES(key, m.seed)
 	} else {
-		hash = memHash64Fallback(unsafe.Pointer(&k), m.seed)
+		hash = memHash64Fallback(key, m.seed)
 	}
 
 	// Select table.
@@ -195,15 +192,12 @@
 		fatal("concurrent map writes")
 	}
 
-	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
 	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
 	if memHashAESImplemented && UseAeshash {
-		hash = memHash64AES(unsafe.Pointer(&k), m.seed)
+		hash = memHash64AES(key, m.seed)
 	} else {
-		hash = memHash64Fallback(unsafe.Pointer(&k), m.seed)
+		hash = memHash64Fallback(key, m.seed)
 	}
 
 	// Set writing after calling Hasher, since Hasher may panic, in which
@@ -416,15 +410,12 @@
 		fatal("concurrent map writes")
 	}
 
-	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
 	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
 	if memHashAESImplemented && UseAeshash {
-		hash = memHash64AES(unsafe.Pointer(&k), m.seed)
+		hash = memHash64AES(*(*uint64)((unsafe.Pointer)(&key)), m.seed)
 	} else {
-		hash = memHash64Fallback(unsafe.Pointer(&k), m.seed)
+		hash = memHash64Fallback(*(*uint64)((unsafe.Pointer)(&key)), m.seed)
 	}
 
 	// Set writing after calling Hasher, since Hasher may panic, in which
diff --git a/src/internal/runtime/maps/runtime_faststr.go b/src/internal/runtime/maps/runtime_faststr.go
index 4d6839f..85cdccd 100644
--- a/src/internal/runtime/maps/runtime_faststr.go
+++ b/src/internal/runtime/maps/runtime_faststr.go
@@ -64,11 +64,13 @@
 
 dohash:
 	// This path will cost 1 hash and 1+ε comparisons.
-
+	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
-	hash := StrHash(unsafe.Pointer(&k), m.seed)
+	if memHashAESImplemented && UseAeshash {
+		hash = memHashAES(unsafe.Pointer(unsafe.StringData(key)), m.seed, uintptr(len(key)))
+	} else {
+		hash = memHashFallback(unsafe.Pointer(unsafe.StringData(key)), m.seed, uintptr(len(key)))
+	}
 	h2 := uint8(h2(hash))
 	ctrls = *g.ctrls()
 	slotKey = g.key(typ, 0)
@@ -94,26 +96,19 @@
 	if len(a) != len(b) {
 		return false
 	}
-	x, y := stringPtr(a), stringPtr(b)
+	x, y := unsafe.Pointer(unsafe.StringData(a)), unsafe.Pointer(unsafe.StringData(b))
 	// Check first 8 bytes.
 	if *(*[8]byte)(x) != *(*[8]byte)(y) {
 		return false
 	}
 	// Check last 8 bytes.
-	x = unsafe.Pointer(uintptr(x) + uintptr(len(a)) - 8)
-	y = unsafe.Pointer(uintptr(y) + uintptr(len(a)) - 8)
+	x = add(x, uintptr(len(a)-8))
+	y = add(y, uintptr(len(a)-8))
 	if *(*[8]byte)(x) != *(*[8]byte)(y) {
 		return false
 	}
 	return true
 }
-func stringPtr(s string) unsafe.Pointer {
-	type stringStruct struct {
-		ptr unsafe.Pointer
-		len int
-	}
-	return (*stringStruct)(unsafe.Pointer(&s)).ptr
-}
 
 //go:linkname runtime_mapaccess1_faststr runtime.mapaccess1_faststr
 func runtime_mapaccess1_faststr(typ *abi.MapType, m *Map, key string) unsafe.Pointer {
@@ -146,10 +141,13 @@
 		return elem, true
 	}
 
+	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
-	hash := StrHash(unsafe.Pointer(&k), m.seed)
+	if memHashAESImplemented && UseAeshash {
+		hash = memHashAES(unsafe.Pointer(unsafe.StringData(key)), m.seed, uintptr(len(key)))
+	} else {
+		hash = memHashFallback(unsafe.Pointer(unsafe.StringData(key)), m.seed, uintptr(len(key)))
+	}
 
 	// Select table.
 	idx := m.directoryIndex(hash)
@@ -273,10 +271,13 @@
 		fatal("concurrent map writes")
 	}
 
+	var hash uintptr
 	// See the related comment in runtime_mapaccess2_fast32
-	// for why we pass local copy of key.
-	k := key
-	hash := StrHash(unsafe.Pointer(&k), m.seed)
+	if memHashAESImplemented && UseAeshash {
+		hash = memHashAES(unsafe.Pointer(unsafe.StringData(key)), m.seed, uintptr(len(key)))
+	} else {
+		hash = memHashFallback(unsafe.Pointer(unsafe.StringData(key)), m.seed, uintptr(len(key)))
+	}
 
