From 6697e3354fbe51242729888529d047c646acdab3 Mon Sep 17 00:00:00 2001 From: Simon Pilgrim Date: Tue, 29 Mar 2022 22:24:02 +0100 Subject: [PATCH] [X86] combineADC - fold ADC(C1,C2,Carry) -> ADC(0,C1+C2,Carry) If we're not relying on the flag result, we can fold the constants together into the RHS immediate operand and set the LHS operand to zero, simplifying for further folds. We could do something similar if the flag result is in use and the constant fold doesn't affect it, but I don't have any real test cases for this yet. As suggested by @davezarzycki on Issue #35256 Differential Revision: https://reviews.llvm.org/D122482 --- llvm/lib/Target/X86/X86ISelLowering.cpp | 11 +++ llvm/test/CodeGen/X86/call-rv-marker.ll | 23 +++--- llvm/test/CodeGen/X86/combine-adc.ll | 17 +++-- llvm/test/CodeGen/X86/combine-add.ll | 5 +- llvm/test/CodeGen/X86/pr16031.ll | 19 +++-- .../CodeGen/X86/scheduler-backtracking.ll | 72 +++++++++---------- llvm/test/CodeGen/X86/setcc.ll | 8 +-- 7 files changed, 83 insertions(+), 72 deletions(-) diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 6c571c06a22f..712ef6195127 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -52305,6 +52305,17 @@ static SDValue combineADC(SDNode *N, SelectionDAG &DAG, return DCI.CombineTo(N, Res1, CarryOut); } + // Fold ADC(C1,C2,Carry) -> ADC(0,C1+C2,Carry) + // iff the flag result is dead. + // TODO: Allow flag result if C1+C2 doesn't signed/unsigned overflow. + if (LHSC && RHSC && !LHSC->isZero() && !N->hasAnyUseOfValue(1)) { + SDLoc DL(N); + APInt Sum = LHSC->getAPIntValue() + RHSC->getAPIntValue(); + return DAG.getNode(X86ISD::ADC, DL, N->getVTList(), + DAG.getConstant(0, DL, LHS.getValueType()), + DAG.getConstant(Sum, DL, LHS.getValueType()), CarryIn); + } + if (SDValue Flags = combineCarryThroughADD(CarryIn, DAG)) { MVT VT = N->getSimpleValueType(0); SDVTList VTs = DAG.getVTList(VT, MVT::i32); diff --git a/llvm/test/CodeGen/X86/call-rv-marker.ll b/llvm/test/CodeGen/X86/call-rv-marker.ll index ea63873d811e..0050a7d796ea 100644 --- a/llvm/test/CodeGen/X86/call-rv-marker.ll +++ b/llvm/test/CodeGen/X86/call-rv-marker.ll @@ -55,17 +55,18 @@ entry: define void @rv_marker_2_select(i32 %c) { ; CHECK-LABEL: rv_marker_2_select: -; CHECK: pushq %rax -; CHECK-NEXT: .cfi_def_cfa_offset 16 -; CHECK-NEXT: cmpl $1, %edi -; CHECK-NEXT: movl $1, %edi -; CHECK-NEXT: adcl $0, %edi -; CHECK-NEXT: callq _foo0 -; CHECK-NEXT: movq %rax, %rdi -; CHECK-NEXT: callq _objc_retainAutoreleasedReturnValue -; CHECK-NEXT: movq %rax, %rdi -; CHECK-NEXT: popq %rax -; CHECK-NEXT: jmp _foo2 +; CHECK: pushq %rax +; CHECK-NEXT: .cfi_def_cfa_offset 16 +; CHECK-NEXT: xorl %eax, %eax +; CHECK-NEXT: cmpl $1, %edi +; CHECK-NEXT: adcl $1, %eax +; CHECK-NEXT: movl %eax, %edi +; CHECK-NEXT: callq _foo0 +; CHECK-NEXT: movq %rax, %rdi +; CHECK-NEXT: callq _objc_retainAutoreleasedReturnValue +; CHECK-NEXT: movq %rax, %rdi +; CHECK-NEXT: popq %rax +; CHECK-NEXT: jmp _foo2 ; entry: %tobool.not = icmp eq i32 %c, 0 diff --git a/llvm/test/CodeGen/X86/combine-adc.ll b/llvm/test/CodeGen/X86/combine-adc.ll index 3a6e26cbd8c4..37f4fe195d80 100644 --- a/llvm/test/CodeGen/X86/combine-adc.ll +++ b/llvm/test/CodeGen/X86/combine-adc.ll @@ -67,26 +67,25 @@ define i32 @PR40483_add2(i32*, i32) nounwind { ret i32 %10 } -; FIXME: Fail to add (non-overflowing) constants together ; FIXME: Fail to convert add+lshr+and to BT define i32 @adc_merge_constants(i32 %a0) nounwind { ; X86-LABEL: adc_merge_constants: ; X86: # %bb.0: -; X86-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-NEXT: shrl $11, %eax -; X86-NEXT: andb $1, %al -; X86-NEXT: addb $-1, %al -; X86-NEXT: movl $55, %eax -; X86-NEXT: adcl $-1, %eax +; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx +; X86-NEXT: shrl $11, %ecx +; X86-NEXT: andb $1, %cl +; X86-NEXT: xorl %eax, %eax +; X86-NEXT: addb $-1, %cl +; X86-NEXT: adcl $54, %eax ; X86-NEXT: retl ; ; X64-LABEL: adc_merge_constants: ; X64: # %bb.0: ; X64-NEXT: shrl $11, %edi ; X64-NEXT: andb $1, %dil +; X64-NEXT: xorl %eax, %eax ; X64-NEXT: addb $-1, %dil -; X64-NEXT: movl $55, %eax -; X64-NEXT: adcl $-1, %eax +; X64-NEXT: adcl $54, %eax ; X64-NEXT: retq %bit = lshr i32 %a0, 11 %mask = and i32 %bit, 1 diff --git a/llvm/test/CodeGen/X86/combine-add.ll b/llvm/test/CodeGen/X86/combine-add.ll index 3d7fb1192083..fb60bc852189 100644 --- a/llvm/test/CodeGen/X86/combine-add.ll +++ b/llvm/test/CodeGen/X86/combine-add.ll @@ -438,9 +438,10 @@ define i1 @PR51238(i1 %b, i8 %x, i8 %y, i8 %z) { ; CHECK-LABEL: PR51238: ; CHECK: # %bb.0: ; CHECK-NEXT: notb %cl +; CHECK-NEXT: xorl %eax, %eax ; CHECK-NEXT: addb %dl, %cl -; CHECK-NEXT: movb $1, %al -; CHECK-NEXT: adcb $0, %al +; CHECK-NEXT: adcb $1, %al +; CHECK-NEXT: # kill: def $al killed $al killed $eax ; CHECK-NEXT: retq %ny = xor i8 %y, -1 %nz = xor i8 %z, -1 diff --git a/llvm/test/CodeGen/X86/pr16031.ll