[TwoAddressInstruction] Improve tests for register killed by instruction

Define and use a MachineOperand overload of isPlainlyKilled. This
improves codegen in a couple of tests because it catches the case where
MO does not kill Reg but another operand of the same instruction does.

Differential Revision: https://reviews.llvm.org/D147167

diff --git a/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp b/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp
index 8b0751f..256c121 100644 (file)
--- a/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp
+++ b/llvm/lib/CodeGen/TwoAddressInstructionPass.cpp
@@ -198,8 +198,6 @@ INITIALIZE_PASS_DEPENDENCY(AAResultsWrapperPass)
 INITIALIZE_PASS_END(TwoAddressInstructionPass, DEBUG_TYPE,
                 "Two-Address instruction pass", false, false)
 
-static bool isPlainlyKilled(MachineInstr *MI, Register Reg, LiveIntervals *LIS);
-
 /// Return the MachineInstr* if it is the single def of the Reg in current BB.
 static MachineInstr *getSingleDef(Register Reg, MachineBasicBlock *BB,
                                   const MachineRegisterInfo *MRI) {
@@ -287,7 +285,7 @@ static bool isCopyToReg(MachineInstr &MI, const TargetInstrInfo *TII,
 
 /// Test if the given register value, which is used by the
 /// given instruction, is killed by the given instruction.
-static bool isPlainlyKilled(MachineInstr *MI, Register Reg,
+static bool isPlainlyKilled(const MachineInstr *MI, Register Reg,
                             LiveIntervals *LIS) {
   if (LIS && Reg.isVirtual() && !LIS->isNotInMIMap(*MI)) {
     // FIXME: Sometimes tryInstructionTransform() will add instructions and
@@ -311,6 +309,12 @@ static bool isPlainlyKilled(MachineInstr *MI, Register Reg,
   return MI->killsRegister(Reg);
 }
 
+/// Test if the register used by the given operand is killed by the operand's
+/// instruction.
+static bool isPlainlyKilled(const MachineOperand &MO, LiveIntervals *LIS) {
+  return MO.isKill() || isPlainlyKilled(MO.getParent(), MO.getReg(), LIS);
+}
+
 /// Test if the given register value, which is used by the given
 /// instruction, is killed by the given instruction. This looks through
 /// coalescable copies to see if the original value is potentially not killed.
@@ -861,8 +865,7 @@ bool TwoAddressInstructionPass::rescheduleMIBelowKill(
       Defs.push_back(MOReg);
     else {
       Uses.push_back(MOReg);
-      if (MOReg != Reg && (MO.isKill() ||
-                           (LIS && isPlainlyKilled(MI, MOReg, LIS))))
+      if (MOReg != Reg && isPlainlyKilled(MO, LIS))
         Kills.push_back(MOReg);
     }
   }
@@ -913,8 +916,7 @@ bool TwoAddressInstructionPass::rescheduleMIBelowKill(
       } else {
         if (regOverlapsSet(Defs, MOReg, TRI))
           return false;
-        bool isKill =
-            MO.isKill() || (LIS && isPlainlyKilled(&OtherMI, MOReg, LIS));
+        bool isKill = isPlainlyKilled(MO, LIS);
         if (MOReg != Reg && ((isKill && regOverlapsSet(Uses, MOReg, TRI)) ||
                              regOverlapsSet(Kills, MOReg, TRI)))
           // Don't want to extend other live ranges and update kills.
@@ -1042,7 +1044,7 @@ bool TwoAddressInstructionPass::rescheduleKillAboveMI(
         continue;
       if (isDefTooClose(MOReg, DI->second, MI))
         return false;
-      bool isKill = MO.isKill() || (LIS && isPlainlyKilled(KillMI, MOReg, LIS));
+      bool isKill = isPlainlyKilled(MO, LIS);
       if (MOReg == Reg && !isKill)
         return false;
       Uses.push_back(MOReg);
@@ -1084,8 +1086,7 @@ bool TwoAddressInstructionPass::rescheduleKillAboveMI(
         if (regOverlapsSet(Kills, MOReg, TRI))
           // Don't want to extend other live ranges and update kills.
           return false;
-        if (&OtherMI != MI && MOReg == Reg &&
-            !(MO.isKill() || (LIS && isPlainlyKilled(&OtherMI, MOReg, LIS))))
+        if (&OtherMI != MI && MOReg == Reg && !isPlainlyKilled(MO, LIS))
           // We can't schedule across a use of the register in question.
           return false;
       } else {

