x86-32: PUSHF/POPF use/def EFLAGS

[jfb]

Summary

As a side-quest for D6629 jvoung pointed out that I should use -verify-machineinstrs and this found a bug in x86-32's handling of EFLAGS for PUSHF/POPF. This patch fixes the use/def, and adds -verify-machineinstrs to all x86 tests which contain 'EFLAGS'. One exception: this patch leaves inline-asm-fpstack.ll as-is because it fails -verify-machineinstrs in a way unrelated to EFLAGS. This patch also modifies cmpxchg-clobber-flags.ll along the lines of what D6629 already does by also testing i386.

[t.p.northover]

Oh, well spotted. I think this looks fine.

Tim.

[jfb]

I filed a bug for the test that I left as failing: http://llvm.org/bugs/show_bug.cgi?id=21931

llvm/trunk/lib/Target/X86/X86InstrInfo.td

 def POP16rmr: I<0x8F, MRM0r, (outs GR16:$reg), (ins), "pop{w}\t$reg", [],
                 IIC_POP_REG>, OpSize16;
 def POP16rmm: I<0x8F, MRM0m, (outs), (ins i16mem:$dst), "pop{w}\t$dst", [],
                 IIC_POP_MEM>, OpSize16;
 def POP32rmr: I<0x8F, MRM0r, (outs GR32:$reg), (ins), "pop{l}\t$reg", [],
                 IIC_POP_REG>, OpSize32, Requires<[Not64BitMode]>;
 def POP32rmm: I<0x8F, MRM0m, (outs), (ins i32mem:$dst), "pop{l}\t$dst", [],
                 IIC_POP_MEM>, OpSize32, Requires<[Not64BitMode]>;
-
-def POPF16   : I<0x9D, RawFrm, (outs), (ins), "popf{w}", [], IIC_POP_F>,
-                OpSize16;
-def POPF32   : I<0x9D, RawFrm, (outs), (ins), "popf{l|d}", [], IIC_POP_FD>,
-                OpSize32, Requires<[Not64BitMode]>;
 } // mayLoad, SchedRW
 
 let mayStore = 1, SchedRW = [WriteStore] in {
 def PUSH16r  : I<0x50, AddRegFrm, (outs), (ins GR16:$reg), "push{w}\t$reg",[],
                  IIC_PUSH_REG>, OpSize16;
 def PUSH32r  : I<0x50, AddRegFrm, (outs), (ins GR32:$reg), "push{l}\t$reg",[],
                  IIC_PUSH_REG>, OpSize32, Requires<[Not64BitMode]>;
 def PUSH16rmr: I<0xFF, MRM6r, (outs), (ins GR16:$reg), "push{w}\t$reg",[],
def PUSH32i8 : Ii8<0x6a, RawFrm, (outs), (ins i32i8imm:$imm),
                    "push{l}\t$imm", [], IIC_PUSH_IMM>, OpSize32,
                    Requires<[Not64BitMode]>;
 def PUSHi16  : Ii16<0x68, RawFrm, (outs), (ins i16imm:$imm),
                    "push{w}\t$imm", [], IIC_PUSH_IMM>, OpSize16,
                    Requires<[Not64BitMode]>;
 def PUSHi32  : Ii32<0x68, RawFrm, (outs), (ins i32imm:$imm),
                    "push{l}\t$imm", [], IIC_PUSH_IMM>, OpSize32,
                    Requires<[Not64BitMode]>;
+} // mayStore, SchedRW
+}
+
+let Defs = [ESP, EFLAGS], Uses = [ESP], mayLoad = 1, hasSideEffects=0,
+    SchedRW = [WriteLoad] in {
+def POPF16   : I<0x9D, RawFrm, (outs), (ins), "popf{w}", [], IIC_POP_F>,
+                OpSize16;
+def POPF32   : I<0x9D, RawFrm, (outs), (ins), "popf{l|d}", [], IIC_POP_FD>,
+                OpSize32, Requires<[Not64BitMode]>;
+}
 
