From f8fcee70a30475ca6a8f960d2569db05f91c0332 Mon Sep 17 00:00:00 2001 From: Craig Topper Date: Tue, 24 Jul 2018 23:44:12 +0000 Subject: [PATCH] [X86] Change multiply by 26 to use two multiplies by 5 and an add instead of multiply by 3 and 9 and a subtract. Same number of operations, but ending in an add is friendlier due to it being commutable. llvm-svn: 337869 --- llvm/lib/Target/X86/X86ISelLowering.cpp | 14 +++++++------- llvm/test/CodeGen/X86/mul-constant-i16.ll | 12 ++++++------ llvm/test/CodeGen/X86/mul-constant-i32.ll | 18 +++++++++--------- llvm/test/CodeGen/X86/mul-constant-i64.ll | 18 +++++++++--------- llvm/test/CodeGen/X86/mul-constant-result.ll | 16 ++++++++++------ 5 files changed, 41 insertions(+), 37 deletions(-) diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp index 12f6d40..352e817 100644 --- a/llvm/lib/Target/X86/X86ISelLowering.cpp +++ b/llvm/lib/Target/X86/X86ISelLowering.cpp @@ -33698,10 +33698,10 @@ static SDValue combineMulSpecial(uint64_t MulAmt, SDNode *N, SelectionDAG &DAG, return Result; }; - auto combineMulMulAddOrSub = [&](bool isAdd) { + auto combineMulMulAddOrSub = [&](int Mul1, int Mul2, bool isAdd) { SDValue Result = DAG.getNode(X86ISD::MUL_IMM, DL, VT, N->getOperand(0), - DAG.getConstant(9, DL, VT)); - Result = DAG.getNode(ISD::MUL, DL, VT, Result, DAG.getConstant(3, DL, VT)); + DAG.getConstant(Mul1, DL, VT)); + Result = DAG.getNode(ISD::MUL, DL, VT, Result, DAG.getConstant(Mul2, DL, VT)); Result = DAG.getNode(isAdd ? ISD::ADD : ISD::SUB, DL, VT, Result, N->getOperand(0)); return Result; @@ -33730,15 +33730,15 @@ static SDValue combineMulSpecial(uint64_t MulAmt, SDNode *N, SelectionDAG &DAG, // mul x, 23 => sub ((shl (mul x, 3), 3), x) return combineMulShlAddOrSub(3, 3, /*isAdd*/ false); case 26: - // mul x, 26 => sub ((mul (mul x, 9), 3), x) - return combineMulMulAddOrSub(/*isAdd*/ false); + // mul x, 26 => add ((mul (mul x, 5), 5), x) + return combineMulMulAddOrSub(5, 5, /*isAdd*/ true); case 28: // mul x, 28 => add ((mul (mul x, 9), 3), x) - return combineMulMulAddOrSub(/*isAdd*/ true); + return combineMulMulAddOrSub(9, 3, /*isAdd*/ true); case 29: // mul x, 29 => add (add ((mul (mul x, 9), 3), x), x) return DAG.getNode(ISD::ADD, DL, VT, N->getOperand(0), - combineMulMulAddOrSub(/*isAdd*/ true)); + combineMulMulAddOrSub(9, 3, /*isAdd*/ true)); } return SDValue(); } diff --git a/llvm/test/CodeGen/X86/mul-constant-i16.ll b/llvm/test/CodeGen/X86/mul-constant-i16.ll index 044e7ce..83980f9 100644 --- a/llvm/test/CodeGen/X86/mul-constant-i16.ll +++ b/llvm/test/CodeGen/X86/mul-constant-i16.ll @@ -494,18 +494,18 @@ define i16 @test_mul_by_26(i16 %x) { ; X86-LABEL: