OSchip
/
llvm-project
mirror of https://github.com/THU-DSP-LAB/llvm-project.git
13
0
Fork 6
Code Issues Packages Projects Releases Wiki Activity

[X86] Mark FP32_TO_INT16_IN_MEM/FP32_TO_INT32_IN_MEM/FP32_TO_INT64_IN_MEM as clobbering EFLAGS to prevent mis-scheduling during conversion from SelectionDAG to MIR. 

After r354178, these instruction expand to a sequence that uses an OR instruction. That OR clobbers EFLAGS so we need to state that to avoid accidentally using the clobbered flags.

Our tests show the bug, but I didn't notice because the SETcc instructions didn't move after r354178 since it used to be safe to do the fp->int conversion first.

We should probably convert this whole sequence to SelectionDAG instead of a custom inserter to avoid mistakes like this.

Fixes PR40779

llvm-svn: 354395
This commit is contained in:
Craig Topper 2019-02-19 22:37:00 +00:00
parent b6bc11d406
commit 8eade09249
3 changed files with 55 additions and 53 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
llvm/lib/Target/X86/X86InstrFPStack.td
View File

@ -113,7 +113,9 @@ def fpimmneg1 : FPImmLeaf<fAny, [{
}]>; }]>;
// Some 'special' instructions - expanded after instruction selection. // Some 'special' instructions - expanded after instruction selection.
let usesCustomInserter = 1, hasNoSchedulingInfo = 1 in { // Clobbers EFLAGS due to OR instruction used internally.
// FIXME: Can we model this in SelectionDAG?
let usesCustomInserter = 1, hasNoSchedulingInfo = 1, Defs = [EFLAGS] in {
def FP32_TO_INT16_IN_MEM : PseudoI<(outs), (ins i16mem:$dst, RFP32:$src), def FP32_TO_INT16_IN_MEM : PseudoI<(outs), (ins i16mem:$dst, RFP32:$src),
[(X86fp_to_i16mem RFP32:$src, addr:$dst)]>; [(X86fp_to_i16mem RFP32:$src, addr:$dst)]>;
def FP32_TO_INT32_IN_MEM : PseudoI<(outs), (ins i32mem:$dst, RFP32:$src), def FP32_TO_INT32_IN_MEM : PseudoI<(outs), (ins i32mem:$dst, RFP32:$src),

