diff options
author | Sanjay Patel <spatel@rotateright.com> | 2016-10-19 16:58:59 +0000 |
---|---|---|
committer | Sanjay Patel <spatel@rotateright.com> | 2016-10-19 16:58:59 +0000 |
commit | 3a3aaf67e07ade866b9500a9466086a68d0dd70c (patch) | |
tree | 15006d3cec17160af880cfd3669c039ad95dd5ea /llvm/test/CodeGen/Mips | |
parent | 383803230bba0c0dea3d93d9eb4f97d6fa8f3718 (diff) | |
download | bcm5719-llvm-3a3aaf67e07ade866b9500a9466086a68d0dd70c.tar.gz bcm5719-llvm-3a3aaf67e07ade866b9500a9466086a68d0dd70c.zip |
[DAG] optimize negation of bool
Use mask and negate for legalization of i1 source type with SIGN_EXTEND_INREG.
With the mask, this should be no worse than 2 shifts. The mask can be eliminated
in some cases, so that should be better than 2 shifts.
This change exposed some missing folds related to negation:
https://reviews.llvm.org/rL284239
https://reviews.llvm.org/rL284395
There may be others, so please let me know if you see any regressions.
Differential Revision: https://reviews.llvm.org/D25485
llvm-svn: 284611
Diffstat (limited to 'llvm/test/CodeGen/Mips')
-rw-r--r-- | llvm/test/CodeGen/Mips/llvm-ir/add.ll | 46 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/llvm-ir/mul.ll | 51 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/llvm-ir/sdiv.ll | 22 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/llvm-ir/srem.ll | 18 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/llvm-ir/sub.ll | 9 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/llvm-ir/urem.ll | 16 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/select.ll | 3 |
7 files changed, 99 insertions, 66 deletions
diff --git a/llvm/test/CodeGen/Mips/llvm-ir/add.ll b/llvm/test/CodeGen/Mips/llvm-ir/add.ll index 3f69a966dee..eece0309104 100644 --- a/llvm/test/CodeGen/Mips/llvm-ir/add.ll +++ b/llvm/test/CodeGen/Mips/llvm-ir/add.ll @@ -31,21 +31,27 @@ ; RUN: llc < %s -march=mips -mcpu=mips64r6 -target-abi n64 -mattr=+micromips -O2 | FileCheck %s \ ; RUN: -check-prefixes=ALL,MMR6,MM64 + +; FIXME: This code sequence is inefficient as it should be 'subu $[[T0]], $zero, $[[T0]'. +; This sequence is even better as it's a single instruction. See D25485 for the rest of +; the cases where this sequence occurs. + define signext i1 @add_i1(i1 signext %a, i1 signext %b) { entry: ; ALL-LABEL: add_i1: - ; NOT-R2-R6: addu $[[T0:[0-9]+]], $4, $5 - ; NOT-R2-R6: sll $[[T0]], $[[T0]], 31 - ; NOT-R2-R6: sra $2, $[[T0]], 31 + ; NOT-R2-R6: addu $[[T0:[0-9]+]], $4, $5 + ; NOT-R2-R6: andi $[[T0]], $[[T0]], 1 + ; NOT-R2-R6: negu $2, $[[T0]] - ; R2-R6: addu $[[T0:[0-9]+]], $4, $5 - ; R2-R6: sll $[[T0]], $[[T0]], 31 - ; R2-R6: sra $2, $[[T0]], 31 + ; R2-R6: addu $[[T0:[0-9]+]], $4, $5 + ; R2-R6: andi $[[T0]], $[[T0]], 1 + ; R2-R6: negu $2, $[[T0]] ; MMR6: addu16 $[[T0:[0-9]+]], $4, $5 - ; MMR6: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; MMR6: sra $2, $[[T1]], 31 + ; MMR6: andi16 $[[T0]], $[[T0]], 1 + ; MMR6: li16 $[[T1:[0-9]+]], 0 + ; MMR6: subu16 $[[T0]], $[[T1]], $[[T0]] %r = add i1 %a, %b ret i1 %r @@ -303,18 +309,18 @@ define signext i128 @add_i128_4(i128 signext %a) { define signext i1 @add_i1_3(i1 signext %a) { ; ALL-LABEL: add_i1_3: - - ; ALL: sll $[[T0:[0-9]+]], $4, 31 - ; ALL: lui $[[T1:[0-9]+]], 32768 - - ; GP32: addu $[[T0]], $[[T0]], $[[T1]] - ; GP32: sra $[[T1]], $[[T0]], 31 - - ; GP64: addu $[[T0]], $[[T0]], $[[T1]] - ; GP64: sra $[[T1]], $[[T0]], 31 - - ; MMR6: addu16 $[[T0]], $[[T0]], $[[T1]] - ; MMR6: sra $[[T0]], $[[T0]], 31 + ; GP32: addiu $[[T0:[0-9]+]], $4, 1 + ; GP32: andi $[[T0]], $[[T0]], 1 + ; GP32: negu $2, $[[T0]] + + ; GP64: addiu $[[T0:[0-9]+]], $4, 1 + ; GP64: andi $[[T0]], $[[T0]], 1 + ; GP64: negu $2, $[[T0]] + + ; MMR6: addiur2 $[[T0:[0-9]+]], $4, 1 + ; MMR6: andi16 $[[T0]], $[[T0]], 1 + ; MMR6: li16 $[[T1:[0-9]+]], 0 + ; MMR6: subu16 $2, $[[T1]], $[[T0]] %r = add i1 3, %a ret i1 %r diff --git a/llvm/test/CodeGen/Mips/llvm-ir/mul.ll b/llvm/test/CodeGen/Mips/llvm-ir/mul.ll index 3e1293b3331..1562372ce9a 100644 --- a/llvm/test/CodeGen/Mips/llvm-ir/mul.ll +++ b/llvm/test/CodeGen/Mips/llvm-ir/mul.ll @@ -27,7 +27,7 @@ ; RUN: llc < %s -march=mips -mcpu=mips32r6 -mattr=+micromips -relocation-model=pic | \ ; RUN: FileCheck %s -check-prefixes=MM32,MM32R6 ; RUN: llc < %s -march=mips -mcpu=mips64r6 -mattr=+micromips -target-abi n64 -relocation-model=pic | \ -; RUN: FileCheck %s -check-prefix=64R6 +; RUN: FileCheck %s -check-prefix=MM64R6 define signext i1 @mul_i1(i1 signext %a, i1 signext %b) { entry: @@ -35,33 +35,39 @@ entry: ; M2: mult $4, $5 ; M2: mflo $[[T0:[0-9]+]] - ; M2: sll $[[T0]], $[[T0]], 31 - ; M2: sra $2, $[[T0]], 31 + ; M2: andi $[[T0]], $[[T0]], 1 + ; M2: negu $2, $[[T0]] ; 32R1-R5: mul $[[T0:[0-9]+]], $4, $5 - ; 32R1-R5: sll $[[T0]], $[[T0]], 31 - ; 32R1-R5: sra $2, $[[T0]], 31 + ; 32R1-R5: andi $[[T0]], $[[T0]], 1 + ; 32R1-R5: negu $2, $[[T0]] ; 32R6: mul $[[T0:[0-9]+]], $4, $5 - ; 32R6: sll $[[T0]], $[[T0]], 31 - ; 32R6: sra $2, $[[T0]], 31 + ; 32R6: andi $[[T0]], $[[T0]], 1 + ; 32R6: negu $2, $[[T0]] ; M4: mult $4, $5 ; M4: mflo $[[T0:[0-9]+]] - ; M4: sll $[[T0]], $[[T0]], 31 - ; M4: sra $2, $[[T0]], 31 + ; M4: andi $[[T0]], $[[T0]], 1 + ; M4: negu $2, $[[T0]] ; 64R1-R5: mul $[[T0:[0-9]+]], $4, $5 - ; 64R1-R5: sll $[[T0]], $[[T0]], 31 - ; 64R1-R5: sra $2, $[[T0]], 31 + ; 64R1-R5: andi $[[T0]], $[[T0]], 1 + ; 64R1-R5: negu $2, $[[T0]] ; 64R6: mul $[[T0:[0-9]+]], $4, $5 - ; 64R6: sll $[[T0]], $[[T0]], 31 - ; 64R6: sra $2, $[[T0]], 31 + ; 64R6: andi $[[T0]], $[[T0]], 1 + ; 64R6: negu $2, $[[T0]] + + ; MM64R6: mul $[[T0:[0-9]+]], $4, $5 + ; MM64R6: andi16 $[[T0]], $[[T0]], 1 + ; MM64R6: li16 $[[T1:[0-9]+]], 0 + ; MM64R6: subu16 $2, $[[T1]], $[[T0]] ; MM32: mul $[[T0:[0-9]+]], $4, $5 - ; MM32: sll $[[T0]], $[[T0]], 31 - ; MM32: sra $2, $[[T0]], 31 + ; MM32: andi16 $[[T0]], $[[T0]], 1 + ; MM32: li16 $[[T1:[0-9]+]], 0 + ; MM32: subu16 $2, $[[T1]], $[[T0]] %r = mul i1 %a, %b ret i1 %r @@ -101,6 +107,9 @@ entry: ; 64R6: mul $[[T0:[0-9]+]], $4, $5 ; 64R6: seb $2, $[[T0]] + ; MM64R6: mul $[[T0:[0-9]+]], $4, $5 + ; MM64R6: seb $2, $[[T0]] + ; MM32: mul $[[T0:[0-9]+]], $4, $5 ; MM32: seb $2, $[[T0]] @@ -142,6 +151,9 @@ entry: ; 64R6: mul $[[T0:[0-9]+]], $4, $5 ; 64R6: seh $2, $[[T0]] + ; MM64R6: mul $[[T0:[0-9]+]], $4, $5 + ; MM64R6: seh $2, $[[T0]] + ; MM32: mul $[[T0:[0-9]+]], $4, $5 ; MM32: seh $2, $[[T0]] @@ -161,6 +173,7 @@ entry: ; 64R1-R5: mul $2, $4, $5 ; 64R6: mul $2, $4, $5 + ; MM64R6: mul $2, $4, $5 ; MM32: mul $2, $4, $5 @@ -204,6 +217,7 @@ entry: ; 64R1-R5: mflo $2 ; 64R6: dmul $2, $4, $5 + ; MM64R6: dmul $2, $4, $5 ; MM32R3: multu $[[T0:[0-9]+]], $7 ; MM32R3: mflo $[[T1:[0-9]+]] @@ -247,6 +261,13 @@ entry: ; 64R6: daddu $2, $[[T1]], $[[T0]] ; 64R6-DAG: dmul $3, $5, $7 + ; MM64R6-DAG: dmul $[[T1:[0-9]+]], $5, $6 + ; MM64R6: dmuhu $[[T2:[0-9]+]], $5, $7 + ; MM64R6: daddu $[[T3:[0-9]+]], $[[T2]], $[[T1]] + ; MM64R6-DAG: dmul $[[T0:[0-9]+]], $4, $7 + ; MM64R6: daddu $2, $[[T1]], $[[T0]] + ; MM64R6-DAG: dmul $3, $5, $7 + ; MM32: lw $25, %call16(__multi3)($16) %r = mul i128 %a, %b diff --git a/llvm/test/CodeGen/Mips/llvm-ir/sdiv.ll b/llvm/test/CodeGen/Mips/llvm-ir/sdiv.ll index 0bbdae8f5e2..defd25bb41a 100644 --- a/llvm/test/CodeGen/Mips/llvm-ir/sdiv.ll +++ b/llvm/test/CodeGen/Mips/llvm-ir/sdiv.ll @@ -40,26 +40,28 @@ entry: ; NOT-R6: div $zero, $4, $5 ; NOT-R6: teq $5, $zero, 7 ; NOT-R6: mflo $[[T0:[0-9]+]] - ; FIXME: The sll/sra instructions are redundant since div is signed. - ; NOT-R6: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; NOT-R6: sra $2, $[[T1]], 31 + ; FIXME: The andi/negu instructions are redundant since div is signed. + ; NOT-R6: andi $[[T0]], $[[T0]], 1 + ; NOT-R6: negu $2, $[[T0]] ; R6: div $[[T0:[0-9]+]], $4, $5 ; R6: teq $5, $zero, 7 - ; FIXME: The sll/sra instructions are redundant since div is signed. - ; R6: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; R6: sra $2, $[[T1]], 31 + ; FIXME: The andi/negu instructions are redundant since div is signed. + ; R6: andi $[[T0]], $[[T0]], 1 + ; R6: negu $2, $[[T0]] ; MMR3: div $zero, $4, $5 ; MMR3: teq $5, $zero, 7 ; MMR3: mflo $[[T0:[0-9]+]] - ; MMR3: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; MMR3: sra $2, $[[T1]], 31 + ; MMR3: andi16 $[[T0]], $[[T0]], 1 + ; MMR3: li16 $[[T1:[0-9]+]], 0 + ; MMR3: subu16 $2, $[[T1]], $[[T0]] ; MMR6: div $[[T0:[0-9]+]], $4, $5 ; MMR6: teq $5, $zero, 7 - ; MMR6: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; MMR6: sra $2, $[[T1]], 31 + ; MMR6: andi16 $[[T0]], $[[T0]], 1 + ; MMR6: li16 $[[T1:[0-9]+]], 0 + ; MMR6: subu16 $2, $[[T1]], $[[T0]] %r = sdiv i1 %a, %b ret i1 %r diff --git a/llvm/test/CodeGen/Mips/llvm-ir/srem.ll b/llvm/test/CodeGen/Mips/llvm-ir/srem.ll index f67477e6093..42664d7457e 100644 --- a/llvm/test/CodeGen/Mips/llvm-ir/srem.ll +++ b/llvm/test/CodeGen/Mips/llvm-ir/srem.ll @@ -40,24 +40,26 @@ entry: ; NOT-R6: div $zero, $4, $5 ; NOT-R6: teq $5, $zero, 7 ; NOT-R6: mfhi $[[T0:[0-9]+]] - ; NOT-R6: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; NOT-R6: sra $2, $[[T1]], 31 + ; NOT-R6: andi $[[T0]], $[[T0]], 1 + ; NOT-R6: negu $2, $[[T0]] ; R6: mod $[[T0:[0-9]+]], $4, $5 ; R6: teq $5, $zero, 7 - ; R6: sll $[[T3:[0-9]+]], $[[T0]], 31 - ; R6: sra $2, $[[T3]], 31 + ; R6: andi $[[T0]], $[[T0]], 1 + ; R6: negu $2, $[[T0]] ; MMR3: div $zero, $4, $5 ; MMR3: teq $5, $zero, 7 ; MMR3: mfhi $[[T0:[0-9]+]] - ; MMR3: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; MMR3: sra $2, $[[T1]], 31 + ; MMR3: andi16 $[[T0]], $[[T0]], 1 + ; MMR3: li16 $[[T1:[0-9]+]], 0 + ; MMR3: subu16 $2, $[[T1]], $[[T0]] ; MMR6: mod $[[T0:[0-9]+]], $4, $5 ; MMR6: teq $5, $zero, 7 - ; MMR6: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; MMR6: sra $2, $[[T1]], 31 + ; MMR6: andi16 $[[T0]], $[[T0]], 1 + ; MMR6: li16 $[[T1:[0-9]+]], 0 + ; MMR6: subu16 $2, $[[T1]], $[[T0]] %r = srem i1 %a, %b ret i1 %r diff --git a/llvm/test/CodeGen/Mips/llvm-ir/sub.ll b/llvm/test/CodeGen/Mips/llvm-ir/sub.ll index 55b486f83ab..617ab3c1a21 100644 --- a/llvm/test/CodeGen/Mips/llvm-ir/sub.ll +++ b/llvm/test/CodeGen/Mips/llvm-ir/sub.ll @@ -36,12 +36,13 @@ entry: ; ALL-LABEL: sub_i1: ; NOT-MM: subu $[[T0:[0-9]+]], $4, $5 - ; NOT-MM: sll $[[T0]], $[[T0]], 31 - ; NOT-MM: sra $2, $[[T0]], 31 + ; NOT-MM: andi $[[T0]], $[[T0]], 1 + ; NOT-MM: negu $2, $[[T0]] ; MM: subu16 $[[T0:[0-9]+]], $4, $5 - ; MM: sll $[[T1:[0-9]+]], $[[T0]], 31 - ; MM: sra $[[T0]], $[[T1]], 31 + ; MM: andi16 $[[T0]], $[[T0]], 1 + ; MM: li16 $[[T1:[0-9]+]], 0 + ; MM: subu16 $2, $[[T1]], $[[T0]] %r = sub i1 %a, %b ret i1 %r diff --git a/llvm/test/CodeGen/Mips/llvm-ir/urem.ll b/llvm/test/CodeGen/Mips/llvm-ir/urem.ll index bad7c507b05..160c126c7e3 100644 --- a/llvm/test/CodeGen/Mips/llvm-ir/urem.ll +++ b/llvm/test/CodeGen/Mips/llvm-ir/urem.ll @@ -42,30 +42,30 @@ entry: ; NOT-R6: divu $zero, $[[T1]], $[[T0]] ; NOT-R6: teq $[[T0]], $zero, 7 ; NOT-R6: mfhi $[[T2:[0-9]+]] - ; NOT-R6: sll $[[T3:[0-9]+]], $[[T2]], 31 - ; NOT-R6: sra $2, $[[T3]], 31 + ; NOT-R6: andi $[[T0]], $[[T0]], 1 + ; NOT-R6: negu $2, $[[T0]] ; R6: andi $[[T0:[0-9]+]], $5, 1 ; R6: andi $[[T1:[0-9]+]], $4, 1 ; R6: modu $[[T2:[0-9]+]], $[[T1]], $[[T0]] ; R6: teq $[[T0]], $zero, 7 - ; R6: sll $[[T3:[0-9]+]], $[[T2]], 31 - ; R6: sra $2, $[[T3]], 31 + ; R6: negu $2, $[[T2]] ; MMR3: andi16 $[[T0:[0-9]+]], $5, 1 ; MMR3: andi16 $[[T1:[0-9]+]], $4, 1 ; MMR3: divu $zero, $[[T1]], $[[T0]] ; MMR3: teq $[[T0]], $zero, 7 ; MMR3: mfhi $[[T2:[0-9]+]] - ; MMR3: sll $[[T3:[0-9]+]], $[[T2]], 31 - ; MMR3: sra $2, $[[T3]], 31 + ; MMR3: andi16 $[[T0]], $[[T0]], 1 + ; MMR3: li16 $[[T1:[0-9]+]], 0 + ; MMR3: subu16 $2, $[[T1]], $[[T0]] ; MMR6: andi16 $[[T0:[0-9]+]], $5, 1 ; MMR6: andi16 $[[T1:[0-9]+]], $4, 1 ; MMR6: modu $[[T2:[0-9]+]], $[[T1]], $[[T0]] ; MMR6: teq $[[T0]], $zero, 7 - ; MMR6: sll $[[T3:[0-9]+]], $[[T2]], 31 - ; MMR6: sra $2, $[[T3]], 31 + ; MMR6: li16 $[[T3:[0-9]+]], 0 + ; MMR6: subu16 $2, $[[T3]], $[[T2]] %r = urem i1 %a, %b ret i1 %r diff --git a/llvm/test/CodeGen/Mips/select.ll b/llvm/test/CodeGen/Mips/select.ll index 0ef8f36333f..8c1b0286bb5 100644 --- a/llvm/test/CodeGen/Mips/select.ll +++ b/llvm/test/CodeGen/Mips/select.ll @@ -140,9 +140,10 @@ entry: ; 32R2-DAG: mtc1 $6, $[[F1:f0]] ; 32R2: movn.s $[[F1]], $[[F0]], $4 +; 32R6: sltu $[[T0:[0-9]+]], $zero, $4 +; 32R6: negu $[[T0]], $[[T0]] ; 32R6-DAG: mtc1 $5, $[[F0:f[0-9]+]] ; 32R6-DAG: mtc1 $6, $[[F1:f[0-9]+]] -; 32R6: sltu $[[T0:[0-9]+]], $zero, $4 ; 32R6: mtc1 $[[T0]], $[[CC:f0]] ; 32R6: sel.s $[[CC]], $[[F1]], $[[F0]] |