diff options
author | YunQiang Su <syq@debian.org> | 2025-03-28 16:49:45 +0800 |
---|---|---|
committer | GitHub <noreply@github.com> | 2025-03-28 16:49:45 +0800 |
commit | 71f43a7c42a37d18be98b0885d62ef76e658f242 (patch) | |
tree | 72a1ca7ef7390da3c0e059a161e394a760f8f220 | |
parent | 1318a7bb098ff055533d7b3fe2f5d9201f3d79a6 (diff) | |
download | llvm-71f43a7c42a37d18be98b0885d62ef76e658f242.zip llvm-71f43a7c42a37d18be98b0885d62ef76e658f242.tar.gz llvm-71f43a7c42a37d18be98b0885d62ef76e658f242.tar.bz2 |
MIPS: Set EnableLoopTermFold (#133378)
Setting `EnableLoopTermFold` enables `loop-term-fold` pass.
-rw-r--r-- | llvm/lib/Target/Mips/MipsTargetMachine.cpp | 1 | ||||
-rw-r--r-- | llvm/test/CodeGen/Mips/brdelayslot.ll | 9 | ||||
-rw-r--r-- | llvm/test/Transforms/LoopStrengthReduce/Mips/long-array-initialize.ll | 85 |
3 files changed, 63 insertions, 32 deletions
diff --git a/llvm/lib/Target/Mips/MipsTargetMachine.cpp b/llvm/lib/Target/Mips/MipsTargetMachine.cpp index 30eb739..4d1bcb1 100644 --- a/llvm/lib/Target/Mips/MipsTargetMachine.cpp +++ b/llvm/lib/Target/Mips/MipsTargetMachine.cpp @@ -233,6 +233,7 @@ public: // can break this requirement, so disable it when long branch pass is // enabled. EnableTailMerge = !getMipsSubtarget().enableLongBranchPass(); + EnableLoopTermFold = true; } MipsTargetMachine &getMipsTargetMachine() const { diff --git a/llvm/test/CodeGen/Mips/brdelayslot.ll b/llvm/test/CodeGen/Mips/brdelayslot.ll index 7920537..da14564 100644 --- a/llvm/test/CodeGen/Mips/brdelayslot.ll +++ b/llvm/test/CodeGen/Mips/brdelayslot.ll @@ -134,9 +134,12 @@ declare void @foo11() ; ; SUCCBB-LABEL: succbbs_loop1: ; SUCCBB: blez $5, $BB -; SUCCBB-NEXT: addiu -; SUCCBB: bnez ${{[0-9]+}}, $BB -; SUCCBB-NEXT: addiu +; SUCCBB-NEXT: sll +; SUCCBB-NEXT: # %bb +; SUCCBB-NEXT: addu +; SUCCBB: addiu ${{[0-9]+}}, ${{[0-9]+}}, 4 +; SUCCBB-NEXT: bne ${{[0-9]+}}, ${{[0-9]+}}, $BB +; SUCCBB-NEXT: nop define i32 @succbbs_loop1(ptr nocapture %a, i32 %n) { entry: diff --git a/llvm/test/Transforms/LoopStrengthReduce/Mips/long-array-initialize.ll b/llvm/test/Transforms/LoopStrengthReduce/Mips/long-array-initialize.ll index 9f562451..2a40797 100644 --- a/llvm/test/Transforms/LoopStrengthReduce/Mips/long-array-initialize.ll +++ b/llvm/test/Transforms/LoopStrengthReduce/Mips/long-array-initialize.ll @@ -1,5 +1,6 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 -; RUN: opt < %s -passes=loop-reduce -S | FileCheck %s +; RUN: opt < %s -passes=loop-reduce,loop-term-fold -S | FileCheck %s --check-prefix=CHECK-OPT +; RUN: llc < %s -o - | FileCheck %s --check-prefix=CHECK-LLC target datalayout = "E-m:m-p:32:32-i8:8:32-i16:16:32-i64:64-n32-S64" target triple = "mips-unknown-linux-gnu" @@ -8,20 +9,34 @@ target triple = "mips-unknown-linux-gnu" ; Function Attrs: nofree norecurse nosync nounwind memory(write, argmem: none, inaccessiblemem: none) define dso_local void @in128000(i32 noundef signext %k, i32 noundef signext %n) local_unnamed_addr #0 { -; CHECK-LABEL: define dso_local void @in128000( -; CHECK-SAME: i32 noundef signext [[K:%.*]], i32 noundef signext [[N:%.*]]) local_unnamed_addr { -; CHECK-NEXT: [[ENTRY:.*]]: -; CHECK-NEXT: br label %[[FOR_BODY:.*]] -; CHECK: [[FOR_COND_CLEANUP:.*]]: -; CHECK-NEXT: ret void -; CHECK: [[FOR_BODY]]: -; CHECK-NEXT: [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_BODY]] ], [ @x, %[[ENTRY]] ] -; CHECK-NEXT: [[LSR_IV:%.