runtime: record argument size in assembly functions
I have not done the system call stubs in sys_*.s.
I hope to avoid that, because those do not block, so those
frames will not appear in stack traces during garbage
collection.
R=golang-dev, dvyukov, khr
CC=golang-dev
https://golang.org/cl/11360043
diff --git a/src/pkg/runtime/asm_386.s b/src/pkg/runtime/asm_386.s
index 3bf3321..67c8854 100644
--- a/src/pkg/runtime/asm_386.s
+++ b/src/pkg/runtime/asm_386.s
@@ -3,6 +3,7 @@
// license that can be found in the LICENSE file.
#include "zasm_GOOS_GOARCH.h"
+#include "funcdata.h"
TEXT _rt0_go(SB),7,$0
// copy arguments forward on an even stack
@@ -95,7 +96,9 @@
// create a new goroutine to start program
PUSHL $runtime·main·f(SB) // entry
PUSHL $0 // arg size
+ ARGSIZE(8)
CALL runtime·newproc(SB)
+ ARGSIZE(-1)
POPL AX
POPL AX
@@ -108,11 +111,11 @@
DATA runtime·main·f+0(SB)/4,$runtime·main(SB)
GLOBL runtime·main·f(SB),8,$4
-TEXT runtime·breakpoint(SB),7,$0
+TEXT runtime·breakpoint(SB),7,$0-0
INT $3
RET
-TEXT runtime·asminit(SB),7,$0
+TEXT runtime·asminit(SB),7,$0-0
// Linux and MinGW start the FPU in extended double precision.
// Other operating systems use double precision.
// Change to double precision to match them,
@@ -128,7 +131,7 @@
// void gosave(Gobuf*)
// save state in Gobuf; setjmp
-TEXT runtime·gosave(SB), 7, $0
+TEXT runtime·gosave(SB), 7, $0-4
MOVL 4(SP), AX // gobuf
LEAL 4(SP), BX // caller's SP
MOVL BX, gobuf_sp(AX)
@@ -143,7 +146,7 @@
// void gogo(Gobuf*)
// restore state from Gobuf; longjmp
-TEXT runtime·gogo(SB), 7, $0
+TEXT runtime·gogo(SB), 7, $0-4
MOVL 4(SP), BX // gobuf
MOVL gobuf_g(BX), DX
MOVL 0(DX), CX // make sure g != nil
@@ -162,7 +165,7 @@
// Switch to m->g0's stack, call fn(g).
// Fn must never return. It should gogo(&g->sched)
// to keep running g.
-TEXT runtime·mcall(SB), 7, $0
+TEXT runtime·mcall(SB), 7, $0-4
MOVL fn+0(FP), DI
get_tls(CX)
@@ -241,7 +244,7 @@
// with the desired args running the desired function.
//
// func call(fn *byte, arg *byte, argsize uint32).
-TEXT reflect·call(SB), 7, $0
+TEXT reflect·call(SB), 7, $0-12
get_tls(CX)
MOVL m(CX), BX
@@ -307,7 +310,7 @@
// return 1;
// }else
// return 0;
-TEXT runtime·cas(SB), 7, $0
+TEXT runtime·cas(SB), 7, $0-12
MOVL 4(SP), BX
MOVL 8(SP), AX
MOVL 12(SP), CX
@@ -327,7 +330,7 @@
// } else {
// return 0;
// }
-TEXT runtime·cas64(SB), 7, $0
+TEXT runtime·cas64(SB), 7, $0-20
MOVL 4(SP), BP
MOVL 8(SP), AX