28
llvm/test/CodeGen/X86/fp-cvt.ll
View File

@ -457,15 +457,15 @@ define i64 @fptoui_i64_fp80(x86_fp80 %a0) nounwind {
; X86-NEXT: fldz ; X86-NEXT: fldz
; X86-NEXT: .LBB10_2: ; X86-NEXT: .LBB10_2:
; X86-NEXT: fstp %st(0) ; X86-NEXT: fstp %st(0)
; X86-NEXT: setbe %al
; X86-NEXT: fnstcw {{[0-9]+}}(%esp) ; X86-NEXT: fnstcw {{[0-9]+}}(%esp)
; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: xorl %edx, %edx ; X86-NEXT: orl $3072, %ecx # imm = 0xC00
; X86-NEXT: orl $3072, %eax # imm = 0xC00 ; X86-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X86-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X86-NEXT: fldcw {{[0-9]+}}(%esp) ; X86-NEXT: fldcw {{[0-9]+}}(%esp)
; X86-NEXT: fistpll {{[0-9]+}}(%esp) ; X86-NEXT: fistpll {{[0-9]+}}(%esp)
; X86-NEXT: fldcw {{[0-9]+}}(%esp) ; X86-NEXT: fldcw {{[0-9]+}}(%esp)
; X86-NEXT: setbe %dl ; X86-NEXT: movzbl %al, %edx
; X86-NEXT: shll $31, %edx ; X86-NEXT: shll $31, %edx
; X86-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X86-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -479,19 +479,19 @@ define i64 @fptoui_i64_fp80(x86_fp80 %a0) nounwind {
; X64-X87-NEXT: flds {{.*}}(%rip) ; X64-X87-NEXT: flds {{.*}}(%rip)
; X64-X87-NEXT: fld %st(1) ; X64-X87-NEXT: fld %st(1)
; X64-X87-NEXT: fsub %st(1), %st ; X64-X87-NEXT: fsub %st(1), %st
; X64-X87-NEXT: xorl %eax, %eax
; X64-X87-NEXT: fxch %st(1) ; X64-X87-NEXT: fxch %st(1)
; X64-X87-NEXT: fucompi %st(2), %st ; X64-X87-NEXT: fucompi %st(2), %st
; X64-X87-NEXT: fcmovnbe %st(1), %st ; X64-X87-NEXT: fcmovnbe %st(1), %st
; X64-X87-NEXT: fstp %st(1) ; X64-X87-NEXT: fstp %st(1)
; X64-X87-NEXT: setbe %al
; X64-X87-NEXT: fnstcw -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fnstcw -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx ; X64-X87-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx
; X64-X87-NEXT: xorl %eax, %eax
; X64-X87-NEXT: orl $3072, %ecx # imm = 0xC00 ; X64-X87-NEXT: orl $3072, %ecx # imm = 0xC00
; X64-X87-NEXT: movw %cx, -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: movw %cx, -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: fistpll -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fistpll -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: setbe %al
; X64-X87-NEXT: shlq $63, %rax ; X64-X87-NEXT: shlq $63, %rax
; X64-X87-NEXT: xorq -{{[0-9]+}}(%rsp), %rax ; X64-X87-NEXT: xorq -{{[0-9]+}}(%rsp), %rax
; X64-X87-NEXT: retq ; X64-X87-NEXT: retq
@ -539,15 +539,15 @@ define i64 @fptoui_i64_fp80_ld(x86_fp80 *%a0) nounwind {
