[llvm] c7ee57f - [LV] Directly use incoming value for single VPBlendRecipes.
Florian Hahn via llvm-commits
llvm-commits at lists.llvm.org
Mon Feb 22 08:11:38 PST 2021
Author: Florian Hahn
Date: 2021-02-22T16:10:08Z
New Revision: c7ee57f1dccf965aa73ad708937443d649fbb0a7
URL: https://github.com/llvm/llvm-project/commit/c7ee57f1dccf965aa73ad708937443d649fbb0a7
DIFF: https://github.com/llvm/llvm-project/commit/c7ee57f1dccf965aa73ad708937443d649fbb0a7.diff
LOG: [LV] Directly use incoming value for single VPBlendRecipes.
VPBlendRecipes with single incoming (value, mask) pair are no-ops. Use
the incoming value directly.
Added:
llvm/test/Transforms/LoopVectorize/single-value-blend-phis.ll
Modified:
llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
Removed:
################################################################################
diff --git a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
index 625a82548150..006fcc7b6f4e 100644
--- a/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/llvm/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -8604,6 +8604,15 @@ VPlanPtr LoopVectorizationPlanner::buildVPlanWithVPRecipes(
if (auto Recipe =
RecipeBuilder.tryToCreateWidenRecipe(Instr, Range, Plan)) {
+
+ // VPBlendRecipes with a single incoming (value, mask) pair are no-ops.
+ // Use the incoming value directly.
+ if (isa<VPBlendRecipe>(Recipe) && Recipe->getNumOperands() <= 2) {
+ Plan->removeVPValueFor(Instr);
+ Plan->addVPValue(Instr, Recipe->getOperand(0));
+ delete Recipe;
+ continue;
+ }
for (auto *Def : Recipe->definedValues()) {
auto *UV = Def->getUnderlyingValue();
Plan->addVPValue(UV, Def);
diff --git a/llvm/test/Transforms/LoopVectorize/single-value-blend-phis.ll b/llvm/test/Transforms/LoopVectorize/single-value-blend-phis.ll
new file mode 100644
index 000000000000..b98a3ab7336d
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/single-value-blend-phis.ll
@@ -0,0 +1,384 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -loop-vectorize -force-vector-width=2 -S %s | FileCheck %s
+
+ at src = external global [32 x i16], align 1
+ at dst = external global [32 x i16], align 1
+
+; The load in the loop does not need predication, because the accessed memory
+; is de-referenceable for all loop iterations.
+define void @single_incoming_phi_no_blend_mask(i64 %a, i64 %b) {
+; CHECK-LABEL: @single_incoming_phi_no_blend_mask(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDEX]] to i16
+; CHECK-NEXT: [[TMP3:%.*]] = add i16 [[TMP2]], 0
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP3]]
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i16, i16* [[TMP4]], i32 0
+; CHECK-NEXT: [[TMP6:%.*]] = bitcast i16* [[TMP5]] to <2 x i16>*
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP6]], align 1
+; CHECK-NEXT: [[TMP7:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP8:%.*]] = xor <2 x i1> [[TMP7]], <i1 true, i1 true>
+; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP7]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[WIDE_LOAD]]
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i16, i16* [[TMP9]], i32 0
+; CHECK-NEXT: [[TMP11:%.*]] = bitcast i16* [[TMP10]] to <2 x i16>*
+; CHECK-NEXT: store <2 x i16> [[PREDPHI]], <2 x i16>* [[TMP11]], align 2
+; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
+; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
+; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 32, 32
+; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
+; CHECK: loop.header:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i16
+; CHECK-NEXT: br label [[LOOP_COND:%.*]]
+; CHECK: loop.cond:
+; CHECK-NEXT: [[BLEND:%.*]] = phi i16 [ [[IV_TRUNC]], [[LOOP_HEADER]] ]
+; CHECK-NEXT: [[SRC_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[BLEND]]
+; CHECK-NEXT: [[LV:%.*]] = load i16, i16* [[SRC_PTR]], align 1
+; CHECK-NEXT: [[CMP_B:%.*]] = icmp sgt i64 [[IV]], [[A]]
+; CHECK-NEXT: br i1 [[CMP_B]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