diff --git a/llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll b/llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll
index 88a0b06..bb4f131 100644 (file)
--- a/llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll
+++ b/llvm/test/CodeGen/AArch64/sve-fixed-length-limit-duplane.ll
@@ -6,15 +6,14 @@ target triple = "aarch64-unknown-linux-gnu"
 define <4 x i32> @test(ptr %arg1, ptr %arg2) {
 ; CHECK-LABEL: test:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    mov x8, #8
+; CHECK-NEXT:    mov x8, #8 // =0x8
 ; CHECK-NEXT:    ptrue p0.s, vl8
-; CHECK-NEXT:    ld1w { z1.s }, p0/z, [x0, x8, lsl #2]
+; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
 ; CHECK-NEXT:    ld1w { z2.s }, p0/z, [x0]
-; CHECK-NEXT:    mov z0.d, z1.d
-; CHECK-NEXT:    add z2.s, z2.s, z2.s
-; CHECK-NEXT:    ext z0.b, z0.b, z1.b, #16
-; CHECK-NEXT:    add z1.s, z1.s, z1.s
+; CHECK-NEXT:    add z1.s, z0.s, z0.s
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #16
 ; CHECK-NEXT:    dup v0.4s, v0.s[2]
+; CHECK-NEXT:    add z2.s, z2.s, z2.s
 ; CHECK-NEXT:    st1w { z1.s }, p0, [x0, x8, lsl #2]
 ; CHECK-NEXT:    st1w { z2.s }, p0, [x0]
 ; CHECK-NEXT:    ret
@@ -30,15 +29,14 @@ entry:
 define <2 x i32> @test2(ptr %arg1, ptr %arg2) {
 ; CHECK-LABEL: test2:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    mov x8, #8
+; CHECK-NEXT:    mov x8, #8 // =0x8
 ; CHECK-NEXT:    ptrue p0.s, vl8
-; CHECK-NEXT:    ld1w { z1.s }, p0/z, [x0, x8, lsl #2]
+; CHECK-NEXT:    ld1w { z0.s }, p0/z, [x0, x8, lsl #2]
 ; CHECK-NEXT:    ld1w { z2.s }, p0/z, [x0]
-; CHECK-NEXT:    mov z0.d, z1.d
-; CHECK-NEXT:    add z2.s, z2.s, z2.s
-; CHECK-NEXT:    ext z0.b, z0.b, z1.b, #24
-; CHECK-NEXT:    add z1.s, z1.s, z1.s
+; CHECK-NEXT:    add z1.s, z0.s, z0.s
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #24
 ; CHECK-NEXT:    dup v0.2s, v0.s[0]
+; CHECK-NEXT:    add z2.s, z2.s, z2.s
 ; CHECK-NEXT:    st1w { z1.s }, p0, [x0, x8, lsl #2]
 ; CHECK-NEXT:    st1w { z2.s }, p0, [x0]
 ; CHECK-NEXT:    ret