MOVL 12(SP), DX
@@ -349,7 +352,7 @@
// return 1;
// }else
// return 0;
-TEXT runtime·casp(SB), 7, $0
+TEXT runtime·casp(SB), 7, $0-12
MOVL 4(SP), BX
MOVL 8(SP), AX
MOVL 12(SP), CX
@@ -365,7 +368,7 @@
// Atomically:
// *val += delta;
// return *val;
-TEXT runtime·xadd(SB), 7, $0
+TEXT runtime·xadd(SB), 7, $0-8
MOVL 4(SP), BX
MOVL 8(SP), AX
MOVL AX, CX
@@ -374,13 +377,13 @@
ADDL CX, AX
RET
-TEXT runtime·xchg(SB), 7, $0
+TEXT runtime·xchg(SB), 7, $0-8
MOVL 4(SP), BX
MOVL 8(SP), AX
XCHGL AX, 0(BX)
RET
-TEXT runtime·procyield(SB),7,$0
+TEXT runtime·procyield(SB),7,$0-0
MOVL 4(SP), AX
again:
PAUSE
@@ -388,13 +391,13 @@
JNZ again
RET
-TEXT runtime·atomicstorep(SB), 7, $0
+TEXT runtime·atomicstorep(SB), 7, $0-8
MOVL 4(SP), BX
MOVL 8(SP), AX
XCHGL AX, 0(BX)
RET
-TEXT runtime·atomicstore(SB), 7, $0
+TEXT runtime·atomicstore(SB), 7, $0-8
MOVL 4(SP), BX
MOVL 8(SP), AX
XCHGL AX, 0(BX)
@@ -403,7 +406,7 @@
// uint64 atomicload64(uint64 volatile* addr);
// so actually
// void atomicload64(uint64 *res, uint64 volatile *addr);
-TEXT runtime·atomicload64(SB), 7, $0
+TEXT runtime·atomicload64(SB), 7, $0-8
MOVL 4(SP), BX
MOVL 8(SP), AX
// MOVQ (%EAX), %MM0
@@ -415,7 +418,7 @@
RET
// void runtime·atomicstore64(uint64 volatile* addr, uint64 v);
-TEXT runtime·atomicstore64(SB), 7, $0
+TEXT runtime·atomicstore64(SB), 7, $0-12
MOVL 4(SP), AX
// MOVQ and EMMS were introduced on the Pentium MMX.
// MOVQ 0x8(%ESP), %MM0
@@ -464,7 +467,7 @@
// Call fn(arg) on the scheduler stack,
// aligned appropriately for the gcc ABI.
// See cgocall.c for more details.
-TEXT runtime·asmcgocall(SB),7,$0
+TEXT runtime·asmcgocall(SB),7,$0-8
MOVL fn+0(FP), AX
MOVL arg+4(FP), BX
MOVL SP, DX
@@ -500,7 +503,7 @@
// cgocallback(void (*fn)(void*), void *frame, uintptr framesize)
// Turn the fn into a Go func (by taking its address) and call
// cgocallback_gofunc.
-TEXT runtime·cgocallback(SB),7,$12
+TEXT runtime·cgocallback(SB),7,$12-12
LEAL fn+0(FP), AX
MOVL AX, 0(SP)
MOVL frame+4(FP), AX
@@ -513,7 +516,7 @@
// cgocallback_gofunc(FuncVal*, void *frame, uintptr framesize)
// See cgocall.c for more details.
-TEXT runtime·cgocallback_gofunc(SB),7,$12
+TEXT runtime·cgocallback_gofunc(SB),7,$12-12
// If m is nil, Go did not create the current thread.
// Call needm to obtain one for temporary use.
// In this case, we're running on the thread stack, so there's
@@ -613,7 +616,7 @@
RET
// void setmg(M*, G*); set m and g. for use by needm.