+; CHECK: loop.next:
+; CHECK-NEXT: br label [[LOOP_LATCH]]
+; CHECK: loop.latch:
+; CHECK-NEXT: [[RES:%.*]] = phi i16 [ [[LV]], [[LOOP_COND]] ], [ 1, [[LOOP_NEXT]] ]
+; CHECK-NEXT: [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[IV]]
+; CHECK-NEXT: store i16 [[RES]], i16* [[DST_PTR]], align 2
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
+; CHECK-NEXT: br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], [[LOOP2:!llvm.loop !.*]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop.header
+
+loop.header:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
+ %iv.trunc = trunc i64 %iv to i16
+ br label %loop.cond
+
+loop.cond:
+ %blend = phi i16 [ %iv.trunc, %loop.header ]
+ %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
+ %lv = load i16, i16* %src.ptr, align 1
+ %cmp.b = icmp sgt i64 %iv, %a
+ br i1 %cmp.b, label %loop.next, label %loop.latch
+
+loop.next:
+ br label %loop.latch
+
+loop.latch:
+ %res = phi i16 [ %lv, %loop.cond ], [ 1, %loop.next ]
+ %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
+ store i16 %res, i16* %dst.ptr
+ %iv.next = add nuw nsw i64 %iv, 1
+ %cmp439 = icmp ult i64 %iv, 31
+ br i1 %cmp439, label %loop.header, label %exit
+
+exit:
+ ret void
+}
+
+; The load in the loop does not need predication, because the accessed memory
+; is de-referenceable for all loop iterations.
+define void @single_incoming_phi_with_blend_mask(i64 %a, i64 %b) {
+; CHECK-LABEL: @single_incoming_phi_with_blend_mask(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = trunc i64 [[INDEX]] to i16
+; CHECK-NEXT: [[TMP3:%.*]] = add i16 [[TMP2]], 0
+; CHECK-NEXT: [[TMP4:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP3]]
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i16, i16* [[TMP5]], i32 0
+; CHECK-NEXT: [[TMP7:%.*]] = bitcast i16* [[TMP6]] to <2 x i16>*
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP7]], align 1
+; CHECK-NEXT: [[TMP8:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP9:%.*]] = xor <2 x i1> [[TMP4]], <i1 true, i1 true>
+; CHECK-NEXT: [[TMP10:%.*]] = xor <2 x i1> [[TMP8]], <i1 true, i1 true>
+; CHECK-NEXT: [[TMP11:%.*]] = select <2 x i1> [[TMP4]], <2 x i1> [[TMP10]], <2 x i1> zeroinitializer
+; CHECK-NEXT: [[TMP12:%.*]] = select <2 x i1> [[TMP4]], <2 x i1> [[TMP8]], <2 x i1> zeroinitializer
+; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP11]], <2 x i16> [[WIDE_LOAD]], <2 x i16> zeroinitializer
+; CHECK-NEXT: [[PREDPHI1:%.*]] = select <2 x i1> [[TMP12]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[PREDPHI]]
+; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i16, i16* [[TMP13]], i32 0
+; CHECK-NEXT: [[TMP15:%.*]] = bitcast i16* [[TMP14]] to <2 x i16>*
+; CHECK-NEXT: store <2 x i16> [[PREDPHI1]], <2 x i16>* [[TMP15]], align 2
+; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
+; CHECK-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
+; CHECK-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 32, 32
+; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
+; CHECK: loop.header:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i16
+; CHECK-NEXT: [[CMP_A:%.*]] = icmp ugt i64 [[IV]], [[A]]
+; CHECK-NEXT: br i1 [[CMP_A]], label [[LOOP_COND:%.*]], label [[LOOP_LATCH]]
+; CHECK: loop.cond:
+; CHECK-NEXT: [[BLEND:%.*]] = phi i16 [ [[IV_TRUNC]], [[LOOP_HEADER]] ]
+; CHECK-NEXT: [[SRC_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[BLEND]]
+; CHECK-NEXT: [[LV:%.*]] = load i16, i16* [[SRC_PTR]], align 1
+; CHECK-NEXT: [[CMP_B:%.*]] = icmp sgt i64 [[IV]], [[A]]
+; CHECK-NEXT: br i1 [[CMP_B]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
+; CHECK: loop.next:
+; CHECK-NEXT: br label [[LOOP_LATCH]]
+; CHECK: loop.latch:
+; CHECK-NEXT: [[RES:%.*]] = phi i16 [ 0, [[LOOP_HEADER]] ], [ [[LV]], [[LOOP_COND]] ], [ 1, [[LOOP_NEXT]] ]
+; CHECK-NEXT: [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[IV]]
+; CHECK-NEXT: store i16 [[RES]], i16* [[DST_PTR]], align 2
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