test_mul_by_26: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: leal (%ecx,%ecx,8), %eax -; X86-NEXT: leal (%eax,%eax,2), %eax -; X86-NEXT: subl %ecx, %eax +; X86-NEXT: leal (%ecx,%ecx,4), %eax +; X86-NEXT: leal (%eax,%eax,4), %eax +; X86-NEXT: addl %ecx, %eax ; X86-NEXT: # kill: def $ax killed $ax killed $eax ; X86-NEXT: retl ; ; X64-LABEL: test_mul_by_26: ; X64: # %bb.0: ; X64-NEXT: # kill: def $edi killed $edi def $rdi -; X64-NEXT: leal (%rdi,%rdi,8), %eax -; X64-NEXT: leal (%rax,%rax,2), %eax -; X64-NEXT: subl %edi, %eax +; X64-NEXT: leal (%rdi,%rdi,4), %eax +; X64-NEXT: leal (%rax,%rax,4), %eax +; X64-NEXT: addl %edi, %eax ; X64-NEXT: # kill: def $ax killed $ax killed $eax ; X64-NEXT: retq %mul = mul nsw i16 %x, 26 diff --git a/llvm/test/CodeGen/X86/mul-constant-i32.ll b/llvm/test/CodeGen/X86/mul-constant-i32.ll index 6b2a054..83b4d9f 100644 --- a/llvm/test/CodeGen/X86/mul-constant-i32.ll +++ b/llvm/test/CodeGen/X86/mul-constant-i32.ll @@ -1304,25 +1304,25 @@ define i32 @test_mul_by_26(i32 %x) { ; X86-LABEL: test_mul_by_26: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %ecx -; X86-NEXT: leal (%ecx,%ecx,8), %eax -; X86-NEXT: leal (%eax,%eax,2), %eax -; X86-NEXT: subl %ecx, %eax +; X86-NEXT: leal (%ecx,%ecx,4), %eax +; X86-NEXT: leal (%eax,%eax,4), %eax +; X86-NEXT: addl %ecx, %eax ; X86-NEXT: retl ; ; X64-HSW-LABEL: test_mul_by_26: ; X64-HSW: # %bb.0: ; X64-HSW-NEXT: # kill: def $edi killed $edi def $rdi -; X64-HSW-NEXT: leal (%rdi,%rdi,8), %eax # sched: [1:0.50] -; X64-HSW-NEXT: leal (%rax,%rax,2), %eax # sched: [1:0.50] -; X64-HSW-NEXT: subl %edi, %eax # sched: [1:0.25] +; X64-HSW-NEXT: leal (%rdi,%rdi,4), %eax # sched: [1:0.50] +; X64-HSW-NEXT: leal (%rax,%rax,4), %eax # sched: [1:0.50] +; X64-HSW-NEXT: addl %edi, %eax # sched: [1:0.25] ; X64-HSW-NEXT: retq # sched: [7:1.00] ; ; X64-JAG-LABEL: test_mul_by_26: ; X64-JAG: # %bb.0: ; X64-JAG-NEXT: # kill: def $edi killed $edi def $rdi -; X64-JAG-NEXT: leal (%rdi,%rdi,8), %eax # sched: [2:1.00] -; X64-JAG-NEXT: leal (%rax,%rax,2), %eax # sched: [2:1.00] -; X64-JAG-NEXT: subl %edi, %eax # sched: [1:0.50] +; X64-JAG-NEXT: leal (%rdi,%rdi,4), %eax # sched: [2:1.00] +; X64-JAG-NEXT: leal (%rax,%rax,4), %eax # sched: [2:1.00] +; X64-JAG-NEXT: addl %edi, %eax # sched: [1:0.50] ; X64-JAG-NEXT: retq # sched: [4:1.00] ; ; X86-NOOPT-LABEL: test_mul_by_26: diff --git a/llvm/test/CodeGen/X86/mul-constant-i64.ll b/llvm/test/CodeGen/X86/mul-constant-i64.ll index e241f84..32fe937 100644 --- a/llvm/test/CodeGen/X86/mul-constant-i64.ll +++ b/llvm/test/CodeGen/X86/mul-constant-i64.ll @@ -1358,9 +1358,9 @@ define i64 @test_mul_by_26(i64 %x) { ; X86-LABEL: