diff --git a/llvm/lib/Analysis/Loads.cpp b/llvm/lib/Analysis/Loads.cpp index 1d54a66705a2a..c19a276464e1e 100644 --- a/llvm/lib/Analysis/Loads.cpp +++ b/llvm/lib/Analysis/Loads.cpp @@ -313,6 +313,13 @@ bool llvm::isDereferenceableAndAlignedInLoop(LoadInst *LI, Loop *L, const auto *Offset = dyn_cast(StartS->getOperand(0)); const auto *NewBase = dyn_cast(StartS->getOperand(1)); if (StartS->getNumOperands() == 2 && Offset && NewBase) { + // The following code below assumes the offset is unsigned, but GEP + // offsets are treated as signed so we can end up with a signed value + // here too. For example, suppose the initial PHI value is (i8 255), + // the offset will be treated as (i8 -1) and sign-extended to (i64 -1). + if (Offset->getAPInt().isNegative()) + return false; + // For the moment, restrict ourselves to the case where the offset is a // multiple of the requested alignment and the base is aligned. // TODO: generalize if a case found which warrants diff --git a/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll b/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll index b85d68c574a82..a7c9a18127ade 100644 --- a/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll +++ b/llvm/test/Transforms/LoopVectorize/load-deref-pred-align.ll @@ -1,7 +1,7 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py ; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 -S %s | FileCheck %s -target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:1" +target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:1-p:16:16:16:16" declare void @init(ptr nocapture nofree) @@ -17,7 +17,7 @@ define i16 @test_access_size_not_multiple_of_align(i64 %len, ptr %test_base) { ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE2:%.*]] ] -; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i16> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[PRED_LOAD_CONTINUE2]] ] +; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i16> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[PRED_LOAD_CONTINUE2]] ] ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[TEST_BASE:%.*]], i64 [[TMP0]] ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 0 @@ -43,16 +43,16 @@ define i16 @test_access_size_not_multiple_of_align(i64 %len, ptr %test_base) { ; CHECK: pred.load.continue2: ; CHECK-NEXT: [[TMP14:%.*]] = phi <2 x i16> [ [[TMP8]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP13]], [[PRED_LOAD_IF1]] ] ; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP3]], <2 x i16> [[TMP14]], <2 x i16> zeroinitializer -; CHECK-NEXT: [[TMP16]] = add <2 x i16> [[VEC_PHI]], [[PREDPHI]] +; CHECK-NEXT: [[TMP15]] = add <2 x i16> [[VEC_PHI]], [[PREDPHI]] ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 -; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096 -; CHECK-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] +; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096 +; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK: middle.block: -; CHECK-NEXT: [[TMP18:%.*]] = call i16 @llvm.vector.reduce.add.v2i16(<2 x i16> [[TMP16]]) +; CHECK-NEXT: [[TMP17:%.*]] = call i16 @llvm.vector.reduce.add.v2i16(<2 x i16> [[TMP15]]) ; CHECK-NEXT: br i1 true, label [[LOOP_EXIT:%.*]], label [[SCALAR_PH]] ; CHECK: scalar.ph: ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 4096, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] -; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i16 [ [[TMP18]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] +; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i16 [ [[TMP17]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: loop: ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LATCH:%.*]] ] @@ -72,7 +72,7 @@ define i16 @test_access_size_not_multiple_of_align(i64 %len, ptr %test_base) { ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV]], 4095 ; CHECK-NEXT: br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK: loop_exit: -; CHECK-NEXT: [[ACCUM_NEXT_LCSSA:%.*]] = phi i16 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP18]], [[MIDDLE_BLOCK]] ] +; CHECK-NEXT: [[ACCUM_NEXT_LCSSA:%.