From b822a32659c3fd5afd75c8f028aaa142fbd68e20 Mon Sep 17 00:00:00 2001 From: Florian Hahn Date: Sun, 29 Jun 2025 21:29:12 +0100 Subject: [PATCH] [VPlan] Fix crash when trying to narrow interleave group storing const. Use dyn_cast_null to handle the case where an interleave groups stores a constant in any of its lanes. --- .../Transforms/Vectorize/VPlanTransforms.cpp | 2 +- ...-narrow-interleave-to-widen-memory-cost.ll | 137 ++++++++++++++++++ 2 files changed, 138 insertions(+), 1 deletion(-) diff --git a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp index bcfb889469ea..7298b24fb559 100644 --- a/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp +++ b/llvm/lib/Transforms/Vectorize/VPlanTransforms.cpp @@ -3251,7 +3251,7 @@ void VPlanTransforms::narrowInterleaveGroups(VPlan &Plan, ElementCount VF, if (!WideMember0) return; for (const auto &[I, V] : enumerate(InterleaveR->getStoredValues())) { - auto *R = dyn_cast(V->getDefiningRecipe()); + auto *R = dyn_cast_or_null(V->getDefiningRecipe()); if (!R || R->getOpcode() != WideMember0->getOpcode() || R->getNumOperands() > 2) return; diff --git a/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll b/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll index 217127765054..1ca118572d54 100644 --- a/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll +++ b/llvm/test/Transforms/LoopVectorize/AArch64/transform-narrow-interleave-to-widen-memory-cost.ll @@ -141,3 +141,140 @@ loop: exit: ret void } + +define void @test_interleave_store_one_constant(ptr noalias %src, ptr noalias %dst, i64 %n) #0 { +; CHECK-LABEL: define void @test_interleave_store_one_constant( +; CHECK-SAME: ptr noalias [[SRC:%.*]], ptr noalias [[DST:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] { +; CHECK-NEXT: [[ITER_CHECK:.*]]: +; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[N]], 1 +; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2 +; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_SCEVCHECK:.*]] +; CHECK: [[VECTOR_SCEVCHECK]]: +; CHECK-NEXT: [[MUL1:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]]) +; CHECK-NEXT: [[MUL_RESULT:%.*]] = extractvalue { i64, i1 } [[MUL1]], 0 +; CHECK-NEXT: [[MUL_OVERFLOW:%.*]] = extractvalue { i64, i1 } [[MUL1]], 1 +; CHECK-NEXT: [[TMP1:%.*]] = sub i64 0, [[MUL_RESULT]] +; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[DST]], i64 [[MUL_RESULT]] +; CHECK-NEXT: [[TMP3:%.*]] = icmp ult ptr [[TMP2]], [[DST]] +; CHECK-NEXT: [[TMP4:%.*]] = or i1 [[TMP3]], [[MUL_OVERFLOW]] +; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[DST]], i64 8 +; CHECK-NEXT: [[MUL2:%.*]] = call { i64, i1 } @llvm.umul.with.overflow.i64(i64 16, i64 [[N]]) +; CHECK-NEXT: [[MUL_RESULT3:%.*]] = extractvalue { i64, i1 } [[MUL2]], 0 +; CHECK-NEXT: [[MUL_OVERFLOW4:%.*]] = extractvalue { i64, i1 } [[MUL2]], 1 +; CHECK-NEXT: [[TMP5:%.*]] = sub i64 0, [[MUL_RESULT3]] +; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[SCEVGEP]], i64 [[MUL_RESULT3]] +; CHECK-NEXT: [[TMP7:%.*]] = icmp ult ptr [[TMP6]], [[SCEVGEP]] +; CHECK-NEXT: [[TMP8:%.*]] = or i1 [[TMP7]], [[MUL_OVERFLOW4]] +; CHECK-NEXT: [[TMP9:%.*]] = or i1 [[TMP4]], [[TMP8]] +; CHECK-NEXT: br i1 [[TMP9]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]] +; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]: +; CHECK-NEXT: [[MIN_ITERS_CHECK5:%.*]] = icmp ult i64 [[TMP0]], 8 +; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK5]], label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]] +; CHECK: [[VECTOR_PH]]: +; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 8 +; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] +; CHECK-NEXT: br label %[[VECTOR_BODY:.