-TEXT runtime·setmg(SB), 7, $0
+TEXT runtime·setmg(SB), 7, $0-8
#ifdef GOOS_windows
MOVL mm+0(FP), AX
CMPL AX, $0
@@ -633,7 +636,7 @@
RET
// void setmg_gcc(M*, G*); set m and g. for use by gcc
-TEXT setmg_gcc<>(SB), 7, $0
+TEXT setmg_gcc<>(SB), 7, $0
get_tls(AX)
MOVL mm+0(FP), DX
MOVL DX, m(AX)
@@ -642,7 +645,7 @@
RET
// check that SP is in range [g->stackbase, g->stackguard)
-TEXT runtime·stackcheck(SB), 7, $0
+TEXT runtime·stackcheck(SB), 7, $0-0
get_tls(CX)
MOVL g(CX), AX
CMPL g_stackbase(AX), SP
@@ -653,7 +656,7 @@
INT $3
RET
-TEXT runtime·memclr(SB),7,$0
+TEXT runtime·memclr(SB),7,$0-8
MOVL 4(SP), DI // arg 1 addr
MOVL 8(SP), CX // arg 2 count
MOVL CX, BX
@@ -668,31 +671,31 @@
STOSB
RET
-TEXT runtime·getcallerpc(SB),7,$0
+TEXT runtime·getcallerpc(SB),7,$0-4
MOVL x+0(FP),AX // addr of first arg
MOVL -4(AX),AX // get calling pc
RET
-TEXT runtime·setcallerpc(SB),7,$0
+TEXT runtime·setcallerpc(SB),7,$0-8
MOVL x+0(FP),AX // addr of first arg
MOVL x+4(FP), BX
MOVL BX, -4(AX) // set calling pc
RET
-TEXT runtime·getcallersp(SB), 7, $0
+TEXT runtime·getcallersp(SB), 7, $0-4
MOVL sp+0(FP), AX
RET
// int64 runtime·cputicks(void), so really
// void runtime·cputicks(int64 *ticks)
-TEXT runtime·cputicks(SB),7,$0
+TEXT runtime·cputicks(SB),7,$0-4
RDTSC
MOVL ret+0(FP), DI
MOVL AX, 0(DI)
MOVL DX, 4(DI)
RET
-TEXT runtime·ldt0setup(SB),7,$16
+TEXT runtime·ldt0setup(SB),7,$16-0
// set up ldt 7 to point at tls0
// ldt 1 would be fine on Linux, but on OS X, 7 is as low as we can go.
// the entry number is just a hint. setldt will set up GS with what it used.
@@ -703,13 +706,13 @@
CALL runtime·setldt(SB)
RET
-TEXT runtime·emptyfunc(SB),0,$0
+TEXT runtime·emptyfunc(SB),0,$0-0
RET
-TEXT runtime·abort(SB),7,$0
+TEXT runtime·abort(SB),7,$0-0
INT $0x3
-TEXT runtime·stackguard(SB),7,$0
+TEXT runtime·stackguard(SB),7,$0-8
MOVL SP, DX
MOVL DX, sp+0(FP)
get_tls(CX)
@@ -721,13 +724,13 @@
GLOBL runtime·tls0(SB), $32
// hash function using AES hardware instructions
-TEXT runtime·aeshash(SB),7,$0
+TEXT runtime·aeshash(SB),7,$0-12
MOVL 4(SP), DX // ptr to hash value
MOVL 8(SP), CX // size
MOVL 12(SP), AX // ptr to data
JMP runtime·aeshashbody(SB)
-TEXT runtime·aeshashstr(SB),7,$0
+TEXT runtime·aeshashstr(SB),7,$0-12
MOVL 4(SP), DX // ptr to hash value
MOVL 12(SP), AX // ptr to string struct
MOVL 4(AX), CX // length of string
@@ -737,7 +740,7 @@
// AX: data
// CX: length
// DX: ptr to seed input / hash output
-TEXT runtime·aeshashbody(SB),7,$0
+TEXT runtime·aeshashbody(SB),7,$0-12
MOVL (DX), X0 // seed to low 32 bits of xmm0
PINSRD $1, CX, X0 // size to next 32 bits of xmm0
MOVO runtime·aeskeysched+0(SB), X2
@@ -771,7 +774,7 @@
// a page boundary, so we can load it directly.
MOVOU (AX), X1
ADDL CX, CX
- PAND masks(SB)(CX*8), X1
+ PAND masks<>(SB)(CX*8), X1
JMP partial
highpartial:
// address ends in 1111xxxx. Might be up against
@@ -779,7 +782,7 @@
// Then shift bytes down using pshufb.
MOVOU -16(AX)(CX*1), X1
ADDL CX, CX
- PSHUFB shifts(SB)(CX*8), X1
+ PSHUFB shifts<>(SB)(CX*8), X1
partial:
// incorporate partial block into hash
AESENC X3, X0
@@ -792,7 +795,7 @@
MOVL X0, (DX)
RET
-TEXT runtime·aeshash32(SB),7,$0
+TEXT runtime·aeshash32(SB),7,$0-12
MOVL 4(SP), DX // ptr to hash value
MOVL 12(SP), AX // ptr to data
MOVL (DX), X0 // seed
@@ -803,7 +806,7 @@
MOVL X0, (DX)
RET
-TEXT runtime·aeshash64(SB),7,$0
+TEXT runtime·aeshash64(SB),7,$0-12
MOVL 4(SP), DX // ptr to hash value
MOVL 12(SP), AX // ptr to data
MOVQ (AX), X0 // data
@@ -814,181 +817,181 @@
MOVL X0, (DX)
RET
-
// simple mask to get rid of data in the high part of the register.