*]] = phi i16 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ] ; CHECK-NEXT: ret i16 [[ACCUM_NEXT_LCSSA]] ; entry: @@ -114,7 +114,7 @@ define i32 @test_access_size_multiple_of_align_but_offset_by_1(i64 %len, ptr %te ; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK: vector.body: ; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE2:%.*]] ] -; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[PRED_LOAD_CONTINUE2]] ] +; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP15:%.*]], [[PRED_LOAD_CONTINUE2]] ] ; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 ; CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[TEST_BASE:%.*]], i64 [[TMP0]] ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i32 0 @@ -140,16 +140,16 @@ define i32 @test_access_size_multiple_of_align_but_offset_by_1(i64 %len, ptr %te ; CHECK: pred.load.continue2: ; CHECK-NEXT: [[TMP14:%.*]] = phi <2 x i32> [ [[TMP8]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP13]], [[PRED_LOAD_IF1]] ] ; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP3]], <2 x i32> [[TMP14]], <2 x i32> zeroinitializer -; CHECK-NEXT: [[TMP16]] = add <2 x i32> [[VEC_PHI]], [[PREDPHI]] +; CHECK-NEXT: [[TMP15]] = add <2 x i32> [[VEC_PHI]], [[PREDPHI]] ; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 -; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096 -; CHECK-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] +; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 4096 +; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK: middle.block: -; CHECK-NEXT: [[TMP18:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP16]]) +; CHECK-NEXT: [[TMP17:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[TMP15]]) ; CHECK-NEXT: br i1 true, label [[LOOP_EXIT:%.*]], label [[SCALAR_PH]] ; CHECK: scalar.ph: ; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 4096, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] -; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP18]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] +; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ [[TMP17]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] ; CHECK-NEXT: br label [[LOOP:%.*]] ; CHECK: loop: ; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LATCH:%.*]] ] @@ -169,7 +169,7 @@ define i32 @test_access_size_multiple_of_align_but_offset_by_1(i64 %len, ptr %te ; CHECK-NEXT: [[EXIT:%.*]] = icmp eq i64 [[IV]], 4095 ; CHECK-NEXT: br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]] ; CHECK: loop_exit: -; CHECK-NEXT: [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP18]], [[MIDDLE_BLOCK]] ] +; CHECK-NEXT: [[ACCUM_NEXT_LCSSA:%.*]] = phi i32 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP17]], [[MIDDLE_BLOCK]] ] ; CHECK-NEXT: ret i32 [[ACCUM_NEXT_LCSSA]] ; entry: @@ -198,3 +198,101 @@ latch: loop_exit: ret i32 %accum.next } + + +; Test where offset relative to alloca is negative and we shouldn't +; treat predicated loads as being always dereferenceable. +define i8 @test_negative_off(i16 %len, ptr %test_base) { +; CHECK-LABEL: @test_negative_off( +; CHECK-NEXT: entry: +; CHECK-NEXT: [[ALLOCA:%.*]] = alloca [64638 x i8], align 1 +; CHECK-NEXT: call void @init(ptr [[ALLOCA]]) +; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] +; CHECK: vector.ph: +; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] +; CHECK: vector.body: +; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE2:%.*]] ] +; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i8> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP18:%.*]], [[PRED_LOAD_CONTINUE2]] ] +; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i32 [[INDEX]] to i16 +; CHECK-NEXT: [[OFFSET_IDX:%.*]] = add i16 -1000, [[DOTCAST]] +; CHECK-NEXT: [[TMP0:%.*]] = add i16 [[OFFSET_IDX]], 0 +; CHECK-NEXT: [[TMP1:%.*]] = add i16 [[OFFSET_IDX]], 1 +; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds i1, ptr [[TEST_BASE:%.*]], i16 [[TMP0]] +; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i1, ptr [[TEST_BASE]], i16 [[TMP1]] +; CHECK-NEXT: [[TMP4:%.*]] = load i1, ptr [[TMP2]], align 1 +; CHECK-NEXT: [[TMP5:%.*]] = load i1, ptr [[TMP3]], align 1 +; CHECK-NEXT: [[TMP6:%.