*]] = phi i32 [ [[LSR_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ 128000, %[[ENTRY]] ] -; CHECK-NEXT: store i32 [[K]], ptr [[LSR_IV1]], align 4 -; CHECK-NEXT: [[LSR_IV_NEXT]] = add nsw i32 [[LSR_IV]], -1 -; CHECK-NEXT: [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i32 4 -; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[LSR_IV_NEXT]], 0 -; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]] +; CHECK-LLC-LABEL: in128000: +; CHECK-LLC: # %bb.0: # %entry +; CHECK-LLC-NEXT: lui $1, %hi(x) +; CHECK-LLC-NEXT: addiu $2, $1, %lo(x) +; CHECK-LLC-NEXT: lui $1, 7 +; CHECK-LLC-NEXT: ori $1, $1, 53248 +; CHECK-LLC-NEXT: addu $3, $2, $1 +; CHECK-LLC-NEXT: $BB0_1: # %for.body +; CHECK-LLC-NEXT: # =>This Inner Loop Header: Depth=1 +; CHECK-LLC-NEXT: sw $4, 0($2) +; CHECK-LLC-NEXT: addiu $2, $2, 4 +; CHECK-LLC-NEXT: bne $2, $3, $BB0_1 +; CHECK-LLC-NEXT: nop +; CHECK-LLC-NEXT: # %bb.2: # %for.cond.cleanup +; CHECK-LLC-NEXT: jr $ra +; CHECK-LLC-NEXT: nop +; CHECK-OPT-LABEL: define dso_local void @in128000( +; CHECK-OPT-SAME: i32 noundef signext [[K:%.*]], i32 noundef signext [[N:%.*]]) local_unnamed_addr { +; CHECK-OPT-NEXT: [[ENTRY:.*]]: +; CHECK-OPT-NEXT: br label %[[FOR_BODY:.*]] +; CHECK-OPT: [[FOR_COND_CLEANUP:.*]]: +; CHECK-OPT-NEXT: ret void +; CHECK-OPT: [[FOR_BODY]]: +; CHECK-OPT-NEXT: [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_BODY]] ], [ @x, %[[ENTRY]] ] +; CHECK-OPT-NEXT: store i32 [[K]], ptr [[LSR_IV1]], align 4 +; CHECK-OPT-NEXT: [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i32 4 +; CHECK-OPT-NEXT: [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND:%.*]] = icmp eq ptr [[SCEVGEP]], getelementptr inbounds nuw (i8, ptr @x, i32 512000) +; CHECK-OPT-NEXT: br i1 [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]] ; entry: br label %for.body @@ -40,20 +55,32 @@ for.body: ; preds = %entry, %for.body ; Function Attrs: nofree norecurse nosync nounwind memory(write, argmem: none, inaccessiblemem: none) define dso_local void @in1000(i32 noundef signext %k, i32 noundef signext %n) local_unnamed_addr #0 { -; CHECK-LABEL: define dso_local void @in1000( -; CHECK-SAME: i32 noundef signext [[K:%.*]], i32 noundef signext [[N:%.*]]) local_unnamed_addr { -; CHECK-NEXT: [[ENTRY:.*]]: -; CHECK-NEXT: br label %[[FOR_BODY:.*]] -; CHECK: [[FOR_COND_CLEANUP:.*]]: -; CHECK-NEXT: ret void -; CHECK: [[FOR_BODY]]: -; CHECK-NEXT: [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_BODY]] ], [ @x, %[[ENTRY]] ] -; CHECK-NEXT: [[LSR_IV:%.*]] = phi i32 [ [[LSR_IV_NEXT:%.*]], %[[FOR_BODY]] ], [ 1000, %[[ENTRY]] ] -; CHECK-NEXT: store i32 [[K]], ptr [[LSR_IV1]], align 4 -; CHECK-NEXT: [[LSR_IV_NEXT]] = add nsw i32 [[LSR_IV]], -1 -; CHECK-NEXT: [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i32 4 -; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[LSR_IV_NEXT]], 0 -; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]] +; CHECK-LLC-LABEL: in1000: +; CHECK-LLC: # %bb.0: # %entry +; CHECK-LLC-NEXT: lui $1, %hi(x) +; CHECK-LLC-NEXT: addiu $2, $1, %lo(x) +; CHECK-LLC-NEXT: addiu $3, $2, 4000 +; CHECK-LLC-NEXT: $BB1_1: # %for.body +; CHECK-LLC-NEXT: # =>This Inner Loop Header: Depth=1 +; CHECK-LLC-NEXT: sw $4, 0($2) +; CHECK-LLC-NEXT: addiu $2, $2, 4 +; CHECK-LLC-NEXT: bne $2, $3, $BB1_1 +; CHECK-LLC-NEXT: nop +; CHECK-LLC-NEXT: # %bb.2: # %for.cond.cleanup +; CHECK-LLC-NEXT: jr $ra +; CHECK-LLC-NEXT: nop +; CHECK-OPT-LABEL: define dso_local void @in1000( +; CHECK-OPT-SAME: i32 noundef signext [[K:%.*]], i32 noundef signext [[N:%.*]]) local_unnamed_addr { +; CHECK-OPT-NEXT: [[ENTRY:.*]]: +; CHECK-OPT-NEXT: br label %[[FOR_BODY:.*]] +; CHECK-OPT: [[FOR_COND_CLEANUP:.*]]: +; CHECK-OPT-NEXT: ret void +; CHECK-OPT: [[FOR_BODY]]: +; CHECK-OPT-NEXT: [[LSR_IV1:%.*]] = phi ptr [ [[SCEVGEP:%.*]], %[[FOR_BODY]] ], [ @x, %[[ENTRY]] ] +; CHECK-OPT-NEXT: store i32 [[K]], ptr [[LSR_IV1]], align 4 +; CHECK-OPT-NEXT: [[SCEVGEP]] = getelementptr i8, ptr [[LSR_IV1]], i32 4 +; CHECK-OPT-NEXT: [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND:%.*]] = icmp eq ptr [[SCEVGEP]], getelementptr inbounds nuw (i8, ptr @x, i32 4000) +; CHECK-OPT-NEXT: br i1 [[LSR_FOLD_TERM_COND_REPLACED_TERM_COND]], label %[[FOR_COND_CLEANUP]], label %[[FOR_BODY]] ; entry: br label %for.body |