 	// Set writing after calling Hasher, since Hasher may panic, in which
 	// case we have not actually done a write.
diff --git a/src/internal/runtime/maps/runtime_hash32.go b/src/internal/runtime/maps/runtime_hash32.go
index 2244c93..72ab13d 100644
--- a/src/internal/runtime/maps/runtime_hash32.go
+++ b/src/internal/runtime/maps/runtime_hash32.go
@@ -11,27 +11,25 @@
 
 import "unsafe"
 
-func memHash32Fallback(p unsafe.Pointer, seed uintptr) uintptr {
+func memHash32Fallback(k uint32, seed uintptr) uintptr {
 	a, b := mix32(uint32(seed), uint32(4^hashkey[0]))
-	t := readUnaligned32(p)
-	a ^= t
-	b ^= t
+	a ^= k
+	b ^= k
 	a, b = mix32(a, b)
 	a, b = mix32(a, b)
 	return uintptr(a ^ b)
 }
 
-func memHash64Fallback(p unsafe.Pointer, seed uintptr) uintptr {
+func memHash64Fallback(k uint64, seed uintptr) uintptr {
 	a, b := mix32(uint32(seed), uint32(8^hashkey[0]))
-	a ^= readUnaligned32(p)
-	b ^= readUnaligned32(add(p, 4))
+	a ^= uint32(k)
+	b ^= uint32(k >> 32)
 	a, b = mix32(a, b)
 	a, b = mix32(a, b)
 	return uintptr(a ^ b)
 }
 
 func memHashFallback(p unsafe.Pointer, seed, s uintptr) uintptr {
-
 	a, b := mix32(uint32(seed), uint32(s^hashkey[0]))
 	if s == 0 {
 		return uintptr(a ^ b)
diff --git a/src/internal/runtime/maps/runtime_hash64.go b/src/internal/runtime/maps/runtime_hash64.go
index 733cecb..0e3a35a 100644
--- a/src/internal/runtime/maps/runtime_hash64.go
+++ b/src/internal/runtime/maps/runtime_hash64.go
@@ -64,13 +64,13 @@
 	return mix(m5^s, mix(a^hashkey[1], b^seed))
 }
 
-func memHash32Fallback(p unsafe.Pointer, seed uintptr) uintptr {
-	a := r4(p)
+func memHash32Fallback(k uint32, seed uintptr) uintptr {
+	a := uintptr(k)
 	return mix(m5^4, mix(a^hashkey[1], a^seed^hashkey[0]))
 }
 
-func memHash64Fallback(p unsafe.Pointer, seed uintptr) uintptr {
-	a := r8(p)
+func memHash64Fallback(k uint64, seed uintptr) uintptr {
+	a := uintptr(k)
 	return mix(m5^8, mix(a^hashkey[1], a^seed^hashkey[0]))
 }
 
diff --git a/src/runtime/alg.go b/src/runtime/alg.go
index e8fd6f6..ff1ec3e 100644
--- a/src/runtime/alg.go
+++ b/src/runtime/alg.go
@@ -84,12 +84,12 @@
 
 //go:nosplit
 func memhash64(p unsafe.Pointer, seed uintptr) uintptr {
-	return maps.MemHash64(p, seed)
+	return maps.MemHash64(*(*uint64)(p), seed)
 }
 
 //go:nosplit
 func memhash32(p unsafe.Pointer, seed uintptr) uintptr {
-	return maps.MemHash32(p, seed)
+	return maps.MemHash32(*(*uint32)(p), seed)
 }
 
 // strhash should be an internal detail,
@@ -104,9 +104,10 @@
 // Do not remove or change the type signature.
 // See go.dev/issue/67401.
 //
+//go:nosplit
 //go:linkname strhash
 func strhash(p unsafe.Pointer, h uintptr) uintptr {
-	return maps.StrHash(p, h)
+	return maps.StrHash(*(*string)(p), h)
 }
 
 // NOTE: Because NaN != NaN, a map can contain any

Change information

Files:
  • M src/internal/runtime/maps/memhash_386.s
  • M src/internal/runtime/maps/memhash_aes.go
  • M src/internal/runtime/maps/memhash_aes_asm.go
  • M src/internal/runtime/maps/memhash_aes_simd.go
  • M src/internal/runtime/maps/memhash_amd64.s
  • M src/internal/runtime/maps/memhash_arm64.s
  • M src/internal/runtime/maps/memhash_noaes.go
  • M src/internal/runtime/maps/memhash_nosimd_amd64.s
  • M src/internal/runtime/maps/runtime_alg.go
  • M src/internal/runtime/maps/runtime_fast32.go
  • M src/internal/runtime/maps/runtime_fast64.go
  • M src/internal/runtime/maps/runtime_faststr.go
  • M src/internal/runtime/maps/runtime_hash32.go
  • M src/internal/runtime/maps/runtime_hash64.go
  • M src/runtime/alg.go
Change size: L
Delta: 15 files changed, 107 insertions(+), 179 deletions(-)
Open in Gerrit