+let Defs = [ESP], Uses = [ESP, EFLAGS], mayStore = 1, hasSideEffects=0,
+    SchedRW = [WriteStore] in {
 def PUSHF16  : I<0x9C, RawFrm, (outs), (ins), "pushf{w}", [], IIC_PUSH_F>,
                  OpSize16;
 def PUSHF32  : I<0x9C, RawFrm, (outs), (ins), "pushf{l|d}", [], IIC_PUSH_F>,
                OpSize32, Requires<[Not64BitMode]>;
-
-} // mayStore, SchedRW
 }
 
 let Defs = [RSP], Uses = [RSP], hasSideEffects=0 in {
 let mayLoad = 1, SchedRW = [WriteLoad] in {
 def POP64r   : I<0x58, AddRegFrm, (outs GR64:$reg), (ins), "pop{q}\t$reg", [],
                  IIC_POP_REG>, OpSize32, Requires<[In64BitMode]>;
 def POP64rmr: I<0x8F, MRM0r, (outs GR64:$reg), (ins), "pop{q}\t$reg", [],
                 IIC_POP_REG>, OpSize32, Requires<[In64BitMode]>;

llvm/trunk/test/CodeGen/X86/2012-01-16-mfence-nosse-flags.ll

-; RUN: llc < %s -mtriple=i686-linux -mattr=-sse | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mtriple=i686-linux -mattr=-sse | FileCheck %s
 ; PR11768
 
 @ptr = external global i8*
 
 define void @baz() nounwind ssp {
 entry:
   %0 = load i8** @ptr, align 4
   %cmp = icmp eq i8* %0, null

llvm/trunk/test/CodeGen/X86/clobber-fi0.ll

-; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mcpu=generic -mtriple=x86_64-linux | FileCheck %s
 
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.7.0"
 
 ; In the code below we need to copy the EFLAGS because of scheduling constraints.
 ; When copying the EFLAGS we need to write to the stack with push/pop. This forces
 ; us to emit the prolog.
 

llvm/trunk/test/CodeGen/X86/cmov.ll

-; RUN: llc < %s -mtriple=x86_64-apple-darwin10 -disable-cgp-select2branch | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mtriple=x86_64-apple-darwin10 -disable-cgp-select2branch | FileCheck %s
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
 
 define i32 @test1(i32 %x, i32 %n, i32 %w, i32* %vp) nounwind readnone {
 entry:
 ; CHECK-LABEL: test1:
 ; CHECK: btl
 ; CHECK-NEXT: movl  $12, %eax
 ; CHECK-NEXT: cmovael (%rcx), %eax

llvm/trunk/test/CodeGen/X86/cmpxchg-clobber-flags.ll

-; RUN: llc -mtriple=x86_64-linux-gnu %s -o - | FileCheck %s
-; RUN: llc -mtriple=x86_64-linux-gnu -pre-RA-sched=fast %s -o - | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=i386-linux-gnu %s -o - | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=i386-linux-gnu -pre-RA-sched=fast %s -o - | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=x86_64-linux-gnu %s -o - | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=x86_64-linux-gnu -pre-RA-sched=fast %s -o - | FileCheck %s
 
 declare i32 @bar()
 
 define i64 @test_intervening_call(i64* %foo, i64 %bar, i64 %baz) {
 ; CHECK-LABEL: test_intervening_call:
 ; CHECK: cmpxchg
-; CHECK: pushfq
-; CHECK: popq [[FLAGS:%.*]]
+; CHECK: pushf[[LQ:[lq]]]
+; CHECK-NEXT: pop[[LQ]] [[FLAGS:%.*]]
 
-; CHECK: callq bar
+; CHECK-NEXT: call[[LQ]] bar
 
-; CHECK: pushq [[FLAGS]]
-; CHECK: popfq
-; CHECK: jne
+; CHECK-NEXT: push[[LQ]] [[FLAGS]]
+; CHECK-NEXT: popf[[LQ]]
+; CHECK-NEXT: jne
   %cx = cmpxchg i64* %foo, i64 %bar, i64 %baz seq_cst seq_cst
   %p = extractvalue { i64, i1 } %cx, 1
   call i32 @bar()
   br i1 %p, label %t, label %f
 
 t:
   ret i64 42
 

 }
 
 ; This one is an interesting case because CMOV doesn't have a chain
 ; operand. Naive attempts to limit cmpxchg EFLAGS use are likely to fail here.
 define i32 @test_feed_cmov(i32* %addr, i32 %desired, i32 %new) {
 ; CHECK-LABEL: test_feed_cmov:
 