-TEXT masks(SB),7,$0
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x00(SB)/4, $0x00000000
+DATA masks<>+0x04(SB)/4, $0x00000000
+DATA masks<>+0x08(SB)/4, $0x00000000
+DATA masks<>+0x0c(SB)/4, $0x00000000
- LONG $0x000000ff
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x10(SB)/4, $0x000000ff
+DATA masks<>+0x14(SB)/4, $0x00000000
+DATA masks<>+0x18(SB)/4, $0x00000000
+DATA masks<>+0x1c(SB)/4, $0x00000000
- LONG $0x0000ffff
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x20(SB)/4, $0x0000ffff
+DATA masks<>+0x24(SB)/4, $0x00000000
+DATA masks<>+0x28(SB)/4, $0x00000000
+DATA masks<>+0x2c(SB)/4, $0x00000000
- LONG $0x00ffffff
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x30(SB)/4, $0x00ffffff
+DATA masks<>+0x34(SB)/4, $0x00000000
+DATA masks<>+0x38(SB)/4, $0x00000000
+DATA masks<>+0x3c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x40(SB)/4, $0xffffffff
+DATA masks<>+0x44(SB)/4, $0x00000000
+DATA masks<>+0x48(SB)/4, $0x00000000
+DATA masks<>+0x4c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0x000000ff
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x50(SB)/4, $0xffffffff
+DATA masks<>+0x54(SB)/4, $0x000000ff
+DATA masks<>+0x58(SB)/4, $0x00000000
+DATA masks<>+0x5c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0x0000ffff
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x60(SB)/4, $0xffffffff
+DATA masks<>+0x64(SB)/4, $0x0000ffff
+DATA masks<>+0x68(SB)/4, $0x00000000
+DATA masks<>+0x6c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0x00ffffff
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x70(SB)/4, $0xffffffff
+DATA masks<>+0x74(SB)/4, $0x00ffffff
+DATA masks<>+0x78(SB)/4, $0x00000000
+DATA masks<>+0x7c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x00000000
- LONG $0x00000000
+DATA masks<>+0x80(SB)/4, $0xffffffff
+DATA masks<>+0x84(SB)/4, $0xffffffff
+DATA masks<>+0x88(SB)/4, $0x00000000
+DATA masks<>+0x8c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x000000ff
- LONG $0x00000000
+DATA masks<>+0x90(SB)/4, $0xffffffff
+DATA masks<>+0x94(SB)/4, $0xffffffff
+DATA masks<>+0x98(SB)/4, $0x000000ff
+DATA masks<>+0x9c(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x0000ffff
- LONG $0x00000000
+DATA masks<>+0xa0(SB)/4, $0xffffffff
+DATA masks<>+0xa4(SB)/4, $0xffffffff
+DATA masks<>+0xa8(SB)/4, $0x0000ffff
+DATA masks<>+0xac(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x00ffffff
- LONG $0x00000000
+DATA masks<>+0xb0(SB)/4, $0xffffffff
+DATA masks<>+0xb4(SB)/4, $0xffffffff
+DATA masks<>+0xb8(SB)/4, $0x00ffffff
+DATA masks<>+0xbc(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x00000000
+DATA masks<>+0xc0(SB)/4, $0xffffffff
+DATA masks<>+0xc4(SB)/4, $0xffffffff
+DATA masks<>+0xc8(SB)/4, $0xffffffff
+DATA masks<>+0xcc(SB)/4, $0x00000000
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x000000ff
+DATA masks<>+0xd0(SB)/4, $0xffffffff
+DATA masks<>+0xd4(SB)/4, $0xffffffff
+DATA masks<>+0xd8(SB)/4, $0xffffffff
+DATA masks<>+0xdc(SB)/4, $0x000000ff
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x0000ffff
+DATA masks<>+0xe0(SB)/4, $0xffffffff
+DATA masks<>+0xe4(SB)/4, $0xffffffff
+DATA masks<>+0xe8(SB)/4, $0xffffffff
+DATA masks<>+0xec(SB)/4, $0x0000ffff
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0x00ffffff
+DATA masks<>+0xf0(SB)/4, $0xffffffff
+DATA masks<>+0xf4(SB)/4, $0xffffffff
+DATA masks<>+0xf8(SB)/4, $0xffffffff
+DATA masks<>+0xfc(SB)/4, $0x00ffffff
- // these are arguments to pshufb. They move data down from
- // the high bytes of the register to the low bytes of the register.