Related details

Attention set is empty
Submit Requirements:
  • requirement is not satisfiedCode-Review
  • requirement satisfiedNo-Unresolved-Comments
  • requirement is not satisfiedReview-Enforcement
  • requirement is not satisfiedTryBots-Pass
Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
Gerrit-MessageType: newchange
Gerrit-Project: go
Gerrit-Branch: master
Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
Gerrit-Change-Number: 776281
Gerrit-PatchSet: 1
Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
unsatisfied_requirement
satisfied_requirement
open
diffy

Arseny Samoylov (Gerrit)

unread,
May 9, 2026, 10:24:40 AM (10 days ago) May 9
to goph...@pubsubhelper.golang.org, Keith Randall, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
Attention needed from Cherry Mui, Keith Randall and Michael Pratt

Arseny Samoylov added 1 comment

Patchset-level comments
File-level comment, Patchset 1 (Latest):
Arseny Samoylov . resolved

For CL 753740, I moved memhash functions to the maps package (done in CL 770581). This allows us to freely change the API of these hashing functions to take values directly (since we now have compatibility wrappers in runtime that we can update as needed)

With this change, all things that we disscussed in [issue 77892](https://github.com/golang/go/issues/77892) are solved:

  • hashing call devirtualization - CL 750580
  • inlianable hashing - CL 753740 (under `GOEXPERIMENT=simd`)
  • removed redundant store/loads of keys passed to hashing - this CL

With that, my work on hasing fucntions are done. The next thing that I can work on is implementing `memhash{32,64}` for arm64 once the appropriate intrinsics become avalaible =).

Open in Gerrit

Related details

Attention is currently required from:
  • Cherry Mui
  • Keith Randall
  • Michael Pratt
Submit Requirements:
  • requirement is not satisfiedCode-Review
  • requirement satisfiedNo-Unresolved-Comments
  • requirement is not satisfiedReview-Enforcement
  • requirement is not satisfiedTryBots-Pass
Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
Gerrit-MessageType: comment
Gerrit-Project: go
Gerrit-Branch: master
Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
Gerrit-Change-Number: 776281
Gerrit-PatchSet: 1
Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Reviewer: Cherry Mui <cher...@google.com>
Gerrit-Reviewer: Keith Randall <k...@golang.org>
Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
Gerrit-Attention: Keith Randall <k...@golang.org>
Gerrit-Attention: Cherry Mui <cher...@google.com>
Gerrit-Attention: Michael Pratt <mpr...@google.com>
Gerrit-Comment-Date: Sat, 09 May 2026 14:24:30 +0000
Gerrit-HasComments: Yes
Gerrit-Has-Labels: No
unsatisfied_requirement
satisfied_requirement
open
diffy

David Chase (Gerrit)

unread,
May 10, 2026, 11:16:08 AM (9 days ago) May 10
to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
Attention needed from Arseny Samoylov, Cherry Mui, Keith Randall and Michael Pratt

David Chase voted Commit-Queue+1

Commit-Queue+1
Open in Gerrit

Related details

Attention is currently required from:
  • Arseny Samoylov
  • Cherry Mui
  • Keith Randall
  • Michael Pratt
Submit Requirements:
  • requirement is not satisfiedCode-Review
  • requirement satisfiedNo-Unresolved-Comments
  • requirement is not satisfiedReview-Enforcement
  • requirement is not satisfiedTryBots-Pass
Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
Gerrit-MessageType: comment
Gerrit-Project: go
Gerrit-Branch: master
Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
Gerrit-Change-Number: 776281
Gerrit-PatchSet: 1
Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Reviewer: Cherry Mui <cher...@google.com>
Gerrit-Reviewer: David Chase <drc...@google.com>
Gerrit-Reviewer: Keith Randall <k...@golang.org>
Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Attention: Keith Randall <k...@golang.org>
Gerrit-Attention: Cherry Mui <cher...@google.com>
Gerrit-Attention: Michael Pratt <mpr...@google.com>
Gerrit-Comment-Date: Sun, 10 May 2026 15:16:05 +0000
Gerrit-HasComments: No
Gerrit-Has-Labels: Yes
unsatisfied_requirement
satisfied_requirement
open
diffy

Arseny Samoylov (Gerrit)

unread,
May 10, 2026, 12:41:26 PM (9 days ago) May 10
to goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com
Attention needed from Arseny Samoylov, Cherry Mui, David Chase, Keith Randall and Michael Pratt

Arseny Samoylov uploaded new patchset

Arseny Samoylov uploaded patch set #2 to this change.
Following approvals got outdated and were removed:
Open in Gerrit