diff --git a/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-limit-duplane.ll b/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-limit-duplane.ll
index c752b1a..943098d 100644 (file)
--- a/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-limit-duplane.ll
+++ b/llvm/test/CodeGen/AArch64/sve-streaming-mode-fixed-length-limit-duplane.ll
@@ -29,17 +29,16 @@ entry:
 define <2 x i32> @test2(ptr %arg1, ptr %arg2) {
 ; CHECK-LABEL: test2:
 ; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    ldp q2, q0, [x0, #32]
-; CHECK-NEXT:    ldp q4, q5, [x0]
-; CHECK-NEXT:    mov z1.d, z0.d
-; CHECK-NEXT:    add z3.s, z0.s, z0.s
-; CHECK-NEXT:    ext z1.b, z1.b, z0.b, #8
-; CHECK-NEXT:    mov z0.s, s1
-; CHECK-NEXT:    add z1.s, z2.s, z2.s
-; CHECK-NEXT:    stp q1, q3, [x0, #32]
-; CHECK-NEXT:    add z1.s, z4.s, z4.s
+; CHECK-NEXT:    ldp q1, q0, [x0, #32]
+; CHECK-NEXT:    add z1.s, z1.s, z1.s
+; CHECK-NEXT:    ldp q2, q3, [x0]
+; CHECK-NEXT:    add z4.s, z0.s, z0.s
+; CHECK-NEXT:    ext z0.b, z0.b, z0.b, #8
+; CHECK-NEXT:    mov z0.s, s0
+; CHECK-NEXT:    stp q1, q4, [x0, #32]
 ; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $z0
-; CHECK-NEXT:    add z2.s, z5.s, z5.s
+; CHECK-NEXT:    add z1.s, z2.s, z2.s
+; CHECK-NEXT:    add z2.s, z3.s, z3.s
 ; CHECK-NEXT:    stp q1, q2, [x0]
 ; CHECK-NEXT:    ret
 entry:

diff --git a/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll b/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll
index 004f9ab..38fdc6a 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll
+++ b/llvm/test/CodeGen/AMDGPU/mad-mix-lo.ll
@@ -1190,24 +1190,22 @@ define <2 x half> @v_mad_mix_v2f32_clamp_postcvt_lo(<2 x half> %src0, <2 x half>
 ; GISEL-GFX900:       ; %bb.0:
 ; GISEL-GFX900-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GISEL-GFX900-NEXT:    v_mad_mixlo_f16 v3, v0, v1, v2 op_sel_hi:[1,1,1]
-; GISEL-GFX900-NEXT:    v_mov_b32_e32 v4, v3
-; GISEL-GFX900-NEXT:    v_mad_mixhi_f16 v4, v0, v1, v2 op_sel:[1,1,1] op_sel_hi:[1,1,1]
-; GISEL-GFX900-NEXT:    v_max_f16_e64 v0, v3, v3 clamp
-; GISEL-GFX900-NEXT:    v_bfe_u32 v0, v0, 0, 16
+; GISEL-GFX900-NEXT:    v_max_f16_e64 v4, v3, v3 clamp
+; GISEL-GFX900-NEXT:    v_mad_mixhi_f16 v3, v0, v1, v2 op_sel:[1,1,1] op_sel_hi:[1,1,1]
+; GISEL-GFX900-NEXT:    v_bfe_u32 v0, v4, 0, 16
 ; GISEL-GFX900-NEXT:    v_mov_b32_e32 v1, 0xffff0000
-; GISEL-GFX900-NEXT:    v_and_or_b32 v0, v4, v1, v0
+; GISEL-GFX900-NEXT:    v_and_or_b32 v0, v3, v1, v0
 ; GISEL-GFX900-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GISEL-GFX906-LABEL: v_mad_mix_v2f32_clamp_postcvt_lo:
 ; GISEL-GFX906:       ; %bb.0:
 ; GISEL-GFX906-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
 ; GISEL-GFX906-NEXT:    v_fma_mixlo_f16 v3, v0, v1, v2 op_sel_hi:[1,1,1]
-; GISEL-GFX906-NEXT:    v_mov_b32_e32 v4, v3
-; GISEL-GFX906-NEXT:    v_fma_mixhi_f16 v4, v0, v1, v2 op_sel:[1,1,1] op_sel_hi:[1,1,1]
-; GISEL-GFX906-NEXT:    v_max_f16_e64 v0, v3, v3 clamp
-; GISEL-GFX906-NEXT:    v_bfe_u32 v0, v0, 0, 16
+; GISEL-GFX906-NEXT:    v_max_f16_e64 v4, v3, v3 clamp
+; GISEL-GFX906-NEXT:    v_fma_mixhi_f16 v3, v0, v1, v2 op_sel:[1,1,1] op_sel_hi:[1,1,1]
+; GISEL-GFX906-NEXT:    v_bfe_u32 v0, v4, 0, 16
 ; GISEL-GFX906-NEXT:    v_mov_b32_e32 v1, 0xffff0000
-; GISEL-GFX906-NEXT:    v_and_or_b32 v0, v4, v1, v0
+; GISEL-GFX906-NEXT:    v_and_or_b32 v0, v3, v1, v0
 ; GISEL-GFX906-NEXT:    s_setpc_b64 s[30:31]
 ;
 ; GISEL-VI-LABEL: v_mad_mix_v2f32_clamp_postcvt_lo:

