124 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph" --version 52; REQUIRES: asserts3 4; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-ios %s -S -debug -disable-output 2>&1 | FileCheck --check-prefix=CM %s5; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 %s -S | FileCheck --check-prefix=FORCED %s6 7; Test case from PR41294.8 9; Check scalar cost for extractvalue. The constant and loop invariant operands are free,10; leaving cost 3 for scalarizing the result + 2 for executing the op with VF 2.11 12; CM: LV: Found uniform instruction: %a = extractvalue { i64, i64 } %sv, 013; CM: LV: Found uniform instruction: %b = extractvalue { i64, i64 } %sv, 114 15; Ensure the extractvalue + add instructions are hoisted out16; CM: vector.ph:17; CM: CLONE ir<%a> = extractvalue ir<%sv>18; CM: CLONE ir<%b> = extractvalue ir<%sv>19; CM: CLONE ir<%add> = add ir<%a>, ir<%b>20; CM: Successor(s): vector loop21 22; CM: LV: Scalar loop costs: 5.23 24; Check that the extractvalue operands are actually free in vector code.25 26define void @test1(ptr %dst, {i64, i64} %sv) {27; FORCED-LABEL: define void @test1(28; FORCED-SAME: ptr [[DST:%.*]], { i64, i64 } [[SV:%.*]]) {29; FORCED-NEXT: [[ENTRY:.*:]]30; FORCED-NEXT: br label %[[VECTOR_PH:.*]]31; FORCED: [[VECTOR_PH]]:32; FORCED-NEXT: [[TMP0:%.*]] = extractvalue { i64, i64 } [[SV]], 033; FORCED-NEXT: [[TMP4:%.*]] = extractvalue { i64, i64 } [[SV]], 134; FORCED-NEXT: [[TMP5:%.*]] = add i64 [[TMP0]], [[TMP4]]35; FORCED-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x i64> poison, i64 [[TMP5]], i64 036; FORCED-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x i64> [[BROADCAST_SPLATINSERT1]], <2 x i64> poison, <2 x i32> zeroinitializer37; FORCED-NEXT: br label %[[VECTOR_BODY:.*]]38; FORCED: [[VECTOR_BODY]]:39; FORCED-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]40; FORCED-NEXT: [[TMP2:%.*]] = getelementptr i64, ptr [[DST]], i32 [[INDEX]]41; FORCED-NEXT: store <2 x i64> [[BROADCAST_SPLAT2]], ptr [[TMP2]], align 442; FORCED-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 243; FORCED-NEXT: [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 100044; FORCED-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]45; FORCED: [[MIDDLE_BLOCK]]:46; FORCED-NEXT: br label %[[EXIT:.*]]47; FORCED: [[EXIT]]:48; FORCED-NEXT: ret void49;50entry:51 br label %loop.body52 53loop.body:54 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.body ]55 %a = extractvalue { i64, i64 } %sv, 056 %b = extractvalue { i64, i64 } %sv, 157 %addr = getelementptr i64, ptr %dst, i32 %iv58 %add = add i64 %a, %b59 store i64 %add, ptr %addr60 %iv.next = add nsw i32 %iv, 161 %cond = icmp ne i32 %iv.next, 100062 br i1 %cond, label %loop.body, label %exit63 64exit:65 ret void66}67 68 69; Similar to the test case above, but checks getVectorCallCost as well.70declare float @powf(float, float) readnone nounwind71 72; Ensure the extractvalue instructions are hoisted out73; CM-LABEL: Checking a loop in 'test_getVectorCallCost'74; CM: vector.ph:75; CM: CLONE ir<%a> = extractvalue ir<%sv>76; CM: CLONE ir<%b> = extractvalue ir<%sv>77; CM: Successor(s): vector loop78 79; CM: LV: Scalar loop costs: 14.80 81define void @test_getVectorCallCost(ptr %dst, {float, float} %sv) {82; FORCED-LABEL: define void @test_getVectorCallCost(83; FORCED-SAME: ptr [[DST:%.*]], { float, float } [[SV:%.*]]) {84; FORCED-NEXT: [[ENTRY:.*:]]85; FORCED-NEXT: br label %[[VECTOR_PH:.*]]86; FORCED: [[VECTOR_PH]]:87; FORCED-NEXT: [[TMP0:%.*]] = extractvalue { float, float } [[SV]], 088; FORCED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x float> poison, float [[TMP0]], i64 089; FORCED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT]], <2 x float> poison, <2 x i32> zeroinitializer90; FORCED-NEXT: [[TMP4:%.*]] = extractvalue { float, float } [[SV]], 191; FORCED-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <2 x float> poison, float [[TMP4]], i64 092; FORCED-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <2 x float> [[BROADCAST_SPLATINSERT1]], <2 x float> poison, <2 x i32> zeroinitializer93; FORCED-NEXT: br label %[[VECTOR_BODY:.*]]94; FORCED: [[VECTOR_BODY]]:95; FORCED-NEXT: [[INDEX:%.*]] = phi i32 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]96; FORCED-NEXT: [[TMP1:%.*]] = getelementptr float, ptr [[DST]], i32 [[INDEX]]97; FORCED-NEXT: [[TMP2:%.*]] = call <2 x float> @llvm.pow.v2f32(<2 x float> [[BROADCAST_SPLAT]], <2 x float> [[BROADCAST_SPLAT2]])98; FORCED-NEXT: store <2 x float> [[TMP2]], ptr [[TMP1]], align 499; FORCED-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 2100; FORCED-NEXT: [[TMP3:%.*]] = icmp eq i32 [[INDEX_NEXT]], 1000101; FORCED-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]102; FORCED: [[MIDDLE_BLOCK]]:103; FORCED-NEXT: br label %[[EXIT:.*]]104; FORCED: [[EXIT]]:105; FORCED-NEXT: ret void106;107entry:108 br label %loop.body109 110loop.body:111 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.body ]112 %a = extractvalue { float, float } %sv, 0113 %b = extractvalue { float, float } %sv, 1114 %addr = getelementptr float, ptr %dst, i32 %iv115 %p = call float @powf(float %a, float %b)116 store float %p, ptr %addr117 %iv.next = add nsw i32 %iv, 1118 %cond = icmp ne i32 %iv.next, 1000119 br i1 %cond, label %loop.body, label %exit120 121exit:122 ret void123}124