Related details

Attention is currently required from:
  • Arseny Samoylov
  • Cherry Mui
  • David Chase
  • Keith Randall
  • Michael Pratt
Submit Requirements:
  • requirement is not satisfiedCode-Review
  • requirement satisfiedNo-Unresolved-Comments
  • requirement is not satisfiedReview-Enforcement
  • requirement is not satisfiedTryBots-Pass
Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
Gerrit-MessageType: newpatchset
Gerrit-Project: go
Gerrit-Branch: master
Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
Gerrit-Change-Number: 776281
Gerrit-PatchSet: 2
Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Reviewer: Cherry Mui <cher...@google.com>
Gerrit-Reviewer: David Chase <drc...@google.com>
Gerrit-Reviewer: Keith Randall <k...@golang.org>
Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Attention: Keith Randall <k...@golang.org>
Gerrit-Attention: Cherry Mui <cher...@google.com>
Gerrit-Attention: David Chase <drc...@google.com>
Gerrit-Attention: Michael Pratt <mpr...@google.com>
unsatisfied_requirement
satisfied_requirement
open
diffy

Arseny Samoylov (Gerrit)

unread,
May 10, 2026, 12:47:10 PM (9 days ago) May 10
to goph...@pubsubhelper.golang.org, golang...@luci-project-accounts.iam.gserviceaccount.com, David Chase, Keith Randall, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
Attention needed from Cherry Mui, David Chase, Keith Randall and Michael Pratt

Arseny Samoylov added 1 comment

Patchset-level comments
File-level comment, Patchset 2 (Latest):
Arseny Samoylov . resolved

Thank you for starting CI!

I forgot to update asm declaration for `GOARCH=386`.

Interestingly, the way I tested it before, it passed silently:
```
GOARCH=386 ../bin/go test runtime -run=Map
ok runtime 12.667s

```

But running

```
GOARCH=386 ../bin/go test internal/runtime/maps/
```

Correctly tells about the issue:
```
# internal/runtime/maps
# [internal/runtime/maps]
internal/runtime/maps/memhash_386.s:26:1: [386] memHash64AES: invalid offset ret+8(FP); expected ret+12(FP)
```

So my earlier test path didn’t catch the assembly mismatch.

Open in Gerrit

Related details

Attention is currently required from:
  • Cherry Mui
  • David Chase
  • Keith Randall
  • Michael Pratt
Submit Requirements:
  • requirement is not satisfiedCode-Review
  • requirement satisfiedNo-Unresolved-Comments
  • requirement is not satisfiedReview-Enforcement
  • requirement is not satisfiedTryBots-Pass
Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
Gerrit-MessageType: comment
Gerrit-Project: go
Gerrit-Branch: master
Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
Gerrit-Change-Number: 776281
Gerrit-PatchSet: 2
Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Reviewer: Cherry Mui <cher...@google.com>
Gerrit-Reviewer: David Chase <drc...@google.com>
Gerrit-Reviewer: Keith Randall <k...@golang.org>
Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
Gerrit-Attention: Keith Randall <k...@golang.org>
Gerrit-Attention: Cherry Mui <cher...@google.com>
Gerrit-Attention: David Chase <drc...@google.com>
Gerrit-Attention: Michael Pratt <mpr...@google.com>
Gerrit-Comment-Date: Sun, 10 May 2026 16:47:01 +0000
Gerrit-HasComments: Yes
Gerrit-Has-Labels: No
unsatisfied_requirement
satisfied_requirement
open
diffy

Alexander Musman (Gerrit)

unread,
May 12, 2026, 4:17:42 AM (7 days ago) May 12
to Arseny Samoylov, goph...@pubsubhelper.golang.org, golang...@luci-project-accounts.iam.gserviceaccount.com, David Chase, Keith Randall, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
Attention needed from Arseny Samoylov, Cherry Mui, David Chase, Keith Randall and Michael Pratt

Alexander Musman voted Commit-Queue+1

Commit-Queue+1
Open in Gerrit

Related details

Attention is currently required from:
  • Arseny Samoylov
  • Cherry Mui
  • David Chase
  • Keith Randall
  • Michael Pratt
Submit Requirements:
  • requirement is not satisfiedCode-Review
  • requirement satisfiedNo-Unresolved-Comments
  • requirement is not satisfiedReview-Enforcement
  • requirement is not satisfiedTryBots-Pass
Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
Gerrit-MessageType: comment
Gerrit-Project: go
Gerrit-Branch: master
Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
Gerrit-Change-Number: 776281
Gerrit-PatchSet: 2
Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
Gerrit-Reviewer: Cherry Mui <cher...@google.com>
Gerrit-Reviewer: David Chase <drc...@google.com>
Gerrit-Reviewer: Keith Randall <k...@golang.org>
Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
Gerrit-Attention: Keith Randall <k...@golang.org>
Gerrit-Attention: Cherry Mui <cher...@google.com>
Gerrit-Attention: David Chase <drc...@google.com>
Gerrit-Attention: Michael Pratt <mpr...@google.com>
Gerrit-Comment-Date: Tue, 12 May 2026 08:17:34 +0000
Gerrit-HasComments: No
Gerrit-Has-Labels: Yes
unsatisfied_requirement
satisfied_requirement
open
diffy

Arseny Samoylov (Gerrit)

unread,
May 15, 2026, 5:18:32 AM (4 days ago) May 15
to goph...@pubsubhelper.golang.org, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, David Chase, Keith Randall, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
Attention needed from Cherry Mui, David Chase, Keith Randall and Michael Pratt