; X86-NEXT: fldz ; X86-NEXT: fldz
; X86-NEXT: .LBB11_2: ; X86-NEXT: .LBB11_2:
; X86-NEXT: fstp %st(0) ; X86-NEXT: fstp %st(0)
; X86-NEXT: setbe %al
; X86-NEXT: fnstcw {{[0-9]+}}(%esp) ; X86-NEXT: fnstcw {{[0-9]+}}(%esp)
; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X86-NEXT: xorl %edx, %edx ; X86-NEXT: orl $3072, %ecx # imm = 0xC00
; X86-NEXT: orl $3072, %eax # imm = 0xC00 ; X86-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X86-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X86-NEXT: fldcw {{[0-9]+}}(%esp) ; X86-NEXT: fldcw {{[0-9]+}}(%esp)
; X86-NEXT: fistpll {{[0-9]+}}(%esp) ; X86-NEXT: fistpll {{[0-9]+}}(%esp)
; X86-NEXT: fldcw {{[0-9]+}}(%esp) ; X86-NEXT: fldcw {{[0-9]+}}(%esp)
; X86-NEXT: setbe %dl ; X86-NEXT: movzbl %al, %edx
; X86-NEXT: shll $31, %edx ; X86-NEXT: shll $31, %edx
; X86-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X86-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X86-NEXT: movl {{[0-9]+}}(%esp), %eax ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -561,19 +561,19 @@ define i64 @fptoui_i64_fp80_ld(x86_fp80 *%a0) nounwind {
; X64-X87-NEXT: flds {{.*}}(%rip) ; X64-X87-NEXT: flds {{.*}}(%rip)
; X64-X87-NEXT: fld %st(1) ; X64-X87-NEXT: fld %st(1)
; X64-X87-NEXT: fsub %st(1), %st ; X64-X87-NEXT: fsub %st(1), %st
; X64-X87-NEXT: xorl %eax, %eax
; X64-X87-NEXT: fxch %st(1) ; X64-X87-NEXT: fxch %st(1)
; X64-X87-NEXT: fucompi %st(2), %st ; X64-X87-NEXT: fucompi %st(2), %st
; X64-X87-NEXT: fcmovnbe %st(1), %st ; X64-X87-NEXT: fcmovnbe %st(1), %st
; X64-X87-NEXT: fstp %st(1) ; X64-X87-NEXT: fstp %st(1)
; X64-X87-NEXT: setbe %al
; X64-X87-NEXT: fnstcw -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fnstcw -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx ; X64-X87-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx
; X64-X87-NEXT: xorl %eax, %eax
; X64-X87-NEXT: orl $3072, %ecx # imm = 0xC00 ; X64-X87-NEXT: orl $3072, %ecx # imm = 0xC00
; X64-X87-NEXT: movw %cx, -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: movw %cx, -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: fistpll -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fistpll -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp) ; X64-X87-NEXT: fldcw -{{[0-9]+}}(%rsp)
; X64-X87-NEXT: setbe %al
; X64-X87-NEXT: shlq $63, %rax ; X64-X87-NEXT: shlq $63, %rax
; X64-X87-NEXT: xorq -{{[0-9]+}}(%rsp), %rax ; X64-X87-NEXT: xorq -{{[0-9]+}}(%rsp), %rax
; X64-X87-NEXT: retq ; X64-X87-NEXT: retq