*]] +; CHECK: [[VECTOR_BODY]]: +; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 2 +; CHECK-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], 4 +; CHECK-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 6 +; CHECK-NEXT: [[TMP13:%.*]] = getelementptr double, ptr [[SRC]], i64 [[INDEX]] +; CHECK-NEXT: [[TMP14:%.*]] = getelementptr double, ptr [[TMP13]], i32 0 +; CHECK-NEXT: [[TMP15:%.*]] = getelementptr double, ptr [[TMP13]], i32 2 +; CHECK-NEXT: [[TMP16:%.*]] = getelementptr double, ptr [[TMP13]], i32 4 +; CHECK-NEXT: [[TMP17:%.*]] = getelementptr double, ptr [[TMP13]], i32 6 +; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x double>, ptr [[TMP14]], align 8 +; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <2 x double>, ptr [[TMP15]], align 8 +; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x double>, ptr [[TMP16]], align 8 +; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x double>, ptr [[TMP17]], align 8 +; CHECK-NEXT: [[TMP18:%.*]] = fmul <2 x double> [[WIDE_LOAD]], splat (double 5.000000e+00) +; CHECK-NEXT: [[TMP19:%.*]] = fmul <2 x double> [[WIDE_LOAD6]], splat (double 5.000000e+00) +; CHECK-NEXT: [[TMP20:%.*]] = fmul <2 x double> [[WIDE_LOAD7]], splat (double 5.000000e+00) +; CHECK-NEXT: [[TMP21:%.*]] = fmul <2 x double> [[WIDE_LOAD8]], splat (double 5.000000e+00) +; CHECK-NEXT: [[TMP22:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[INDEX]] +; CHECK-NEXT: [[TMP23:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[TMP10]] +; CHECK-NEXT: [[TMP24:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[TMP11]] +; CHECK-NEXT: [[TMP25:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[TMP12]] +; CHECK-NEXT: [[TMP26:%.*]] = shufflevector <2 x double> [[TMP18]], <2 x double> zeroinitializer, <4 x i32> +; CHECK-NEXT: [[INTERLEAVED_VEC:%.*]] = shufflevector <4 x double> [[TMP26]], <4 x double> poison, <4 x i32> +; CHECK-NEXT: store <4 x double> [[INTERLEAVED_VEC]], ptr [[TMP22]], align 8 +; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <2 x double> [[TMP19]], <2 x double> zeroinitializer, <4 x i32> +; CHECK-NEXT: [[INTERLEAVED_VEC9:%.*]] = shufflevector <4 x double> [[TMP27]], <4 x double> poison, <4 x i32> +; CHECK-NEXT: store <4 x double> [[INTERLEAVED_VEC9]], ptr [[TMP23]], align 8 +; CHECK-NEXT: [[TMP28:%.*]] = shufflevector <2 x double> [[TMP20]], <2 x double> zeroinitializer, <4 x i32> +; CHECK-NEXT: [[INTERLEAVED_VEC10:%.*]] = shufflevector <4 x double> [[TMP28]], <4 x double> poison, <4 x i32> +; CHECK-NEXT: store <4 x double> [[INTERLEAVED_VEC10]], ptr [[TMP24]], align 8 +; CHECK-NEXT: [[TMP29:%.*]] = shufflevector <2 x double> [[TMP21]], <2 x double> zeroinitializer, <4 x i32> +; CHECK-NEXT: [[INTERLEAVED_VEC11:%.*]] = shufflevector <4 x double> [[TMP29]], <4 x double> poison, <4 x i32> +; CHECK-NEXT: store <4 x double> [[INTERLEAVED_VEC11]], ptr [[TMP25]], align 8 +; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 +; CHECK-NEXT: [[TMP30:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] +; CHECK-NEXT: br i1 [[TMP30]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] +; CHECK: [[MIDDLE_BLOCK]]: +; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] +; CHECK-NEXT: br i1 [[CMP_N]], label %[[EXIT:.*]], label %[[VEC_EPILOG_ITER_CHECK:.