test_mul_by_26: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax -; X86-NEXT: leal (%eax,%eax,8), %ecx -; X86-NEXT: leal (%ecx,%ecx,2), %ecx -; X86-NEXT: subl %eax, %ecx +; X86-NEXT: leal (%eax,%eax,4), %ecx +; X86-NEXT: leal (%ecx,%ecx,4), %ecx +; X86-NEXT: addl %eax, %ecx ; X86-NEXT: movl $26, %eax ; X86-NEXT: mull {{[0-9]+}}(%esp) ; X86-NEXT: addl %ecx, %edx @@ -1368,16 +1368,16 @@ define i64 @test_mul_by_26(i64 %x) { ; ; X64-HSW-LABEL: test_mul_by_26: ; X64-HSW: # %bb.0: -; X64-HSW-NEXT: leaq (%rdi,%rdi,8), %rax # sched: [1:0.50] -; X64-HSW-NEXT: leaq (%rax,%rax,2), %rax # sched: [1:0.50] -; X64-HSW-NEXT: subq %rdi, %rax # sched: [1:0.25] +; X64-HSW-NEXT: leaq (%rdi,%rdi,4), %rax # sched: [1:0.50] +; X64-HSW-NEXT: leaq (%rax,%rax,4), %rax # sched: [1:0.50] +; X64-HSW-NEXT: addq %rdi, %rax # sched: [1:0.25] ; X64-HSW-NEXT: retq # sched: [7:1.00] ; ; X64-JAG-LABEL: test_mul_by_26: ; X64-JAG: # %bb.0: -; X64-JAG-NEXT: leaq (%rdi,%rdi,8), %rax # sched: [2:1.00] -; X64-JAG-NEXT: leaq (%rax,%rax,2), %rax # sched: [2:1.00] -; X64-JAG-NEXT: subq %rdi, %rax # sched: [1:0.50] +; X64-JAG-NEXT: leaq (%rdi,%rdi,4), %rax # sched: [2:1.00] +; X64-JAG-NEXT: leaq (%rax,%rax,4), %rax # sched: [2:1.00] +; X64-JAG-NEXT: addq %rdi, %rax # sched: [1:0.50] ; X64-JAG-NEXT: retq # sched: [4:1.00] ; ; X86-NOOPT-LABEL: test_mul_by_26: diff --git a/llvm/test/CodeGen/X86/mul-constant-result.ll b/llvm/test/CodeGen/X86/mul-constant-result.ll index 2dd973a..adac93d3 100644 --- a/llvm/test/CodeGen/X86/mul-constant-result.ll +++ b/llvm/test/CodeGen/X86/mul-constant-result.ll @@ -193,10 +193,14 @@ define i32 @mult(i32, i32) local_unnamed_addr #0 { ; X86-NEXT: retl ; X86-NEXT: .LBB0_32: ; X86-NEXT: .cfi_def_cfa_offset 8 -; X86-NEXT: leal (%eax,%eax,8), %ecx -; X86-NEXT: leal (%ecx,%ecx,2), %ecx -; X86-NEXT: jmp .LBB0_12 +; X86-NEXT: leal (%eax,%eax,4), %ecx +; X86-NEXT: leal (%ecx,%ecx,4), %ecx +; X86-NEXT: addl %ecx, %eax +; X86-NEXT: popl %esi +; X86-NEXT: .cfi_def_cfa_offset 4 +; X86-NEXT: retl ; X86-NEXT: .LBB0_33: +; X86-NEXT: .cfi_def_cfa_offset 8 ; X86-NEXT: leal (%eax,%eax,8), %eax ; X86-NEXT: leal (%eax,%eax,2), %eax ; X86-NEXT: popl %esi @@ -369,9 +373,9 @@ define i32 @mult(i32, i32) local_unnamed_addr #0 { ; X64-HSW-NEXT: # kill: def $eax killed $eax killed $rax ; X64-HSW-NEXT: retq ; X64-HSW-NEXT: .LBB0_29: -; X64-HSW-NEXT: leal (%rax,%rax,8), %ecx -; X64-HSW-NEXT: leal (%rcx,%rcx,2), %ecx -; X64-HSW-NEXT: jmp .LBB0_8 +; X64-HSW-NEXT: leal (%rax,%rax,4), %ecx +; X64-HSW-NEXT: leal (%rcx,%rcx,4), %ecx +; X64-HSW-NEXT: jmp .LBB0_20 ; X64-HSW-NEXT: .LBB0_30: ; X64-HSW-NEXT: leal (%rax,%rax,8), %eax ; X64-HSW-NEXT: leal (%rax,%rax,2), %eax -- 2.7.4