76
llvm/test/CodeGen/X86/scalar-fp-to-i64.ll
View File

@ -289,15 +289,15 @@ define i64 @f_to_u64(float %a) nounwind {
; X87_WIN-NEXT: fldz ; X87_WIN-NEXT: fldz
; X87_WIN-NEXT: LBB0_2: ; X87_WIN-NEXT: LBB0_2:
; X87_WIN-NEXT: fstp %st(0) ; X87_WIN-NEXT: fstp %st(0)
; X87_WIN-NEXT: setbe %al
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_WIN-NEXT: xorl %edx, %edx ; X87_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_WIN-NEXT: orl $3072, %eax # imm = 0xC00 ; X87_WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: setbe %dl ; X87_WIN-NEXT: movzbl %al, %edx
; X87_WIN-NEXT: shll $31, %edx ; X87_WIN-NEXT: shll $31, %edx
; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -323,15 +323,15 @@ define i64 @f_to_u64(float %a) nounwind {
; X87_LIN-NEXT: fldz ; X87_LIN-NEXT: fldz
; X87_LIN-NEXT: .LBB0_2: ; X87_LIN-NEXT: .LBB0_2:
; X87_LIN-NEXT: fstp %st(0) ; X87_LIN-NEXT: fstp %st(0)
; X87_LIN-NEXT: setbe %al
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_LIN-NEXT: xorl %edx, %edx ; X87_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_LIN-NEXT: orl $3072, %eax # imm = 0xC00 ; X87_LIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: setbe %dl ; X87_LIN-NEXT: movzbl %al, %edx
; X87_LIN-NEXT: shll $31, %edx ; X87_LIN-NEXT: shll $31, %edx
; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -779,15 +779,15 @@ define i64 @d_to_u64(double %a) nounwind {
; X87_WIN-NEXT: fldz ; X87_WIN-NEXT: fldz
; X87_WIN-NEXT: LBB2_2: ; X87_WIN-NEXT: LBB2_2:
; X87_WIN-NEXT: fstp %st(0) ; X87_WIN-NEXT: fstp %st(0)
; X87_WIN-NEXT: setbe %al
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_WIN-NEXT: xorl %edx, %edx ; X87_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_WIN-NEXT: orl $3072, %eax # imm = 0xC00 ; X87_WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: setbe %dl ; X87_WIN-NEXT: movzbl %al, %edx
; X87_WIN-NEXT: shll $31, %edx ; X87_WIN-NEXT: shll $31, %edx
; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -813,15 +813,15 @@ define i64 @d_to_u64(double %a) nounwind {
; X87_LIN-NEXT: fldz ; X87_LIN-NEXT: fldz
; X87_LIN-NEXT: .LBB2_2: ; X87_LIN-NEXT: .LBB2_2:
; X87_LIN-NEXT: fstp %st(0) ; X87_LIN-NEXT: fstp %st(0)
; X87_LIN-NEXT: setbe %al
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_LIN-NEXT: xorl %edx, %edx ; X87_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_LIN-NEXT: orl $3072, %eax # imm = 0xC00 ; X87_LIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: setbe %dl ; X87_LIN-NEXT: movzbl %al, %edx
; X87_LIN-NEXT: shll $31, %edx ; X87_LIN-NEXT: shll $31, %edx
; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -1189,19 +1189,19 @@ define i64 @x_to_u64(x86_fp80 %a) nounwind {
; SSE2_32_WIN-NEXT: flds __real@5f000000 ; SSE2_32_WIN-NEXT: flds __real@5f000000
; SSE2_32_WIN-NEXT: fld %st(1) ; SSE2_32_WIN-NEXT: fld %st(1)
; SSE2_32_WIN-NEXT: fsub %st(1), %st ; SSE2_32_WIN-NEXT: fsub %st(1), %st
; SSE2_32_WIN-NEXT: xorl %edx, %edx
; SSE2_32_WIN-NEXT: fxch %st(1) ; SSE2_32_WIN-NEXT: fxch %st(1)
; SSE2_32_WIN-NEXT: fucompi %st(2), %st ; SSE2_32_WIN-NEXT: fucompi %st(2), %st
; SSE2_32_WIN-NEXT: fcmovnbe %st(1), %st ; SSE2_32_WIN-NEXT: fcmovnbe %st(1), %st
; SSE2_32_WIN-NEXT: fstp %st(1) ; SSE2_32_WIN-NEXT: fstp %st(1)
; SSE2_32_WIN-NEXT: setbe %dl
; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; SSE2_32_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; SSE2_32_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_WIN-NEXT: xorl %edx, %edx
; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00 ; SSE2_32_WIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp) ; SSE2_32_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp) ; SSE2_32_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; SSE2_32_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_WIN-NEXT: setbe %dl
; SSE2_32_WIN-NEXT: shll $31, %edx ; SSE2_32_WIN-NEXT: shll $31, %edx
; SSE2_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; SSE2_32_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; SSE2_32_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -1216,19 +1216,19 @@ define i64 @x_to_u64(x86_fp80 %a) nounwind {
; SSE2_32_LIN-NEXT: flds {{\.LCPI.*}} ; SSE2_32_LIN-NEXT: flds {{\.LCPI.*}}
; SSE2_32_LIN-NEXT: fld %st(1) ; SSE2_32_LIN-NEXT: fld %st(1)
; SSE2_32_LIN-NEXT: fsub %st(1), %st ; SSE2_32_LIN-NEXT: fsub %st(1), %st
; SSE2_32_LIN-NEXT: xorl %edx, %edx
; SSE2_32_LIN-NEXT: fxch %st(1) ; SSE2_32_LIN-NEXT: fxch %st(1)
; SSE2_32_LIN-NEXT: fucompi %st(2), %st ; SSE2_32_LIN-NEXT: fucompi %st(2), %st
; SSE2_32_LIN-NEXT: fcmovnbe %st(1), %st ; SSE2_32_LIN-NEXT: fcmovnbe %st(1), %st
; SSE2_32_LIN-NEXT: fstp %st(1) ; SSE2_32_LIN-NEXT: fstp %st(1)
; SSE2_32_LIN-NEXT: setbe %dl
; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; SSE2_32_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; SSE2_32_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax
; SSE2_32_LIN-NEXT: xorl %edx, %edx
; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00 ; SSE2_32_LIN-NEXT: orl $3072, %eax # imm = 0xC00
; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp) ; SSE2_32_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp) ; SSE2_32_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; SSE2_32_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; SSE2_32_LIN-NEXT: setbe %dl
; SSE2_32_LIN-NEXT: shll $31, %edx ; SSE2_32_LIN-NEXT: shll $31, %edx
; SSE2_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; SSE2_32_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; SSE2_32_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -1242,19 +1242,19 @@ define i64 @x_to_u64(x86_fp80 %a) nounwind {
; SSE2_64_WIN-NEXT: flds __real@{{.*}}(%rip) ; SSE2_64_WIN-NEXT: flds __real@{{.*}}(%rip)
; SSE2_64_WIN-NEXT: fld %st(1) ; SSE2_64_WIN-NEXT: fld %st(1)
; SSE2_64_WIN-NEXT: fsub %st(1), %st ; SSE2_64_WIN-NEXT: fsub %st(1), %st
; SSE2_64_WIN-NEXT: xorl %eax, %eax
; SSE2_64_WIN-NEXT: fxch %st(1) ; SSE2_64_WIN-NEXT: fxch %st(1)
; SSE2_64_WIN-NEXT: fucompi %st(2), %st ; SSE2_64_WIN-NEXT: fucompi %st(2), %st
; SSE2_64_WIN-NEXT: fcmovnbe %st(1), %st ; SSE2_64_WIN-NEXT: fcmovnbe %st(1), %st
; SSE2_64_WIN-NEXT: fstp %st(1) ; SSE2_64_WIN-NEXT: fstp %st(1)
; SSE2_64_WIN-NEXT: setbe %al
; SSE2_64_WIN-NEXT: fnstcw {{[0-9]+}}(%rsp) ; SSE2_64_WIN-NEXT: fnstcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: movzwl {{[0-9]+}}(%rsp), %ecx ; SSE2_64_WIN-NEXT: movzwl {{[0-9]+}}(%rsp), %ecx
; SSE2_64_WIN-NEXT: xorl %eax, %eax
; SSE2_64_WIN-NEXT: orl $3072, %ecx # imm = 0xC00 ; SSE2_64_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE2_64_WIN-NEXT: movw %cx, {{[0-9]+}}(%rsp) ; SSE2_64_WIN-NEXT: movw %cx, {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp) ; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fistpll {{[0-9]+}}(%rsp) ; SSE2_64_WIN-NEXT: fistpll {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp) ; SSE2_64_WIN-NEXT: fldcw {{[0-9]+}}(%rsp)
; SSE2_64_WIN-NEXT: setbe %al
; SSE2_64_WIN-NEXT: shlq $63, %rax ; SSE2_64_WIN-NEXT: shlq $63, %rax
; SSE2_64_WIN-NEXT: xorq {{[0-9]+}}(%rsp), %rax ; SSE2_64_WIN-NEXT: xorq {{[0-9]+}}(%rsp), %rax
; SSE2_64_WIN-NEXT: addq $16, %rsp ; SSE2_64_WIN-NEXT: addq $16, %rsp
