From a651e2ea47586f1e29688202ff60c719bac6a05b Mon Sep 17 00:00:00 2001
From: Keith Randall <khr@golang.org>
Date: Tue, 17 Jun 2025 10:56:53 -0700
Subject: [PATCH] runtime: remove duff support for amd64

Change-Id: I742b49a3889892b7b1bb354f47f1c0d933c041e4
Reviewed-on: https://go-review.googlesource.com/c/go/+/682395
Auto-Submit: Keith Randall <khr@golang.org>
Reviewed-by: Jorropo <jorropo.pgm@gmail.com>
Reviewed-by: Keith Randall <khr@google.com>
Reviewed-by: Michael Knyszek <mknyszek@google.com>
LUCI-TryBot-Result: Go LUCI <golang-scoped@luci-project-accounts.iam.gserviceaccount.com>
---
 src/cmd/internal/obj/x86/asm6.go |  25 --
 src/runtime/duff_amd64.s         | 427 -------------------------------
 src/runtime/mkduff.go            |   1 -
 3 files changed, 453 deletions(-)
 delete mode 100644 src/runtime/duff_amd64.s

diff --git a/src/cmd/internal/obj/x86/asm6.go b/src/cmd/internal/obj/x86/asm6.go
index 0906f16eaa..b071bd530d 100644
--- a/src/cmd/internal/obj/x86/asm6.go
+++ b/src/cmd/internal/obj/x86/asm6.go
@@ -4013,15 +4013,6 @@ func (ab *AsmBuf) mediaop(ctxt *obj.Link, o *Optab, op int, osize int, z int) in
 	return z
 }
 
-var bpduff1 = []byte{
-	0x48, 0x89, 0x6c, 0x24, 0xf0, // MOVQ BP, -16(SP)
-	0x48, 0x8d, 0x6c, 0x24, 0xf0, // LEAQ -16(SP), BP
-}
-
-var bpduff2 = []byte{
-	0x48, 0x8b, 0x6d, 0x00, // MOVQ 0(BP), BP
-}
-
 // asmevex emits EVEX pregis and opcode byte.
 // In addition to asmvex r/m, vvvv and reg fields also requires optional
 // K-masking register.
@@ -4859,16 +4850,6 @@ func (ab *AsmBuf) doasm(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog) {
 					ctxt.Diag("directly calling duff when dynamically linking Go")
 				}
 
-				if yt.zcase == Zcallduff && ctxt.Arch.Family == sys.AMD64 {
-					// Maintain BP around call, since duffcopy/duffzero can't do it
-					// (the call jumps into the middle of the function).
-					// This makes it possible to see call sites for duffcopy/duffzero in
-					// BP-based profiling tools like Linux perf (which is the
-					// whole point of maintaining frame pointers in Go).
-					// MOVQ BP, -16(SP)
-					// LEAQ -16(SP), BP
-					ab.Put(bpduff1)
-				}
 				ab.Put1(byte(op))
 				cursym.AddRel(ctxt, obj.Reloc{
 					Type: objabi.R_CALL,
@@ -4879,12 +4860,6 @@ func (ab *AsmBuf) doasm(ctxt *obj.Link, cursym *obj.LSym, p *obj.Prog) {
 				})
 				ab.PutInt32(0)
 
-				if yt.zcase == Zcallduff && ctxt.Arch.Family == sys.AMD64 {
-					// Pop BP pushed above.
-					// MOVQ 0(BP), BP
-					ab.Put(bpduff2)
-				}
-
 			// TODO: jump across functions needs reloc
 			case Zbr, Zjmp, Zloop:
 				if p.As == AXBEGIN {
diff --git a/src/runtime/duff_amd64.s b/src/runtime/duff_amd64.s
deleted file mode 100644
index 69e9980a30..0000000000
--- a/src/runtime/duff_amd64.s
+++ /dev/null
@@ -1,427 +0,0 @@
-// Code generated by mkduff.go; DO NOT EDIT.
-// Run go generate from src/runtime to update.
-// See mkduff.go for comments.
-
-#include "textflag.h"
-
-TEXT runtime·duffzero<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	MOVUPS	X15,(DI)
-	MOVUPS	X15,16(DI)
-	MOVUPS	X15,32(DI)
-	MOVUPS	X15,48(DI)
-	LEAQ	64(DI),DI
-
-	RET
-
-TEXT runtime·duffcopy<ABIInternal>(SB), NOSPLIT|NOFRAME, $0-0
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	MOVUPS	(SI), X0
-	ADDQ	$16, SI
-	MOVUPS	X0, (DI)
-	ADDQ	$16, DI
-
-	RET
diff --git a/src/runtime/mkduff.go b/src/runtime/mkduff.go
index a0ec987307..dfff084e80 100644
--- a/src/runtime/mkduff.go
+++ b/src/runtime/mkduff.go
@@ -32,7 +32,6 @@ import (
 )
 
 func main() {
-	gen("amd64", notags, zeroAMD64, copyAMD64)
 	gen("386", notags, zero386, copy386)
 	gen("arm", notags, zeroARM, copyARM)
 	gen("loong64", notags, zeroLOONG64, copyLOONG64)
-- 
2.51.0