- // index is how many bytes to move.
-TEXT shifts(SB),7,$0
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
- LONG $0x00000000
-
- LONG $0xffffff0f
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0xffff0f0e
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0xff0f0e0d
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0x0f0e0d0c
- LONG $0xffffffff
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0x0e0d0c0b
- LONG $0xffffff0f
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0x0d0c0b0a
- LONG $0xffff0f0e
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0x0c0b0a09
- LONG $0xff0f0e0d
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0x0b0a0908
- LONG $0x0f0e0d0c
- LONG $0xffffffff
- LONG $0xffffffff
-
- LONG $0x0a090807
- LONG $0x0e0d0c0b
- LONG $0xffffff0f
- LONG $0xffffffff
-
- LONG $0x09080706
- LONG $0x0d0c0b0a
- LONG $0xffff0f0e
- LONG $0xffffffff
-
- LONG $0x08070605
- LONG $0x0c0b0a09
- LONG $0xff0f0e0d
- LONG $0xffffffff
-
- LONG $0x07060504
- LONG $0x0b0a0908
- LONG $0x0f0e0d0c
- LONG $0xffffffff
-
- LONG $0x06050403
- LONG $0x0a090807
- LONG $0x0e0d0c0b
- LONG $0xffffff0f
-
- LONG $0x05040302
- LONG $0x09080706
- LONG $0x0d0c0b0a
- LONG $0xffff0f0e
-
- LONG $0x04030201
- LONG $0x08070605
- LONG $0x0c0b0a09
- LONG $0xff0f0e0d
+GLOBL masks<>(SB),8,$256
-TEXT runtime·memeq(SB),7,$0
+// these are arguments to pshufb. They move data down from
+// the high bytes of the register to the low bytes of the register.
+// index is how many bytes to move.
+DATA shifts<>+0x00(SB)/4, $0x00000000
+DATA shifts<>+0x04(SB)/4, $0x00000000
+DATA shifts<>+0x08(SB)/4, $0x00000000
+DATA shifts<>+0x0c(SB)/4, $0x00000000
+
+DATA shifts<>+0x10(SB)/4, $0xffffff0f
+DATA shifts<>+0x14(SB)/4, $0xffffffff
+DATA shifts<>+0x18(SB)/4, $0xffffffff
+DATA shifts<>+0x1c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x20(SB)/4, $0xffff0f0e
+DATA shifts<>+0x24(SB)/4, $0xffffffff
+DATA shifts<>+0x28(SB)/4, $0xffffffff
+DATA shifts<>+0x2c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x30(SB)/4, $0xff0f0e0d
+DATA shifts<>+0x34(SB)/4, $0xffffffff
+DATA shifts<>+0x38(SB)/4, $0xffffffff
+DATA shifts<>+0x3c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x40(SB)/4, $0x0f0e0d0c
+DATA shifts<>+0x44(SB)/4, $0xffffffff
+DATA shifts<>+0x48(SB)/4, $0xffffffff
+DATA shifts<>+0x4c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x50(SB)/4, $0x0e0d0c0b
+DATA shifts<>+0x54(SB)/4, $0xffffff0f
+DATA shifts<>+0x58(SB)/4, $0xffffffff