diff --git a/llvm/test/CodeGen/AVR/shift32.ll b/llvm/test/CodeGen/AVR/shift32.ll
index 5ccc80d..820419b 100644 (file)
--- a/llvm/test/CodeGen/AVR/shift32.ll
+++ b/llvm/test/CodeGen/AVR/shift32.ll
@@ -525,10 +525,10 @@ define i32 @ashr_i32_22(i32 %a) {
 ; CHECK-NEXT:    sbc r18, r18
 ; CHECK-NEXT:    lsl r24
 ; CHECK-NEXT:    rol r25
+; CHECK-NEXT:    mov r19, r18
 ; CHECK-NEXT:    mov r23, r18
 ; CHECK-NEXT:    rol r23
 ; CHECK-NEXT:    mov r22, r25
-; CHECK-NEXT:    mov r19, r18
 ; CHECK-NEXT:    movw r24, r18
 ; CHECK-NEXT:    ret
   %res = ashr i32 %a, 22

diff --git a/llvm/test/CodeGen/X86/pr32284.ll b/llvm/test/CodeGen/X86/pr32284.ll
index 3c258a4..90fb76a 100644 (file)
--- a/llvm/test/CodeGen/X86/pr32284.ll
+++ b/llvm/test/CodeGen/X86/pr32284.ll
@@ -504,18 +504,18 @@ define void @f3() #0 {
 ; X86-NEXT:    .cfi_def_cfa_register %ebp
 ; X86-NEXT:    andl $-8, %esp
 ; X86-NEXT:    subl $8, %esp
-; X86-NEXT:    movl var_13, %ecx
-; X86-NEXT:    xorl %eax, %eax
-; X86-NEXT:    testl %ecx, %ecx
-; X86-NEXT:    notl %ecx
-; X86-NEXT:    sete %al
+; X86-NEXT:    movl var_13, %eax
+; X86-NEXT:    xorl %ecx, %ecx
+; X86-NEXT:    testl %eax, %eax
+; X86-NEXT:    notl %eax
+; X86-NEXT:    sete %cl
 ; X86-NEXT:    movl var_16, %edx
-; X86-NEXT:    xorl %ecx, %edx
-; X86-NEXT:    andl %eax, %edx
-; X86-NEXT:    orl %ecx, %edx
+; X86-NEXT:    xorl %eax, %edx
+; X86-NEXT:    andl %ecx, %edx
+; X86-NEXT:    orl %eax, %edx
 ; X86-NEXT:    movl %edx, (%esp)
 ; X86-NEXT:    movl $0, {{[0-9]+}}(%esp)
-; X86-NEXT:    movl %ecx, var_46
+; X86-NEXT:    movl %eax, var_46
 ; X86-NEXT:    movl %ebp, %esp
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    .cfi_def_cfa %esp, 4