Arseny Samoylov added 1 comment

Patchset-level comments
Arseny Samoylov . resolved
Open in Gerrit

Related details

Attention is currently required from:
  • Cherry Mui
  • David Chase
  • Keith Randall
  • Michael Pratt
Submit Requirements:
    • requirement is not satisfiedCode-Review
    • requirement satisfiedNo-Unresolved-Comments
    • requirement is not satisfiedReview-Enforcement
    • requirement satisfiedTryBots-Pass
    Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
    Gerrit-MessageType: comment
    Gerrit-Project: go
    Gerrit-Branch: master
    Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
    Gerrit-Change-Number: 776281
    Gerrit-PatchSet: 2
    Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
    Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
    Gerrit-Reviewer: Cherry Mui <cher...@google.com>
    Gerrit-Reviewer: David Chase <drc...@google.com>
    Gerrit-Reviewer: Keith Randall <k...@golang.org>
    Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
    Gerrit-Attention: Keith Randall <k...@golang.org>
    Gerrit-Attention: Cherry Mui <cher...@google.com>
    Gerrit-Attention: David Chase <drc...@google.com>
    Gerrit-Attention: Michael Pratt <mpr...@google.com>
    Gerrit-Comment-Date: Fri, 15 May 2026 09:18:25 +0000
    Gerrit-HasComments: Yes
    Gerrit-Has-Labels: No
    unsatisfied_requirement
    satisfied_requirement
    open
    diffy

    Keith Randall (Gerrit)

    unread,
    May 15, 2026, 12:45:02 PM (4 days ago) May 15
    to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
    Attention needed from Arseny Samoylov, Cherry Mui, David Chase and Michael Pratt

    Keith Randall voted and added 4 comments

    Votes added by Keith Randall

    Code-Review+2

    4 comments

    File src/internal/runtime/maps/memhash_386.s
    Line 40, Patchset 2 (Latest): MOVL CX, X0 // 32 bits of per-table hash seed
    Keith Randall . unresolved

    This could load directly from h+4(FP), I think.

    File src/internal/runtime/maps/runtime_fast32.go
    Line 352, Patchset 2 (Latest): hash = memHash32AES(*(*uint32)((unsafe.Pointer)(&key)), m.seed)
    Keith Randall . unresolved

    Can we just do `uint32(uintptr(key))`?

    File src/internal/runtime/maps/runtime_fast64.go
    Line 416, Patchset 2 (Latest): hash = memHash64AES(*(*uint64)((unsafe.Pointer)(&key)), m.seed)
    Keith Randall . unresolved

    same for both of these.

    File src/runtime/alg.go
    Line 87, Patchset 2 (Latest): return maps.MemHash64(*(*uint64)(p), seed)
    Keith Randall . unresolved

    Do we need to use readUnaligned here?
    I guess the other possibility would be to require alignment for use of memhash64. Which I think would be in cmd/compile/internal/reflectdata/alg.go:genHashSig, to do what genEqSig does.

    Might be good to have a test with a key like:
    ```
    type Key struct {
        _ [1]byte
        k [8]byte
        _ [7]byte
    }
    ```
    That would cause unaligned memhash64 calls, I think.
    Open in Gerrit

    Related details

    Attention is currently required from:
    • Arseny Samoylov
    • Cherry Mui
    • David Chase
    • Michael Pratt
    Submit Requirements:
    • requirement satisfiedCode-Review
    • requirement is not satisfiedNo-Unresolved-Comments
    • requirement is not satisfiedReview-Enforcement
    • requirement satisfiedTryBots-Pass
    Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
    Gerrit-MessageType: comment
    Gerrit-Project: go
    Gerrit-Branch: master
    Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
    Gerrit-Change-Number: 776281
    Gerrit-PatchSet: 2
    Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
    Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
    Gerrit-Reviewer: Cherry Mui <cher...@google.com>
    Gerrit-Reviewer: David Chase <drc...@google.com>
    Gerrit-Reviewer: Keith Randall <k...@golang.org>
    Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
    Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
    Gerrit-Attention: Cherry Mui <cher...@google.com>
    Gerrit-Attention: David Chase <drc...@google.com>
    Gerrit-Attention: Michael Pratt <mpr...@google.com>
    Gerrit-Comment-Date: Fri, 15 May 2026 16:44:54 +0000
    Gerrit-HasComments: Yes
    Gerrit-Has-Labels: Yes
    satisfied_requirement
    unsatisfied_requirement
    open
    diffy

    Keith Randall (Gerrit)

    unread,
    May 15, 2026, 12:45:16 PM (4 days ago) May 15
    to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
    Attention needed from Arseny Samoylov, Cherry Mui, David Chase and Michael Pratt