*]] +; CHECK: [[VEC_EPILOG_ITER_CHECK]]: +; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[TMP0]], [[N_VEC]] +; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 2 +; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]] +; CHECK: [[VEC_EPILOG_PH]]: +; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ] +; CHECK-NEXT: [[N_MOD_VF12:%.*]] = urem i64 [[TMP0]], 2 +; CHECK-NEXT: [[N_VEC13:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF12]] +; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]] +; CHECK: [[VEC_EPILOG_VECTOR_BODY]]: +; CHECK-NEXT: [[INDEX14:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT17:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ] +; CHECK-NEXT: [[TMP31:%.*]] = getelementptr double, ptr [[SRC]], i64 [[INDEX14]] +; CHECK-NEXT: [[TMP32:%.*]] = getelementptr double, ptr [[TMP31]], i32 0 +; CHECK-NEXT: [[WIDE_LOAD15:%.*]] = load <2 x double>, ptr [[TMP32]], align 8 +; CHECK-NEXT: [[TMP33:%.*]] = fmul <2 x double> [[WIDE_LOAD15]], splat (double 5.000000e+00) +; CHECK-NEXT: [[TMP34:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[INDEX14]] +; CHECK-NEXT: [[TMP35:%.*]] = shufflevector <2 x double> [[TMP33]], <2 x double> zeroinitializer, <4 x i32> +; CHECK-NEXT: [[INTERLEAVED_VEC16:%.*]] = shufflevector <4 x double> [[TMP35]], <4 x double> poison, <4 x i32> +; CHECK-NEXT: store <4 x double> [[INTERLEAVED_VEC16]], ptr [[TMP34]], align 8 +; CHECK-NEXT: [[INDEX_NEXT17]] = add nuw i64 [[INDEX14]], 2 +; CHECK-NEXT: [[TMP36:%.*]] = icmp eq i64 [[INDEX_NEXT17]], [[N_VEC13]] +; CHECK-NEXT: br i1 [[TMP36]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] +; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]: +; CHECK-NEXT: [[CMP_N18:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC13]] +; CHECK-NEXT: br i1 [[CMP_N18]], label %[[EXIT]], label %[[VEC_EPILOG_SCALAR_PH]] +; CHECK: [[VEC_EPILOG_SCALAR_PH]]: +; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC13]], %[[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_SCEVCHECK]] ], [ 0, %[[ITER_CHECK]] ] +; CHECK-NEXT: br label %[[LOOP:.*]] +; CHECK: [[LOOP]]: +; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], %[[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], %[[LOOP]] ] +; CHECK-NEXT: [[GEP_SRC:%.*]] = getelementptr double, ptr [[SRC]], i64 [[IV]] +; CHECK-NEXT: [[L:%.*]] = load double, ptr [[GEP_SRC]], align 8 +; CHECK-NEXT: [[MUL:%.*]] = fmul double [[L]], 5.000000e+00 +; CHECK-NEXT: [[GEP_DST:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[IV]] +; CHECK-NEXT: store double [[MUL]], ptr [[GEP_DST]], align 8 +; CHECK-NEXT: [[GEP_DST_1:%.*]] = getelementptr [2 x double], ptr [[DST]], i64 [[IV]], i32 1 +; CHECK-NEXT: store double 0.000000e+00, ptr [[GEP_DST_1]], align 8 +; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1 +; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV]], [[N]] +; CHECK-NEXT: br i1 [[EC]], label %[[EXIT]], label %[[LOOP]], !llvm.loop [[LOOP8:![0-9]+]] +; CHECK: [[EXIT]]: +; CHECK-NEXT: ret void +; +entry: + br label %loop + +loop: ; preds = %loop, %entry + %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] + %gep.src = getelementptr double, ptr %src, i64 %iv + %l = load double, ptr %gep.src, align 8 + %mul = fmul double %l, 5.000000e+00 + %gep.dst = getelementptr [2 x double], ptr %dst, i64 %iv + store double %mul, ptr %gep.dst, align 8 + %gep.dst.1 = getelementptr [2 x double], ptr %dst, i64 %iv, i32 1 + store double 0.000000e+00, ptr %gep.dst.1, align 8 + %iv.next = add i64 %iv, 1 + %ec = icmp eq i64 %iv, %n + br i1 %ec, label %exit, label %loop + +exit: + ret void +} + +attributes #0 = { "target-cpu"="neoverse-v2" }