+; CHECK-NEXT: br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], [[LOOP5:!llvm.loop !.*]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop.header
+
+loop.header:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
+ %iv.trunc = trunc i64 %iv to i16
+ %cmp.a = icmp ugt i64 %iv, %a
+ br i1 %cmp.a, label %loop.cond, label %loop.latch
+
+loop.cond:
+ %blend = phi i16 [ %iv.trunc, %loop.header ]
+ %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
+ %lv = load i16, i16* %src.ptr, align 1
+ %cmp.b = icmp sgt i64 %iv, %a
+ br i1 %cmp.b, label %loop.next, label %loop.latch
+
+loop.next:
+ br label %loop.latch
+
+loop.latch:
+ %res = phi i16 [ 0, %loop.header ], [ %lv, %loop.cond ], [ 1, %loop.next ]
+ %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
+ store i16 %res, i16* %dst.ptr
+ %iv.next = add nuw nsw i64 %iv, 1
+ %cmp439 = icmp ult i64 %iv, 31
+ br i1 %cmp439, label %loop.header, label %exit
+
+exit:
+ ret void
+}
+
+define void @multiple_incoming_phi_with_blend_mask(i64 %a) {
+; CHECK-LABEL: @multiple_incoming_phi_with_blend_mask(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND1:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[VEC_IND3:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT4:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP3:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
+; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP2]], <2 x i16> [[VEC_IND3]], <2 x i16> [[VEC_IND1]]
+; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 0
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP4]]
+; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x i16> [[PREDPHI]], i32 1
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP6]]
+; CHECK-NEXT: [[TMP8:%.*]] = load i16, i16* [[TMP5]], align 1
+; CHECK-NEXT: [[TMP9:%.*]] = load i16, i16* [[TMP7]], align 1
+; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
+; CHECK-NEXT: [[VEC_IND_NEXT2]] = add <2 x i16> [[VEC_IND1]], <i16 2, i16 2>
+; CHECK-NEXT: [[VEC_IND_NEXT4]] = add <2 x i16> [[VEC_IND3]], <i16 2, i16 2>
+; CHECK-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], 32
+; CHECK-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP6:!llvm.loop !.*]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 32, 32
+; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 32, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
+; CHECK: loop.header:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i16
+; CHECK-NEXT: [[IV_TRUNC_2:%.*]] = trunc i64 [[IV]] to i16
+; CHECK-NEXT: [[CMP_A:%.*]] = icmp ugt i64 [[IV]], [[A]]
+; CHECK-NEXT: br i1 [[CMP_A]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
+; CHECK: loop.next:
+; CHECK-NEXT: br label [[LOOP_LATCH]]
+; CHECK: loop.latch:
+; CHECK-NEXT: [[BLEND:%.*]] = phi i16 [ [[IV_TRUNC]], [[LOOP_HEADER]] ], [ [[IV_TRUNC_2]], [[LOOP_NEXT]] ]
+; CHECK-NEXT: [[SRC_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[BLEND]]
+; CHECK-NEXT: [[LV:%.*]] = load i16, i16* [[SRC_PTR]], align 1
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[CMP439:%.*]] = icmp ult i64 [[IV]], 31
+; CHECK-NEXT: br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], [[LOOP7:!llvm.loop !.*]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop.header
+
+loop.header:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
+ %iv.trunc = trunc i64 %iv to i16
+ %iv.trunc.2 = trunc i64 %iv to i16
+ %cmp.a = icmp ugt i64 %iv, %a
+ br i1 %cmp.a, label %loop.next, label %loop.latch
+
+loop.next:
+ br label %loop.latch
+
+loop.latch:
+ %blend = phi i16 [ %iv.trunc, %loop.header ], [ %iv.trunc.2, %loop.next ]
+ %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
+ %lv = load i16, i16* %src.ptr, align 1
+ %iv.next = add nuw nsw i64 %iv, 1
+ %cmp439 = icmp ult i64 %iv, 31
+ br i1 %cmp439, label %loop.header, label %exit
+
+exit:
+ ret void
+}
+
+; The load in the loop needs predication, because the accessed memory is not
+; de-referencable for all iterations of the loop.