    Keith Randall voted Code-Review+1

    Code-Review+1
    Gerrit-Reviewer: Keith Randall <k...@google.com>
    Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
    Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
    Gerrit-Attention: Cherry Mui <cher...@google.com>
    Gerrit-Attention: David Chase <drc...@google.com>
    Gerrit-Attention: Michael Pratt <mpr...@google.com>
    Gerrit-Comment-Date: Fri, 15 May 2026 16:45:09 +0000
    Gerrit-HasComments: No
    Gerrit-Has-Labels: Yes
    satisfied_requirement
    unsatisfied_requirement
    open
    diffy

    Arseny Samoylov (Gerrit)

    unread,
    May 15, 2026, 4:09:33 PM (4 days ago) May 15
    to goph...@pubsubhelper.golang.org, golang-co...@googlegroups.com
    Attention needed from Arseny Samoylov, Cherry Mui, David Chase, Keith Randall, Keith Randall and Michael Pratt

    Arseny Samoylov uploaded new patchset

    Arseny Samoylov uploaded patch set #3 to this change.
    Following approvals got outdated and were removed:
    Open in Gerrit

    Related details

    Attention is currently required from:
    • Arseny Samoylov
    • Cherry Mui
    • David Chase
    • Keith Randall
    • Keith Randall
    • Michael Pratt
    Submit Requirements:
    • requirement is not satisfiedCode-Review
    • requirement is not satisfiedNo-Unresolved-Comments
    • requirement is not satisfiedReview-Enforcement
    • requirement is not satisfiedTryBots-Pass
    Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
    Gerrit-MessageType: newpatchset
    Gerrit-Project: go
    Gerrit-Branch: master
    Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
    Gerrit-Change-Number: 776281
    Gerrit-PatchSet: 3
    Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
    Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
    Gerrit-Reviewer: Cherry Mui <cher...@google.com>
    Gerrit-Reviewer: David Chase <drc...@google.com>
    Gerrit-Reviewer: Keith Randall <k...@golang.org>
    Gerrit-Reviewer: Keith Randall <k...@google.com>
    Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
    Gerrit-Attention: Keith Randall <k...@golang.org>
    Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
    Gerrit-Attention: Cherry Mui <cher...@google.com>
    Gerrit-Attention: David Chase <drc...@google.com>
    Gerrit-Attention: Michael Pratt <mpr...@google.com>
    Gerrit-Attention: Keith Randall <k...@google.com>
    unsatisfied_requirement
    open
    diffy

    Arseny Samoylov (Gerrit)

    unread,
    May 15, 2026, 4:16:27 PM (4 days ago) May 15
    to goph...@pubsubhelper.golang.org, Keith Randall, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
    Attention needed from Cherry Mui, David Chase, Keith Randall, Keith Randall and Michael Pratt

    Arseny Samoylov added 4 comments

    File src/internal/runtime/maps/memhash_386.s
    Line 40, Patchset 2: MOVL CX, X0 // 32 bits of per-table hash seed
    Keith Randall . resolved

    This could load directly from h+4(FP), I think.

    Arseny Samoylov

    Done

    File src/internal/runtime/maps/runtime_fast32.go
    Line 352, Patchset 2: hash = memHash32AES(*(*uint32)((unsafe.Pointer)(&key)), m.seed)
    Keith Randall . resolved

    Can we just do `uint32(uintptr(key))`?

    Arseny Samoylov

    Done

    File src/internal/runtime/maps/runtime_fast64.go
    Line 416, Patchset 2: hash = memHash64AES(*(*uint64)((unsafe.Pointer)(&key)), m.seed)
    Keith Randall . resolved

    same for both of these.

    Arseny Samoylov

    Done

    File src/runtime/alg.go
    Line 87, Patchset 2: return maps.MemHash64(*(*uint64)(p), seed)
    Keith Randall . resolved

    Do we need to use readUnaligned here?
    I guess the other possibility would be to require alignment for use of memhash64. Which I think would be in cmd/compile/internal/reflectdata/alg.go:genHashSig, to do what genEqSig does.

    Might be good to have a test with a key like:
    ```
    type Key struct {
        _ [1]byte
        k [8]byte
        _ [7]byte
    }
    ```
    That would cause unaligned memhash64 calls, I think.
    Arseny Samoylov

    Do we need to use readUnaligned here?

    Yes - you are spot on, thanks!

    For most arches (everything except arm64,amd64,386), AES hashing is not implemented, so before my changes it would redirect to the fallback, which already used readUnaligned. I missed that.


    > I guess the other possibility would be to require alignment for use of memhash64.

    I don't think that we should enforce the alignment here, especially since the `memcobine` pass optimizes `readUnalgined` to single mov.


