*/
.globl EXT(crosscall1)
EXT(crosscall1):
	pushl %ebp
	movl %esp, %ebp
	pushl %ebx
	pushl %esi
	pushl %edi

	movl 16(%ebp), %eax	/* g */
	pushl %eax
	movl 12(%ebp), %eax	/* setg_gcc */
	call *%eax
	popl %eax

	movl 8(%ebp), %eax	/* fn */
	call *%eax

	popl %edi
	popl %esi
	popl %ebx
	popl %ebp
	ret

#ifdef __ELF__
.section .note.GNU-stack,"",@progbits

push    offset msg
call    printf
mov eax,0
push eax
call exit
sayFooFromAsm endp
end
"""
        } else {
            return """
.text

LC0:
.ascii "fooFromAsm\\12\\0"
.globl _sayFooFromAsm

_sayFooFromAsm:
        pushl   %ebp
        movl    %esp, %ebp
        subl    \$8, %esp
        andl    \$-16, %esp
        movl    \$0, %eax
        movl    %eax, -4(%ebp)

    .text
    .globl  _sum
_sum:
    movl    8(%esp), %eax
    addl    4(%esp), %eax

INCLUDE <ks386.inc>
PUBLIC    _sumx
_TEXT     SEGMENT
_sumx    PROC
mov    eax, DWORD PTR 4[esp]
add    eax, DWORD PTR 8[esp]
ret    0
_sumx    ENDP
_TEXT   ENDS
END
'''

    private static String i386GnuAsmSource = '''
    .text
    .globl  _sumx
_sumx:
    movl    8(%esp), %eax
    addl    4(%esp), %eax
    ret
'''

    private static String x64GnuAsmSource = '''
    .text

    .386
    .model    flat

PUBLIC    _sum
_TEXT     SEGMENT
_sum    PROC
    mov    eax, DWORD PTR 4[esp]
    add    eax, DWORD PTR 8[esp]
    ret    0
_sum    ENDP
_TEXT   ENDS

	/*
	 * Set specific keys in thread local storage.
	 */
	asm volatile (
		"movl %0, %%fs:0(%1)\n"	// MOVL tls0, 0(tls_g)(FS)
		"movl %%fs:0(%1), %%eax\n"	// MOVL 0(tls_g)(FS), tmp
		"movl %2, 0(%%eax)\n"	// MOVL g, 0(AX)
		:: "r"(ts.tls), "r"(*tls_g), "r"(ts.g) : "%eax"
	);

	crosscall1(ts.fn, setg_gcc, ts.g);

#include "textflag.h"

// func cpuid(eaxArg, ecxArg uint32) (eax, ebx, ecx, edx uint32)
TEXT ·cpuid(SB), NOSPLIT, $0-24
	MOVL eaxArg+0(FP), AX
	MOVL ecxArg+4(FP), CX
	CPUID
	MOVL AX, eax+8(FP)
	MOVL BX, ebx+12(FP)
	MOVL CX, ecx+16(FP)
	MOVL DX, edx+20(FP)
	RET

// func xgetbv() (eax, edx uint32)
TEXT ·xgetbv(SB),NOSPLIT,$0-8
	MOVL $0, CX
	XGETBV
	MOVL AX, eax+0(FP)
	MOVL DX, edx+4(FP)
	RET

// Need to wrap __get_cpuid_count because it's declared as static.
int
gccgoGetCpuidCount(uint32_t leaf, uint32_t subleaf,
                   uint32_t *eax, uint32_t *ebx,
                   uint32_t *ecx, uint32_t *edx)
{
	return __get_cpuid_count(leaf, subleaf, eax, ebx, ecx, edx);
}

#pragma GCC diagnostic ignored "-Wunknown-pragmas"
#pragma GCC push_options
#pragma GCC target("xsave")

		return ""
	}

	data := make([]byte, 0, 3*4*4)

	var eax, ebx, ecx, edx uint32
	eax, ebx, ecx, edx = cpuid(0x80000002, 0)
	data = appendBytes(data, eax, ebx, ecx, edx)
	eax, ebx, ecx, edx = cpuid(0x80000003, 0)
	data = appendBytes(data, eax, ebx, ecx, edx)
	eax, ebx, ecx, edx = cpuid(0x80000004, 0)
	data = appendBytes(data, eax, ebx, ecx, edx)

	// Trim leading spaces.
	for len(data) > 0 && data[0] == ' ' {

#include "textflag.h"

// func cpuid(eaxArg, ecxArg uint32) (eax, ebx, ecx, edx uint32)
TEXT ·cpuid(SB), NOSPLIT, $0-24
	MOVL eaxArg+0(FP), AX
	MOVL ecxArg+4(FP), CX
	CPUID
	MOVL AX, eax+8(FP)
	MOVL BX, ebx+12(FP)
	MOVL CX, ecx+16(FP)
	MOVL DX, edx+20(FP)
	RET

// func xgetbv() (eax, edx uint32)
TEXT ·xgetbv(SB),NOSPLIT,$0-8
	MOVL $0, CX
	XGETBV
	MOVL AX, eax+0(FP)
	MOVL DX, edx+4(FP)