@ -1266,19 +1266,19 @@ define i64 @x_to_u64(x86_fp80 %a) nounwind {
; SSE2_64_LIN-NEXT: flds {{.*}}(%rip) ; SSE2_64_LIN-NEXT: flds {{.*}}(%rip)
; SSE2_64_LIN-NEXT: fld %st(1) ; SSE2_64_LIN-NEXT: fld %st(1)
; SSE2_64_LIN-NEXT: fsub %st(1), %st ; SSE2_64_LIN-NEXT: fsub %st(1), %st
; SSE2_64_LIN-NEXT: xorl %eax, %eax
; SSE2_64_LIN-NEXT: fxch %st(1) ; SSE2_64_LIN-NEXT: fxch %st(1)
; SSE2_64_LIN-NEXT: fucompi %st(2), %st ; SSE2_64_LIN-NEXT: fucompi %st(2), %st
; SSE2_64_LIN-NEXT: fcmovnbe %st(1), %st ; SSE2_64_LIN-NEXT: fcmovnbe %st(1), %st
; SSE2_64_LIN-NEXT: fstp %st(1) ; SSE2_64_LIN-NEXT: fstp %st(1)
; SSE2_64_LIN-NEXT: setbe %al
; SSE2_64_LIN-NEXT: fnstcw -{{[0-9]+}}(%rsp) ; SSE2_64_LIN-NEXT: fnstcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx ; SSE2_64_LIN-NEXT: movzwl -{{[0-9]+}}(%rsp), %ecx
; SSE2_64_LIN-NEXT: xorl %eax, %eax
; SSE2_64_LIN-NEXT: orl $3072, %ecx # imm = 0xC00 ; SSE2_64_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; SSE2_64_LIN-NEXT: movw %cx, -{{[0-9]+}}(%rsp) ; SSE2_64_LIN-NEXT: movw %cx, -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp) ; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fistpll -{{[0-9]+}}(%rsp) ; SSE2_64_LIN-NEXT: fistpll -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp) ; SSE2_64_LIN-NEXT: fldcw -{{[0-9]+}}(%rsp)
; SSE2_64_LIN-NEXT: setbe %al
; SSE2_64_LIN-NEXT: shlq $63, %rax ; SSE2_64_LIN-NEXT: shlq $63, %rax
; SSE2_64_LIN-NEXT: xorq -{{[0-9]+}}(%rsp), %rax ; SSE2_64_LIN-NEXT: xorq -{{[0-9]+}}(%rsp), %rax
; SSE2_64_LIN-NEXT: retq ; SSE2_64_LIN-NEXT: retq
@ -1304,15 +1304,15 @@ define i64 @x_to_u64(x86_fp80 %a) nounwind {
; X87_WIN-NEXT: fldz ; X87_WIN-NEXT: fldz
; X87_WIN-NEXT: LBB4_2: ; X87_WIN-NEXT: LBB4_2:
; X87_WIN-NEXT: fstp %st(0) ; X87_WIN-NEXT: fstp %st(0)
; X87_WIN-NEXT: setbe %al
; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X87_WIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_WIN-NEXT: xorl %edx, %edx ; X87_WIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_WIN-NEXT: orl $3072, %eax # imm = 0xC00 ; X87_WIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_WIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_WIN-NEXT: setbe %dl ; X87_WIN-NEXT: movzbl %al, %edx
; X87_WIN-NEXT: shll $31, %edx ; X87_WIN-NEXT: shll $31, %edx
; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X87_WIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; X87_WIN-NEXT: movl {{[0-9]+}}(%esp), %eax
@ -1338,15 +1338,15 @@ define i64 @x_to_u64(x86_fp80 %a) nounwind {
; X87_LIN-NEXT: fldz ; X87_LIN-NEXT: fldz
; X87_LIN-NEXT: .LBB4_2: ; X87_LIN-NEXT: .LBB4_2:
; X87_LIN-NEXT: fstp %st(0) ; X87_LIN-NEXT: fstp %st(0)
; X87_LIN-NEXT: setbe %al
; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fnstcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %eax ; X87_LIN-NEXT: movzwl {{[0-9]+}}(%esp), %ecx
; X87_LIN-NEXT: xorl %edx, %edx ; X87_LIN-NEXT: orl $3072, %ecx # imm = 0xC00
; X87_LIN-NEXT: orl $3072, %eax # imm = 0xC00 ; X87_LIN-NEXT: movw %cx, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: movw %ax, {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fistpll {{[0-9]+}}(%esp)
; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp) ; X87_LIN-NEXT: fldcw {{[0-9]+}}(%esp)
; X87_LIN-NEXT: setbe %dl ; X87_LIN-NEXT: movzbl %al, %edx
; X87_LIN-NEXT: shll $31, %edx ; X87_LIN-NEXT: shll $31, %edx
; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx ; X87_LIN-NEXT: xorl {{[0-9]+}}(%esp), %edx
; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax ; X87_LIN-NEXT: movl {{[0-9]+}}(%esp), %eax