 ; CHECK: cmpxchg
-; CHECK: pushfq
-; CHECK: popq [[FLAGS:%.*]]
-
-; CHECK: callq bar
+; CHECK: pushf[[LQ:[lq]]]
+; CHECK-NEXT: pop[[LQ]] [[FLAGS:%.*]]
 
-; CHECK: pushq [[FLAGS]]
-; CHECK: popfq
+; CHECK-NEXT: call[[LQ]] bar
 
+; CHECK-NEXT: push[[LQ]] [[FLAGS]]
+; CHECK-NEXT: popf[[LQ]]
   %res = cmpxchg i32* %addr, i32 %desired, i32 %new seq_cst seq_cst
   %success = extractvalue { i32, i1 } %res, 1
 
   %rhs = call i32 @bar()
 
   %ret = select i1 %success, i32 %new, i32 %rhs
   ret i32 %ret
 }

llvm/trunk/test/CodeGen/X86/coalescer-dce.ll

-; RUN: llc < %s -disable-fp-elim -disable-machine-dce -verify-coalescing
+; RUN: llc < %s -verify-machineinstrs -disable-fp-elim -disable-machine-dce -verify-coalescing
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-macosx10.7.0"
 
 ; This test case has a sub-register join followed by a remat:
 ;
 ; 256L    %vreg2<def> = COPY %vreg7:sub_32bit<kill>; GR32:%vreg2 GR64:%vreg7
 ;         Considering merging %vreg2 with %vreg7:sub_32bit
 ;         Cross-class to GR64.

llvm/trunk/test/CodeGen/X86/misched-copy.ll

 ; REQUIRES: asserts
-; RUN: llc < %s -march=x86 -mcpu=core2 -pre-RA-sched=source -enable-misched -verify-misched -debug-only=misched -o - 2>&1 > /dev/null | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -march=x86 -mcpu=core2 -pre-RA-sched=source -enable-misched -verify-misched -debug-only=misched -o - 2>&1 > /dev/null | FileCheck %s
 ;
 ; Test scheduling of copy instructions.
 ;
 ; Argument copies should be hoisted to the top of the block.
 ; Return copies should be sunk to the end.
 ; MUL_HiLo PhysReg use copies should be just above the mul.
 ; MUL_HiLo PhysReg def copies should be just below the mul.
 ;

llvm/trunk/test/CodeGen/X86/misched-crash.ll

-; RUN: llc < %s -enable-misched -verify-misched
+; RUN: llc < %s -verify-machineinstrs -enable-misched -verify-misched
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10"
 
 ; This function contains a cmp instruction with two users.
 ; Hoisting the last use requires trimming the EFLAGS live range to the second.
 define void @rdar13353090(i8* %plane, i64 %_x1, i64 %_x2) {
 entry:
   %cmp = icmp ult i64 %_x1, %_x2

llvm/trunk/test/CodeGen/X86/norex-subreg.ll

-; RUN: llc -O0 < %s
-; RUN: llc < %s
+; RUN: llc -O0 < %s -verify-machineinstrs
+; RUN: llc < %s -verify-machineinstrs
 target triple = "x86_64-apple-macosx10.7"
 
 ; This test case extracts a sub_8bit_hi sub-register:
 ;
 ; %R8B<def> = COPY %BH, %EBX<imp-use,kill>
 ; %ESI<def> = MOVZX32_NOREXrr8 %R8B<kill>
 ;
 ; The register allocation above is invalid, %BH can only be encoded without an

llvm/trunk/test/CodeGen/X86/peep-test-2.ll

-; RUN: llc < %s -march=x86 | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -march=x86 | FileCheck %s
 