+define void @single_incoming_needs_predication(i64 %a, i64 %b) {
+; CHECK-LABEL: @single_incoming_needs_predication(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i64> poison, i64 [[A:%.*]], i32 0
+; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT]], <2 x i64> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[PRED_LOAD_CONTINUE4:%.*]] ]
+; CHECK-NEXT: [[VEC_IND:%.*]] = phi <2 x i64> [ <i64 0, i64 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[PRED_LOAD_CONTINUE4]] ]
+; CHECK-NEXT: [[VEC_IND1:%.*]] = phi <2 x i16> [ <i16 0, i16 1>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT2:%.*]], [[PRED_LOAD_CONTINUE4]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1
+; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x i1> [[TMP2]], i32 0
+; CHECK-NEXT: br i1 [[TMP3]], label [[PRED_LOAD_IF:%.*]], label [[PRED_LOAD_CONTINUE:%.*]]
+; CHECK: pred.load.if:
+; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x i16> [[VEC_IND1]], i32 0
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP4]]
+; CHECK-NEXT: [[TMP6:%.*]] = load i16, i16* [[TMP5]], align 1
+; CHECK-NEXT: [[TMP7:%.*]] = insertelement <2 x i16> poison, i16 [[TMP6]], i32 0
+; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE]]
+; CHECK: pred.load.continue:
+; CHECK-NEXT: [[TMP8:%.*]] = phi <2 x i16> [ poison, [[VECTOR_BODY]] ], [ [[TMP7]], [[PRED_LOAD_IF]] ]
+; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x i1> [[TMP2]], i32 1
+; CHECK-NEXT: br i1 [[TMP9]], label [[PRED_LOAD_IF3:%.*]], label [[PRED_LOAD_CONTINUE4]]
+; CHECK: pred.load.if3:
+; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x i16> [[VEC_IND1]], i32 1
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[TMP10]]
+; CHECK-NEXT: [[TMP12:%.*]] = load i16, i16* [[TMP11]], align 1
+; CHECK-NEXT: [[TMP13:%.*]] = insertelement <2 x i16> [[TMP8]], i16 [[TMP12]], i32 1
+; CHECK-NEXT: br label [[PRED_LOAD_CONTINUE4]]
+; CHECK: pred.load.continue4:
+; CHECK-NEXT: [[TMP14:%.*]] = phi <2 x i16> [ [[TMP8]], [[PRED_LOAD_CONTINUE]] ], [ [[TMP13]], [[PRED_LOAD_IF3]] ]
+; CHECK-NEXT: [[TMP15:%.*]] = icmp sgt <2 x i64> [[VEC_IND]], [[BROADCAST_SPLAT]]
+; CHECK-NEXT: [[TMP16:%.*]] = xor <2 x i1> [[TMP2]], <i1 true, i1 true>
+; CHECK-NEXT: [[TMP17:%.*]] = xor <2 x i1> [[TMP15]], <i1 true, i1 true>
+; CHECK-NEXT: [[TMP18:%.*]] = select <2 x i1> [[TMP2]], <2 x i1> [[TMP17]], <2 x i1> zeroinitializer
+; CHECK-NEXT: [[TMP19:%.*]] = select <2 x i1> [[TMP2]], <2 x i1> [[TMP15]], <2 x i1> zeroinitializer
+; CHECK-NEXT: [[PREDPHI:%.*]] = select <2 x i1> [[TMP18]], <2 x i16> [[TMP14]], <2 x i16> zeroinitializer
+; CHECK-NEXT: [[PREDPHI5:%.*]] = select <2 x i1> [[TMP19]], <2 x i16> <i16 1, i16 1>, <2 x i16> [[PREDPHI]]
+; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[TMP0]]
+; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i16, i16* [[TMP20]], i32 0
+; CHECK-NEXT: [[TMP22:%.*]] = bitcast i16* [[TMP21]] to <2 x i16>*