    > Might be good to have a test with a key

    Done

    Open in Gerrit

    Related details

    Attention is currently required from:
    • Cherry Mui
    • David Chase
    • Keith Randall
    • Keith Randall
    • Michael Pratt
    Submit Requirements:
      • requirement is not satisfiedCode-Review
      • requirement satisfiedNo-Unresolved-Comments
      • requirement is not satisfiedReview-Enforcement
      • requirement is not satisfiedTryBots-Pass
      Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
      Gerrit-MessageType: comment
      Gerrit-Project: go
      Gerrit-Branch: master
      Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
      Gerrit-Change-Number: 776281
      Gerrit-PatchSet: 3
      Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
      Gerrit-Reviewer: Cherry Mui <cher...@google.com>
      Gerrit-Reviewer: David Chase <drc...@google.com>
      Gerrit-Reviewer: Keith Randall <k...@golang.org>
      Gerrit-Reviewer: Keith Randall <k...@google.com>
      Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Keith Randall <k...@golang.org>
      Gerrit-Attention: Cherry Mui <cher...@google.com>
      Gerrit-Attention: David Chase <drc...@google.com>
      Gerrit-Attention: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Keith Randall <k...@google.com>
      Gerrit-Comment-Date: Fri, 15 May 2026 20:16:18 +0000
      Gerrit-HasComments: Yes
      Gerrit-Has-Labels: No
      Comment-In-Reply-To: Keith Randall <k...@golang.org>
      unsatisfied_requirement
      satisfied_requirement
      open
      diffy

      Alexander Musman (Gerrit)

      unread,
      May 18, 2026, 4:37:31 AM (yesterday) May 18
      to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
      Attention needed from Arseny Samoylov, Cherry Mui, David Chase, Keith Randall, Keith Randall and Michael Pratt

      Alexander Musman voted Commit-Queue+1

      Commit-Queue+1
      Open in Gerrit

      Related details

      Attention is currently required from:
      • Arseny Samoylov
      Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Attention: Keith Randall <k...@golang.org>
      Gerrit-Attention: Cherry Mui <cher...@google.com>
      Gerrit-Attention: David Chase <drc...@google.com>
      Gerrit-Attention: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Keith Randall <k...@google.com>
      Gerrit-Comment-Date: Mon, 18 May 2026 08:37:23 +0000
      Gerrit-HasComments: No
      Gerrit-Has-Labels: Yes
      unsatisfied_requirement
      satisfied_requirement
      open
      diffy

      Arseny Samoylov (Gerrit)

      unread,
      May 18, 2026, 6:01:26 AM (yesterday) May 18
      to goph...@pubsubhelper.golang.org, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, Keith Randall, Keith Randall, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
      Attention needed from Alexander Musman, Cherry Mui, David Chase, Keith Randall, Keith Randall and Michael Pratt

      Arseny Samoylov added 1 comment

      Patchset-level comments
      File-level comment, Patchset 3 (Latest):
      Arseny Samoylov . resolved

      The CI failure looks like a flaky result. It failed only on single builder (`gotip-windows-amd64`) in http test, without any clear error message.

      @alexande...@gmail.com could you please restart the CI?

      Open in Gerrit

      Related details

      Attention is currently required from:
      • Alexander Musman
      Gerrit-Attention: Keith Randall <k...@golang.org>
      Gerrit-Attention: Cherry Mui <cher...@google.com>
      Gerrit-Attention: David Chase <drc...@google.com>
      Gerrit-Attention: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Alexander Musman <alexande...@gmail.com>
      Gerrit-Attention: Keith Randall <k...@google.com>
      Gerrit-Comment-Date: Mon, 18 May 2026 10:01:18 +0000
      Gerrit-HasComments: Yes
      Gerrit-Has-Labels: No
      unsatisfied_requirement
      satisfied_requirement
      open
      diffy

      Alexander Musman (Gerrit)

      unread,
      May 18, 2026, 6:04:23 AM (yesterday) May 18
      to Arseny Samoylov, goph...@pubsubhelper.golang.org, golang...@luci-project-accounts.iam.gserviceaccount.com, Keith Randall, Keith Randall, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
      Attention needed from Arseny Samoylov, Cherry Mui, David Chase, Keith Randall, Keith Randall and Michael Pratt

      Alexander Musman voted Commit-Queue+1

      Commit-Queue+1
      Open in Gerrit

      Related details

      Attention is currently required from:
      • Arseny Samoylov
      Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Attention: Keith Randall <k...@golang.org>
      Gerrit-Attention: Cherry Mui <cher...@google.com>
      Gerrit-Attention: David Chase <drc...@google.com>
      Gerrit-Attention: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Keith Randall <k...@google.com>
      Gerrit-Comment-Date: Mon, 18 May 2026 10:04:17 +0000
      Gerrit-HasComments: No
      Gerrit-Has-Labels: Yes
      unsatisfied_requirement
      satisfied_requirement
      open
      diffy

      Keith Randall (Gerrit)

      unread,
      May 18, 2026, 11:21:14 AM (21 hours ago) May 18
      to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, Keith Randall, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
      Attention needed from Arseny Samoylov, Cherry Mui, David Chase, Keith Randall and Michael Pratt