*]] = insertelement <2 x i1> poison, i1 [[TMP4]], i32 0 +; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x i1> [[TMP6]], i1 [[TMP5]], i32 1 +; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x i1> [[TMP7]], i32 0 +; CHECK-NEXT: br i1 [[TMP8]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]] +; CHECK: pred.load.if: +; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[ALLOCA]], i16 [[TMP0]] +; CHECK-NEXT: [[TMP10:%.*]] = load i8, ptr [[TMP9]], align 1 +; CHECK-NEXT: [[TMP11:%.*]] = insertelement <2 x i8> poison, i8 [[TMP10]], i32 0 +; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]] +; CHECK: pred.load.continue: +; CHECK-NEXT: [[TMP12:%.*]] = phi <2 x i8> [ poison, [[VECTOR_BODY]] ], [ [[TMP11]], [[PRED_LOAD_IF]] ] +; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x i1> [[TMP7]], i32 1 +; CHECK-NEXT: br i1 [[TMP13]], label [[PRED_LOAD_IF1:%.*]], label [[PRED_LOAD_CONTINUE2]] +; CHECK: pred.load.if1: +; CHECK-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[ALLOCA]], i16 [[TMP1]] +; CHECK-NEXT: [[TMP15:%.*]] = load i8, ptr [[TMP14]], align 1 +; CHECK-NEXT: [[TMP16:%.*]] = insertelement <2 x i8> [[TMP12]], i8 [[TMP15]], i32 1 +; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE2]] +; CHECK: pred.load.continue2: +; CHECK-NEXT: [[TMP17:%.*]] = phi <2 x i8> [ [[TMP12]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP16]], [[PRED_LOAD_IF1]] ] +; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP7]], <2 x i8> [[TMP17]], <2 x i8> zeroinitializer +; CHECK-NEXT: [[TMP18]] = add <2 x i8> [[VEC_PHI]], [[PREDPHI]] +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2 +; CHECK-NEXT: [[TMP19:%.*]] = icmp eq i32 [[INDEX_NEXT]], 12 +; CHECK-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; CHECK: middle.block: +; CHECK-NEXT: [[TMP20:%.*]] = call i8 @llvm.vector.reduce.add.v2i8(<2 x i8> [[TMP18]]) +; CHECK-NEXT: br i1 true, label [[LOOP_EXIT:%.*]], label [[SCALAR_PH]] +; CHECK: scalar.ph: +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i16 [ -988, [[MIDDLE_BLOCK]] ], [ -1000, [[ENTRY:%.*]] ] +; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i8 [ [[TMP20]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ] +; CHECK-NEXT: br label [[LOOP:%.*]] +; CHECK: loop: +; CHECK-NEXT: [[IV:%.*]] = phi i16 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LATCH:%.*]] ] +; CHECK-NEXT: [[ACCUM:%.*]] = phi i8 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ACCUM_NEXT:%.*]], [[LATCH]] ] +; CHECK-NEXT: [[IV_NEXT]] = add i16 [[IV]], 1 +; CHECK-NEXT: [[TEST_ADDR:%.*]] = getelementptr inbounds i1, ptr [[TEST_BASE]], i16 [[IV]] +; CHECK-NEXT: [[EARLYCND:%.*]] = load i1, ptr [[TEST_ADDR]], align 1 +; CHECK-NEXT: br i1 [[EARLYCND]], label [[PRED:%.*]], label [[LATCH]] +; CHECK: pred: +; CHECK-NEXT: [[ADDR:%.*]] = getelementptr i8, ptr [[ALLOCA]], i16 [[IV]] +; CHECK-NEXT: [[VAL:%.*]] = load i8, ptr [[ADDR]], align 1 +; CHECK-NEXT: br label [[LATCH]] +; CHECK: latch: +; CHECK-NEXT: [[VAL_PHI:%.*]] = phi i8 [ 0, [[LOOP]] ], [ [[VAL]], [[PRED]] ] +; CHECK-NEXT: [[ACCUM_NEXT]] = add i8 [[ACCUM]], [[VAL_PHI]] +; CHECK-NEXT: [[EXIT:%.*]] = icmp ugt i16 [[IV]], -990 +; CHECK-NEXT: br i1 [[EXIT]], label [[LOOP_EXIT]], label [[LOOP]], !llvm.loop [[LOOP7:![0-9]+]] +; CHECK: loop_exit: +; CHECK-NEXT: [[ACCUM_NEXT_LCSSA:%.*]] = phi i8 [ [[ACCUM_NEXT]], [[LATCH]] ], [ [[TMP20]], [[MIDDLE_BLOCK]] ] +; CHECK-NEXT: ret i8 [[ACCUM_NEXT_LCSSA]] +; +entry: + %alloca = alloca [64638 x i8] + call void @init(ptr %alloca) + br label %loop +loop: + %iv = phi i16 [ -1000, %entry ], [ %iv.next, %latch ] + %accum = phi i8 [ 0, %entry ], [ %accum.next, %latch ] + %iv.next = add i16 %iv, 1 + %test_addr = getelementptr inbounds i1, ptr %test_base, i16 %iv + %earlycnd = load i1, ptr %test_addr + br i1 %earlycnd, label %pred, label %latch +pred: + %addr = getelementptr i8, ptr %alloca, i16 %iv + %val = load i8, ptr %addr + br label %latch +latch: + %val.phi = phi i8 [ 0, %loop ], [ %val, %pred ] + %accum.next = add i8 %accum, %val.phi + %exit = icmp ugt i16 %iv, -990 + br i1 %exit, label %loop_exit, label %loop +loop_exit: + ret i8 %accum.next +}