+; CHECK-NEXT: store <2 x i16> [[PREDPHI5]], <2 x i16>* [[TMP22]], align 2
+; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
+; CHECK-NEXT: [[VEC_IND_NEXT]] = add <2 x i64> [[VEC_IND]], <i64 2, i64 2>
+; CHECK-NEXT: [[VEC_IND_NEXT2]] = add <2 x i16> [[VEC_IND1]], <i16 2, i16 2>
+; CHECK-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], 64
+; CHECK-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP8:!llvm.loop !.*]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 64, 64
+; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
+; CHECK: scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 64, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
+; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
+; CHECK: loop.header:
+; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
+; CHECK-NEXT: [[IV_TRUNC:%.*]] = trunc i64 [[IV]] to i16
+; CHECK-NEXT: [[CMP_A:%.*]] = icmp ugt i64 [[IV]], [[A]]
+; CHECK-NEXT: br i1 [[CMP_A]], label [[LOOP_COND:%.*]], label [[LOOP_LATCH]]
+; CHECK: loop.cond:
+; CHECK-NEXT: [[BLEND:%.*]] = phi i16 [ [[IV_TRUNC]], [[LOOP_HEADER]] ]
+; CHECK-NEXT: [[SRC_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 [[BLEND]]
+; CHECK-NEXT: [[LV:%.*]] = load i16, i16* [[SRC_PTR]], align 1
+; CHECK-NEXT: [[CMP_B:%.*]] = icmp sgt i64 [[IV]], [[A]]
+; CHECK-NEXT: br i1 [[CMP_B]], label [[LOOP_NEXT:%.*]], label [[LOOP_LATCH]]
+; CHECK: loop.next:
+; CHECK-NEXT: br label [[LOOP_LATCH]]
+; CHECK: loop.latch:
+; CHECK-NEXT: [[RES:%.*]] = phi i16 [ 0, [[LOOP_HEADER]] ], [ [[LV]], [[LOOP_COND]] ], [ 1, [[LOOP_NEXT]] ]
+; CHECK-NEXT: [[DST_PTR:%.*]] = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 [[IV]]
+; CHECK-NEXT: store i16 [[RES]], i16* [[DST_PTR]], align 2
+; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
+; CHECK-NEXT: [[CMP439:%.*]] = icmp ult i64 [[IV]], 63
+; CHECK-NEXT: br i1 [[CMP439]], label [[LOOP_HEADER]], label [[EXIT]], [[LOOP9:!llvm.loop !.*]]
+; CHECK: exit:
+; CHECK-NEXT: ret void
+;
+entry:
+ br label %loop.header
+
+loop.header:
+ %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
+ %iv.trunc = trunc i64 %iv to i16
+ %cmp.a = icmp ugt i64 %iv, %a
+ br i1 %cmp.a, label %loop.cond, label %loop.latch
+
+loop.cond:
+ %blend = phi i16 [ %iv.trunc, %loop.header ]
+ %src.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @src, i16 0, i16 %blend
+ %lv = load i16, i16* %src.ptr, align 1
+ %cmp.b = icmp sgt i64 %iv, %a
+ br i1 %cmp.b, label %loop.next, label %loop.latch
+
+loop.next:
+ br label %loop.latch
+
+loop.latch:
+ %res = phi i16 [ 0, %loop.header ], [ %lv, %loop.cond ], [ 1, %loop.next ]
+ %dst.ptr = getelementptr inbounds [32 x i16], [32 x i16]* @dst, i16 0, i64 %iv
+ store i16 %res, i16* %dst.ptr
+ %iv.next = add nuw nsw i64 %iv, 1
+ %cmp439 = icmp ult i64 %iv, 63
+ br i1 %cmp439, label %loop.header, label %exit
+
+exit:
+ ret void
+}
More information about the llvm-commits
mailing list