+DATA shifts<>+0x5c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x60(SB)/4, $0x0d0c0b0a
+DATA shifts<>+0x64(SB)/4, $0xffff0f0e
+DATA shifts<>+0x68(SB)/4, $0xffffffff
+DATA shifts<>+0x6c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x70(SB)/4, $0x0c0b0a09
+DATA shifts<>+0x74(SB)/4, $0xff0f0e0d
+DATA shifts<>+0x78(SB)/4, $0xffffffff
+DATA shifts<>+0x7c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x80(SB)/4, $0x0b0a0908
+DATA shifts<>+0x84(SB)/4, $0x0f0e0d0c
+DATA shifts<>+0x88(SB)/4, $0xffffffff
+DATA shifts<>+0x8c(SB)/4, $0xffffffff
+
+DATA shifts<>+0x90(SB)/4, $0x0a090807
+DATA shifts<>+0x94(SB)/4, $0x0e0d0c0b
+DATA shifts<>+0x98(SB)/4, $0xffffff0f
+DATA shifts<>+0x9c(SB)/4, $0xffffffff
+
+DATA shifts<>+0xa0(SB)/4, $0x09080706
+DATA shifts<>+0xa4(SB)/4, $0x0d0c0b0a
+DATA shifts<>+0xa8(SB)/4, $0xffff0f0e
+DATA shifts<>+0xac(SB)/4, $0xffffffff
+
+DATA shifts<>+0xb0(SB)/4, $0x08070605
+DATA shifts<>+0xb4(SB)/4, $0x0c0b0a09
+DATA shifts<>+0xb8(SB)/4, $0xff0f0e0d
+DATA shifts<>+0xbc(SB)/4, $0xffffffff
+
+DATA shifts<>+0xc0(SB)/4, $0x07060504
+DATA shifts<>+0xc4(SB)/4, $0x0b0a0908
+DATA shifts<>+0xc8(SB)/4, $0x0f0e0d0c
+DATA shifts<>+0xcc(SB)/4, $0xffffffff
+
+DATA shifts<>+0xd0(SB)/4, $0x06050403
+DATA shifts<>+0xd4(SB)/4, $0x0a090807
+DATA shifts<>+0xd8(SB)/4, $0x0e0d0c0b
+DATA shifts<>+0xdc(SB)/4, $0xffffff0f
+
+DATA shifts<>+0xe0(SB)/4, $0x05040302
+DATA shifts<>+0xe4(SB)/4, $0x09080706
+DATA shifts<>+0xe8(SB)/4, $0x0d0c0b0a
+DATA shifts<>+0xec(SB)/4, $0xffff0f0e
+
+DATA shifts<>+0xf0(SB)/4, $0x04030201
+DATA shifts<>+0xf4(SB)/4, $0x08070605
+DATA shifts<>+0xf8(SB)/4, $0x0c0b0a09
+DATA shifts<>+0xfc(SB)/4, $0xff0f0e0d
+
+GLOBL shifts<>(SB),8,$256
+
+TEXT runtime·memeq(SB),7,$0-12
MOVL a+0(FP), SI
MOVL b+4(FP), DI
MOVL count+8(FP), BX
JMP runtime·memeqbody(SB)
-
-TEXT bytes·Equal(SB),7,$0
+TEXT bytes·Equal(SB),7,$0-25
MOVL a_len+4(FP), BX
MOVL b_len+16(FP), CX
XORL AX, AX
@@ -1004,7 +1007,7 @@
// a in SI
// b in DI
// count in BX
-TEXT runtime·memeqbody(SB),7,$0
+TEXT runtime·memeqbody(SB),7,$0-0
XORL AX, AX
CMPL BX, $4
@@ -1097,7 +1100,7 @@
SETEQ AX
RET
-TEXT runtime·cmpstring(SB),7,$0
+TEXT runtime·cmpstring(SB),7,$0-20
MOVL s1+0(FP), SI
MOVL s1+4(FP), BX
MOVL s2+8(FP), DI
@@ -1106,7 +1109,7 @@
MOVL AX, res+16(FP)
RET
-TEXT bytes·Compare(SB),7,$0
+TEXT bytes·Compare(SB),7,$0-28
MOVL s1+0(FP), SI
MOVL s1+4(FP), BX
MOVL s2+12(FP), DI
@@ -1122,7 +1125,7 @@
// DX = blen
// output:
// AX = 1/0/-1
-TEXT runtime·cmpbody(SB),7,$0
+TEXT runtime·cmpbody(SB),7,$0-0
CMPL SI, DI
JEQ cmp_allsame
CMPL BX, DX