 ; CHECK: testl
 
 ; It's tempting to eliminate the testl instruction here and just use the
 ; EFLAGS value from the incl, however it can't be known whether the add
 ; will overflow, and if it does the incl would set OF, and the
 ; subsequent setg would return true.
 

llvm/trunk/test/CodeGen/X86/phys_subreg_coalesce-3.ll

-; RUN: llc < %s -mtriple=i386-apple-darwin -mcpu=corei7 | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -mtriple=i386-apple-darwin -mcpu=corei7 | FileCheck %s
 ; rdar://5571034
 
 ; This requires physreg joining, %vreg13 is live everywhere:
 ; 304L    %CL<def> = COPY %vreg13:sub_8bit; GR32_ABCD:%vreg13
 ; 320L    %vreg15<def> = COPY %vreg19; GR32:%vreg15 GR32_NOSP:%vreg19
 ; 336L    %vreg15<def> = SAR32rCL %vreg15, %EFLAGS<imp-def,dead>, %CL<imp-use,kill>; GR32:%vreg15
 
 define void @foo(i32* nocapture %quadrant, i32* nocapture %ptr, i32 %bbSize, i32 %bbStart, i32 %shifts) nounwind ssp {

llvm/trunk/test/CodeGen/X86/pre-ra-sched.ll

-; RUN-disabled: llc < %s -mtriple=x86_64-apple-macosx -pre-RA-sched=ilp -debug-only=pre-RA-sched \
+; RUN-disabled: llc < %s -verify-machineinstrs -mtriple=x86_64-apple-macosx -pre-RA-sched=ilp -debug-only=pre-RA-sched \
 ; RUN-disabled:     2>&1 | FileCheck %s
 ; RUN: true
 ; REQUIRES: asserts
 ;
 ; rdar:13279013: pre-RA-sched should not check all interferences and
 ; repush them on the ready queue after scheduling each instruction.
 ;
 ; CHECK: *** List Scheduling

llvm/trunk/test/CodeGen/X86/remat-phys-dead.ll

 ; REQUIRES: asserts
-; RUN: llc -mtriple=x86_64-apple-darwin -debug -o /dev/null < %s 2>&1 | FileCheck %s
+; RUN: llc -verify-machineinstrs -mtriple=x86_64-apple-darwin -debug -o /dev/null < %s 2>&1 | FileCheck %s
 
 ; We need to make sure that rematerialization into a physical register marks the
 ; super- or sub-register as dead after this rematerialization since only the
 ; original register is actually used later. Largely irrelevant for a trivial
 ; example like this, since EAX is never used again, but easy to test.
 
 define i8 @test_remat() {
   ret i8 0

llvm/trunk/test/CodeGen/X86/sink-hoist.ll

-; RUN: llc < %s -march=x86-64 -asm-verbose=false -mtriple=x86_64-unknown-linux-gnu -mcpu=nehalem -post-RA-scheduler=true -schedmodel=false | FileCheck %s
+; RUN: llc < %s -verify-machineinstrs -march=x86-64 -asm-verbose=false -mtriple=x86_64-unknown-linux-gnu -mcpu=nehalem -post-RA-scheduler=true -schedmodel=false | FileCheck %s
 
 ; Currently, floating-point selects are lowered to CFG triangles.
 ; This means that one side of the select is always unconditionally
 ; evaluated, however with MachineSink we can sink the other side so
 ; that it's conditionally evaluated.
 
 ; CHECK-LABEL: foo:
 ; CHECK-NEXT: testb $1, %dil

llvm/trunk/test/CodeGen/X86/vaargs.ll

-; RUN: llc -mcpu=corei7-avx %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=NO-FLAGS
+; RUN: llc -verify-machineinstrs -mcpu=corei7-avx %s -o - | FileCheck %s --check-prefix=CHECK --check-prefix=NO-FLAGS
 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
 target triple = "x86_64-apple-macosx10.9.0"
 
 %struct.__va_list_tag = type { i32, i32, i8*, i8* }
 
 ; Check that vastart gets the right thing.
 define i32 @sum(i32 %count, ...) nounwind optsize ssp uwtable {
 ; CHECK:      testb   %al, %al