runtime: implement getcallersp in Go

This makes it possible to inline getcallersp. getcallersp is on the
hot path of defers, so this slightly speeds up defer:

name           old time/op  new time/op  delta
Defer-4        78.3ns ± 2%  75.1ns ± 1%  -4.00%   (p=0.000 n=9+8)

Updates #14939.

Change-Id: Icc1cc4cd2f0a81fc4c8344432d0b2e783accacdd
Reviewed-on: https://go-review.googlesource.com/29655
TryBot-Result: Gobot Gobot <gobot@golang.org>
Run-TryBot: Austin Clements <austin@google.com>
Reviewed-by: David Crawshaw <crawshaw@golang.org>
Reviewed-by: Keith Randall <khr@golang.org>

9 files changed, 8 insertions, 48 deletions

diff --git a/src/runtime/asm_386.s b/src/runtime/asm_386.s
index 12038220ac..56d495aede 100644
--- a/src/runtime/asm_386.s
+++ b/src/runtime/asm_386.s
@@ -808,11 +808,6 @@ setbar:
 	CALL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB), NOSPLIT, $0-8
-	MOVL	argp+0(FP), AX
-	MOVL	AX, ret+4(FP)
-	RET
-
 // func cputicks() int64
 TEXT runtime·cputicks(SB),NOSPLIT,$0-8
 	TESTL	$0x4000000, runtime·cpuid_edx(SB) // no sse2, no mfence
diff --git a/src/runtime/asm_amd64.s b/src/runtime/asm_amd64.s
index 488c34a233..8d992188de 100644
--- a/src/runtime/asm_amd64.s
+++ b/src/runtime/asm_amd64.s
@@ -825,11 +825,6 @@ setbar:
 	CALL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$0-16
-	MOVQ	argp+0(FP), AX
-	MOVQ	AX, ret+8(FP)
-	RET
-
 // func cputicks() int64
 TEXT runtime·cputicks(SB),NOSPLIT,$0-0
 	CMPB	runtime·lfenceBeforeRdtsc(SB), $1
diff --git a/src/runtime/asm_amd64p32.s b/src/runtime/asm_amd64p32.s
index 6aa230841e..0b42c666ae 100644
--- a/src/runtime/asm_amd64p32.s
+++ b/src/runtime/asm_amd64p32.s
@@ -521,11 +521,6 @@ setbar:
 	CALL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$0-12
-	MOVL	argp+0(FP), AX
-	MOVL	AX, ret+8(FP)
-	RET
-
 // int64 runtime·cputicks(void)
 TEXT runtime·cputicks(SB),NOSPLIT,$0-0
 	RDTSC
diff --git a/src/runtime/asm_arm.s b/src/runtime/asm_arm.s
index 29f39cf15d..d768060af2 100644
--- a/src/runtime/asm_arm.s
+++ b/src/runtime/asm_arm.s
@@ -695,12 +695,6 @@ setbar:
 	BL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$-4-8
-	MOVW	argp+0(FP), R0
-	MOVW	$-4(R0), R0
-	MOVW	R0, ret+4(FP)
-	RET
-
 TEXT runtime·emptyfunc(SB),0,$0-0
 	RET
 
diff --git a/src/runtime/asm_arm64.s b/src/runtime/asm_arm64.s
index 066b534900..c46569f68c 100644
--- a/src/runtime/asm_arm64.s
+++ b/src/runtime/asm_arm64.s
@@ -731,12 +731,6 @@ setbar:
 	BL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$0-16
-	MOVD	argp+0(FP), R0
-	SUB	$8, R0
-	MOVD	R0, ret+8(FP)
-	RET
-
 TEXT runtime·abort(SB),NOSPLIT,$-8-0
 	B	(ZR)
 	UNDEF
diff --git a/src/runtime/asm_mips64x.s b/src/runtime/asm_mips64x.s
index 15105b90c3..138181833c 100644
--- a/src/runtime/asm_mips64x.s
+++ b/src/runtime/asm_mips64x.s
@@ -643,12 +643,6 @@ setbar:
 	JAL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$0-16
-	MOVV	argp+0(FP), R1
-	ADDV	$-8, R1
-	MOVV	R1, ret+8(FP)
-	RET
-
 TEXT runtime·abort(SB),NOSPLIT,$-8-0
 	MOVW	(R0), R0
 	UNDEF
diff --git a/src/runtime/asm_ppc64x.s b/src/runtime/asm_ppc64x.s
index 67b3d50691..b5cd12bb3c 100644
--- a/src/runtime/asm_ppc64x.s
+++ b/src/runtime/asm_ppc64x.s
@@ -742,12 +742,6 @@ setbar:
 	BL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$0-16
-	MOVD	argp+0(FP), R3
-	SUB	$FIXED_FRAME, R3
-	MOVD	R3, ret+8(FP)
-	RET
-
 TEXT runtime·abort(SB),NOSPLIT|NOFRAME,$0-0
 	MOVW	(R0), R0
 	UNDEF
diff --git a/src/runtime/asm_s390x.s b/src/runtime/asm_s390x.s
index 3fe224df37..6d0533a3dc 100644
--- a/src/runtime/asm_s390x.s
+++ b/src/runtime/asm_s390x.s
@@ -715,12 +715,6 @@ setbar:
 	BL	runtime·setNextBarrierPC(SB)
 	RET
 
-TEXT runtime·getcallersp(SB),NOSPLIT,$0-16
-	MOVD	argp+0(FP), R3
-	SUB	$8, R3
-	MOVD	R3, ret+8(FP)
-	RET
-
 TEXT runtime·abort(SB),NOSPLIT|NOFRAME,$0-0
 	MOVW	(R0), R0
 	UNDEF
diff --git a/src/runtime/stubs.go b/src/runtime/stubs.go
index f6bb2fba16..88f4139ba3 100644
--- a/src/runtime/stubs.go
+++ b/src/runtime/stubs.go
@@ -4,7 +4,10 @@
 
 package runtime
 
-import "unsafe"
+import (
+	"runtime/internal/sys"
+	"unsafe"
+)
 
 // Should be a built-in for unsafe.Pointer?
 //go:nosplit
@@ -196,8 +199,10 @@ func setcallerpc(argp unsafe.Pointer, pc uintptr)
 //go:noescape
 func getcallerpc(argp unsafe.Pointer) uintptr
 
-//go:noescape
-func getcallersp(argp unsafe.Pointer) uintptr
+//go:nosplit
+func getcallersp(argp unsafe.Pointer) uintptr {
+	return uintptr(argp) - sys.MinFrameSize
+}
 
 //go:noescape
 func asmcgocall(fn, arg unsafe.Pointer) int32