      Keith Randall voted

      Auto-Submit+1
      Code-Review+2
      Open in Gerrit

      Related details

      Attention is currently required from:
      • Arseny Samoylov
      • Cherry Mui
      • David Chase
      • Keith Randall
      • Michael Pratt
      Submit Requirements:
      • requirement satisfiedCode-Review
      • requirement satisfiedNo-Unresolved-Comments
      • requirement is not satisfiedReview-Enforcement
      • requirement satisfiedTryBots-Pass
      Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
      Gerrit-MessageType: comment
      Gerrit-Project: go
      Gerrit-Branch: master
      Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
      Gerrit-Change-Number: 776281
      Gerrit-PatchSet: 3
      Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
      Gerrit-Reviewer: Cherry Mui <cher...@google.com>
      Gerrit-Reviewer: David Chase <drc...@google.com>
      Gerrit-Reviewer: Keith Randall <k...@golang.org>
      Gerrit-Reviewer: Keith Randall <k...@google.com>
      Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Attention: Cherry Mui <cher...@google.com>
      Gerrit-Attention: David Chase <drc...@google.com>
      Gerrit-Attention: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Keith Randall <k...@google.com>
      Gerrit-Comment-Date: Mon, 18 May 2026 15:21:09 +0000
      Gerrit-HasComments: No
      Gerrit-Has-Labels: Yes
      satisfied_requirement
      unsatisfied_requirement
      open
      diffy

      Keith Randall (Gerrit)

      unread,
      May 18, 2026, 11:21:33 AM (21 hours ago) May 18
      to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, David Chase, Cherry Mui, Michael Pratt, golang-co...@googlegroups.com
      Attention needed from Arseny Samoylov, Cherry Mui, David Chase and Michael Pratt

      Keith Randall voted Code-Review+1

      Code-Review+1
      Open in Gerrit

      Related details

      Attention is currently required from:
      • Arseny Samoylov
      • Cherry Mui
      • David Chase
      • Michael Pratt
      Gerrit-Comment-Date: Mon, 18 May 2026 15:21:29 +0000
      Gerrit-HasComments: No
      Gerrit-Has-Labels: Yes
      satisfied_requirement
      unsatisfied_requirement
      open
      diffy

      Cherry Mui (Gerrit)

      unread,
      May 18, 2026, 10:46:02 PM (9 hours ago) May 18
      to Arseny Samoylov, goph...@pubsubhelper.golang.org, Keith Randall, Keith Randall, golang...@luci-project-accounts.iam.gserviceaccount.com, Alexander Musman, David Chase, Michael Pratt, golang-co...@googlegroups.com
      Attention needed from Arseny Samoylov, David Chase and Michael Pratt

      Cherry Mui voted and added 4 comments

      Votes added by Cherry Mui

      Code-Review+2

      4 comments

      Patchset-level comments
      Cherry Mui . resolved

      Thanks.

      Commit Message
      Line 7, Patchset 3 (Latest):internal/runtime/maps,runtime/: take key by value in memHash{32,64}AES; replace strHashAes by call to memHashAES
      Cherry Mui . unresolved

      Would be good to explain the reason for the change in the CL desciption.

      File src/internal/runtime/maps/memhash_386.s
      Line 11, Patchset 3 (Latest): MOVL h+4(FP), X0 // seed
      Cherry Mui . unresolved

      Comment is not aligned well. Maybe use only spaces, not tab, for aligning.

      File src/internal/runtime/maps/runtime_fast32.go
      Line 352, Patchset 3 (Latest): hash = memHash32AES((uint32)((uintptr)(key)), m.seed)
      Cherry Mui . unresolved

      Parenthese around the type are not needed. `uint32(uintptr(key))`

      Also other places.

      Open in Gerrit

      Related details

      Attention is currently required from:
      • Arseny Samoylov
      • David Chase
      • Michael Pratt
      Submit Requirements:
      • requirement satisfiedCode-Review
      • requirement is not satisfiedNo-Unresolved-Comments
      • requirement satisfiedReview-Enforcement
      • requirement satisfiedTryBots-Pass
      Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. DiffyGerrit
      Gerrit-MessageType: comment
      Gerrit-Project: go
      Gerrit-Branch: master
      Gerrit-Change-Id: I498ae9f437ad089df4298aa85f04a16c2ed6a5f5
      Gerrit-Change-Number: 776281
      Gerrit-PatchSet: 3
      Gerrit-Owner: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Reviewer: Alexander Musman <alexande...@gmail.com>
      Gerrit-Reviewer: Cherry Mui <cher...@google.com>
      Gerrit-Reviewer: David Chase <drc...@google.com>
      Gerrit-Reviewer: Keith Randall <k...@golang.org>
      Gerrit-Reviewer: Keith Randall <k...@google.com>
      Gerrit-Reviewer: Michael Pratt <mpr...@google.com>
      Gerrit-Attention: Arseny Samoylov <samoylo...@gmail.com>
      Gerrit-Attention: David Chase <drc...@google.com>
      Gerrit-Attention: Michael Pratt <mpr...@google.com>
      Gerrit-Comment-Date: Tue, 19 May 2026 02:45:59 +0000
      Gerrit-HasComments: Yes
      Gerrit-Has-Labels: Yes
      satisfied_requirement
      unsatisfied_requirement
      open
      diffy
      Reply all
      Reply to author
      Forward
      0 new messages