runtime/internal/atomic: add arm/arm64 operators for And/Or

This CL continues adding support for And/Or primitives to
more architectures, this time for arm/arm64.

For #61395

Change-Id: Icc44ea65884c825698a345299d8f9511392aceb6
GitHub-Last-Rev: 8267665a03
GitHub-Pull-Request: golang/go#62674
Reviewed-on: https://go-review.googlesource.com/c/go/+/528797
Reviewed-by: Matthew Dempsky <mdempsky@google.com>
Reviewed-by: Keith Randall <khr@golang.org>
Run-TryBot: Mauri de Souza Meneguzzo <mauri870@gmail.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Keith Randall <khr@google.com>

src/runtime/internal/atomic/atomic_andor_test.go

@@ -1,4 +1,4 @@
-//go:build 386 || amd64 || ppc64 || ppc64le || riscv64 || wasm
+//go:build 386 || amd64 || arm || arm64 || ppc64 || ppc64le || riscv64 || wasm
 
 //
 // Copyright 2023 The Go Authors. All rights reserved.

src/runtime/internal/atomic/atomic_arm.go

@@ -208,6 +208,66 @@ func And(addr *uint32, v uint32) {
 	}
 }
 
+//go:nosplit
+func Or32(addr *uint32, v uint32) uint32 {
+	for {
+		old := *addr
+		if Cas(addr, old, old|v) {
+			return old
+		}
+	}
+}
+
+//go:nosplit
+func And32(addr *uint32, v uint32) uint32 {
+	for {
+		old := *addr
+		if Cas(addr, old, old&v) {
+			return old
+		}
+	}
+}
+
+//go:nosplit
+func Or64(addr *uint64, v uint64) uint64 {
+	for {
+		old := *addr
+		if Cas64(addr, old, old|v) {
+			return old
+		}
+	}
+}
+
+//go:nosplit
+func And64(addr *uint64, v uint64) uint64 {
+	for {
+		old := *addr
+		if Cas64(addr, old, old&v) {
+			return old
+		}
+	}
+}
+
+//go:nosplit
+func Oruintptr(addr *uintptr, v uintptr) uintptr {
+	for {
+		old := *addr
+		if Casuintptr(addr, old, old|v) {
+			return old
+		}
+	}
+}
+
+//go:nosplit
+func Anduintptr(addr *uintptr, v uintptr) uintptr {
+	for {
+		old := *addr
+		if Casuintptr(addr, old, old&v) {
+			return old
+		}
+	}
+}
+
 //go:nosplit
 func armcas(ptr *uint32, old, new uint32) bool
 

src/runtime/internal/atomic/atomic_arm64.go

@@ -66,6 +66,24 @@ func And(ptr *uint32, val uint32)
 //go:noescape
 func Or(ptr *uint32, val uint32)
 
+//go:noescape
+func And32(ptr *uint32, val uint32) uint32
+
+//go:noescape
+func Or32(ptr *uint32, val uint32) uint32
+
+//go:noescape
+func And64(ptr *uint64, val uint64) uint64
+
+//go:noescape
+func Or64(ptr *uint64, val uint64) uint64
+
+//go:noescape
+func Anduintptr(ptr *uintptr, val uintptr) uintptr
+
+//go:noescape
+func Oruintptr(ptr *uintptr, val uintptr) uintptr
+
 //go:noescape
 func Cas64(ptr *uint64, old, new uint64) bool
 

src/runtime/internal/atomic/atomic_arm64.s

@@ -331,3 +331,81 @@ load_store_loop:
 	STLXRW	R2, (R0), R3
 	CBNZ	R3, load_store_loop
 	RET
+
+// func Or32(addr *uint32, v uint32) old uint32
+TEXT ·Or32(SB), NOSPLIT, $0-20
+	MOVD	ptr+0(FP), R0
+	MOVW	val+8(FP), R1
+	MOVBU	internal∕cpu·ARM64+const_offsetARM64HasATOMICS(SB), R4
+	CBZ 	R4, load_store_loop
+	LDORALW	R1, (R0), R2
+	MOVD	R2, ret+16(FP)
+	RET
+load_store_loop:
+	LDAXRW	(R0), R2
+	ORR	R1, R2, R3
+	STLXRW	R3, (R0), R4
+	CBNZ	R4, load_store_loop
+	MOVD R2, ret+16(FP)
+	RET
+
+// func And32(addr *uint32, v uint32) old uint32
+TEXT ·And32(SB), NOSPLIT, $0-20
+	MOVD	ptr+0(FP), R0
+	MOVW	val+8(FP), R1
+	MOVBU	internal∕cpu·ARM64+const_offsetARM64HasATOMICS(SB), R4
+	CBZ 	R4, load_store_loop
+	MVN 	R1, R2
+	LDCLRALW	R2, (R0), R3
+	MOVD	R3, ret+16(FP)
+	RET
+load_store_loop:
+	LDAXRW	(R0), R2
+	AND	R1, R2, R3
+	STLXRW	R3, (R0), R4
+	CBNZ	R4, load_store_loop
+	MOVD R2, ret+16(FP)
+	RET
+
+// func Or64(addr *uint64, v uint64) old uint64
+TEXT ·Or64(SB), NOSPLIT, $0-24
+	MOVD	ptr+0(FP), R0
+	MOVD	val+8(FP), R1
+	MOVBU	internal∕cpu·ARM64+const_offsetARM64HasATOMICS(SB), R4
+	CBZ 	R4, load_store_loop
+	LDORALD	R1, (R0), R2
+	MOVD	R2, ret+16(FP)
+	RET
+load_store_loop:
+	LDAXR	(R0), R2
+	ORR	R1, R2, R3
+	STLXR	R3, (R0), R4
+	CBNZ	R4, load_store_loop
+	MOVD 	R2, ret+16(FP)
+	RET
+
+// func And64(addr *uint64, v uint64) old uint64
+TEXT ·And64(SB), NOSPLIT, $0-24
+	MOVD	ptr+0(FP), R0
+	MOVD	val+8(FP), R1
+	MOVBU	internal∕cpu·ARM64+const_offsetARM64HasATOMICS(SB), R4
+	CBZ 	R4, load_store_loop
+	MVN 	R1, R2
+	LDCLRALD	R2, (R0), R3
+	MOVD	R3, ret+16(FP)
+	RET
+load_store_loop:
+	LDAXR	(R0), R2
+	AND	R1, R2, R3
+	STLXR	R3, (R0), R4
+	CBNZ	R4, load_store_loop
+	MOVD 	R2, ret+16(FP)
+	RET
+
+// func Anduintptr(addr *uintptr, v uintptr) old uintptr
+TEXT ·Anduintptr(SB), NOSPLIT, $0-24
+	B	·And64(SB)
+
+// func Oruintptr(addr *uintptr, v uintptr) old uintptr
+TEXT ·Oruintptr(SB), NOSPLIT, $0-24
+	B	·Or64(SB)