b/llvm/test/CodeGen/X86/pr16031.ll index eb8f400969ad..3aeaec1b9ce7 100644 --- a/llvm/test/CodeGen/X86/pr16031.ll +++ b/llvm/test/CodeGen/X86/pr16031.ll @@ -4,18 +4,17 @@ define i64 @main(i1 %tobool1) nounwind { ; CHECK-LABEL: main: ; CHECK: # %bb.0: # %entry -; CHECK-NEXT: pushl %esi ; CHECK-NEXT: testb $1, {{[0-9]+}}(%esp) -; CHECK-NEXT: movl $-12, %eax -; CHECK-NEXT: movl $-1, %ecx -; CHECK-NEXT: cmovel %ecx, %eax +; CHECK-NEXT: movl $-12, %ecx +; CHECK-NEXT: movl $-1, %eax +; CHECK-NEXT: cmovnel %ecx, %eax +; CHECK-NEXT: xorl %ecx, %ecx +; CHECK-NEXT: movl %eax, %edx +; CHECK-NEXT: addl $-1, %edx +; CHECK-NEXT: movl $0, %edx +; CHECK-NEXT: adcl $-2, %edx +; CHECK-NEXT: cmovsl %ecx, %eax ; CHECK-NEXT: xorl %edx, %edx -; CHECK-NEXT: movl %eax, %esi -; CHECK-NEXT: addl $-1, %esi -; CHECK-NEXT: adcl $-1, %ecx -; CHECK-NEXT: cmovsl %edx, %eax -; CHECK-NEXT: xorl %edx, %edx -; CHECK-NEXT: popl %esi ; CHECK-NEXT: retl entry: %0 = zext i1 %tobool1 to i32 diff --git a/llvm/test/CodeGen/X86/scheduler-backtracking.ll b/llvm/test/CodeGen/X86/scheduler-backtracking.ll index 36f17a6628ac..8c3d25fe565f 100644 --- a/llvm/test/CodeGen/X86/scheduler-backtracking.ll +++ b/llvm/test/CodeGen/X86/scheduler-backtracking.ll @@ -690,77 +690,77 @@ declare i256 @llvm.ctlz.i256(i256, i1) nounwind readnone define i64 @test4(i64 %a, i64 %b) nounwind { ; ILP-LABEL: test4: ; ILP: # %bb.0: +; ILP-NEXT: xorl %eax, %eax ; ILP-NEXT: xorl %ecx, %ecx -; ILP-NEXT: xorl %edx, %edx ; ILP-NEXT: incq %rsi -; ILP-NEXT: sete %dl -; ILP-NEXT: movl $2, %eax +; ILP-NEXT: sete %cl ; ILP-NEXT: cmpq %rdi, %rsi -; ILP-NEXT: sbbq $0, %rdx -; ILP-NEXT: movl $0, %edx -; ILP-NEXT: sbbq %rdx, %rdx +; ILP-NEXT: sbbq $0, %rcx +; ILP-NEXT: movl $0, %ecx ; ILP-NEXT: sbbq %rcx, %rcx -; ILP-NEXT: adcq $-1, %rax +; ILP-NEXT: movl $0, %ecx +; ILP-NEXT: sbbq %rcx, %rcx +; ILP-NEXT: adcq $1, %rax ; ILP-NEXT: retq ; ; HYBRID-LABEL: test4: ; HYBRID: # %bb.0: +; HYBRID-NEXT: xorl %eax, %eax ; HYBRID-NEXT: xorl %ecx, %ecx -; HYBRID-NEXT: xorl %edx, %edx ; HYBRID-NEXT: incq %rsi -; HYBRID-NEXT: sete %dl -; HYBRID-NEXT: movl $2, %eax +; HYBRID-NEXT: sete %cl ; HYBRID-NEXT: cmpq %rdi, %rsi -; HYBRID-NEXT: sbbq $0, %rdx -; HYBRID-NEXT: movl $0, %edx -; HYBRID-NEXT: sbbq %rdx, %rdx +; HYBRID-NEXT: sbbq $0, %rcx +; HYBRID-NEXT: movl $0, %ecx ; HYBRID-NEXT: sbbq %rcx, %rcx -; HYBRID-NEXT: adcq $-1, %rax +; HYBRID-NEXT: movl $0, %ecx +; HYBRID-NEXT: sbbq %rcx, %rcx +; HYBRID-NEXT: adcq $1, %rax ; HYBRID-NEXT: retq ; ; BURR-LABEL: test4: ; BURR: # %bb.0: +; BURR-NEXT: xorl %eax, %eax ; BURR-NEXT: xorl %ecx, %ecx -; BURR-NEXT: xorl %edx, %edx ; BURR-NEXT: incq %rsi -; BURR-NEXT: sete %dl -; BURR-NEXT: movl $2, %eax +; BURR-NEXT: sete %cl ; BURR-NEXT: cmpq %rdi, %rsi -; BURR-NEXT: sbbq $0, %rdx -; BURR-NEXT: movl $0, %edx -; BURR-NEXT: sbbq %rdx, %rdx +; BURR-NEXT: sbbq $0, %rcx +; BURR-NEXT: movl $0, %ecx ; BURR-NEXT: sbbq %rcx, %rcx -; BURR-NEXT: adcq $-1, %rax +; BURR-NEXT: movl $0, %ecx +; BURR-NEXT: sbbq %rcx, %rcx +; BURR-NEXT: adcq $1, %rax ; BURR-NEXT: retq ; ; SRC-LABEL: test4: ; SRC: # %bb.0: -; SRC-NEXT: xorl %eax, %eax -; SRC-NEXT: incq %rsi -; SRC-NEXT: sete %al ; SRC-NEXT: xorl %ecx, %ecx +; SRC-NEXT: incq %rsi +; SRC-NEXT: sete %cl +; SRC-NEXT: xorl %eax, %eax ; SRC-NEXT: cmpq %rdi, %rsi -; SRC-NEXT: sbbq $0, %rax -; SRC-NEXT: movl $0, %eax -; SRC-NEXT: sbbq %rax, %rax +; SRC-NEXT: sbbq $0, %rcx +; SRC-NEXT: movl $0, %ecx ; SRC-NEXT: sbbq %rcx, %rcx -; SRC-NEXT: movl $2, %eax -; SRC-NEXT: adcq $-1, %rax +; SRC-NEXT: movl $0, %ecx +; SRC-NEXT: sbbq %rcx, %rcx +; SRC-NEXT: adcq $1, %rax ; SRC-NEXT: retq ; ; LIN-LABEL: test4: ; LIN: # %bb.0: -; LIN-NEXT: movl $2, %eax +; LIN-NEXT: xorl %eax, %eax ; LIN-NEXT: xorl %ecx, %ecx -; LIN-NEXT: xorl %edx, %edx ; LIN-NEXT: incq %rsi -; LIN-NEXT: sete %dl +; LIN-NEXT: sete %cl ; LIN-NEXT: cmpq %rdi, %rsi -; LIN-NEXT: sbbq $0, %rdx -; LIN-NEXT: movl $0, %edx -; LIN-NEXT: sbbq %rdx, %rdx +; LIN-NEXT: sbbq $0, %rcx +; LIN-NEXT: movl $0, %ecx ; LIN-NEXT: sbbq %rcx, %rcx -; LIN-NEXT: adcq $-1, %rax +; LIN-NEXT: movl $0, %ecx +; LIN-NEXT: sbbq %rcx, %rcx +; LIN-NEXT: adcq $1, %rax ; LIN-NEXT: retq %r = zext i64 %b to i256 %u = add i256 %r, 1 diff --git a/llvm/test/CodeGen/X86/setcc.ll b/llvm/test/CodeGen/X86/setcc.ll index 2bbc9ffe6168..50f762de251b 100644 --- a/llvm/test/CodeGen/X86/setcc.ll +++ b/llvm/test/CodeGen/X86/setcc.ll @@ -46,10 +46,10 @@ define i64 @t3(i64 %x) nounwind readnone ssp { define i32 @t4(i32 %a) { ; CHECK-LABEL: t4: ; CHECK: ## %bb.0: -; CHECK-NEXT: movq _v4@GOTPCREL(%rip), %rax -; CHECK-NEXT: cmpl $1, (%rax) -; CHECK-NEXT: movw $1, %ax -; CHECK-NEXT: adcw $0, %ax +; CHECK-NEXT: movq _v4@GOTPCREL(%rip), %rcx +; CHECK-NEXT: xorl %eax, %eax +; CHECK-NEXT: cmpl $1, (%rcx) +; CHECK-NEXT: adcw $1, %ax ; CHECK-NEXT: shll $16, %eax ; CHECK-NEXT: retq %t0 = load i32, i32* @v4, align 4