; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "^scalar.ph:" --version 4 ; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -enable-epilogue-vectorization=false -S < %s | FileCheck %s --check-prefixes=CHECK-INTERLEAVE1 ; RUN: opt -passes=loop-vectorize -enable-epilogue-vectorization=false -S < %s | FileCheck %s --check-prefixes=CHECK-INTERLEAVED ; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -vectorizer-maximize-bandwidth -enable-epilogue-vectorization=false -S < %s | FileCheck %s --check-prefixes=CHECK-MAXBW target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128" target triple = "aarch64-none-unknown-elf" define i32 @dotp(ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @dotp( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 16 ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT1:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX1]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP16]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX1]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP20]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = mul [[TMP4]], [[TMP5]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP6]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT1]] = add nuw i64 [[INDEX1]], [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT1]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i32 @dotp( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 32 ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]] ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX1:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT1:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX1]] ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = shl nuw i64 [[TMP3]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP20]], i64 [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP20]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load , ptr [[TMP5]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX1]] ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = shl nuw i64 [[TMP7]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[TMP28]], i64 [[TMP8]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load , ptr [[TMP28]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[TMP9]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = zext [[WIDE_LOAD3]] to ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = mul [[TMP10]], [[TMP11]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP12]]) ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext [[WIDE_LOAD4]] to ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = zext [[WIDE_LOAD2]] to ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = mul [[TMP13]], [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE5]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI1]], [[TMP15]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT1]] = add nuw i64 [[INDEX1]], [[TMP1]] ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = icmp eq i64 [[INDEX_NEXT1]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP16]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]] ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i32 @dotp( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0:[0-9]+]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16 ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP14]], align 1 ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = mul [[TMP4]], [[TMP5]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP6]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP23]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-MAXBW: scalar.ph: ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i32 [ 0, %entry ], [ %add, %for.body ] %gep.a = getelementptr i8, ptr %a, i64 %iv %load.a = load i8, ptr %gep.a, align 1 %ext.a = zext i8 %load.a to i32 %gep.b = getelementptr i8, ptr %b, i64 %iv %load.b = load i8, ptr %gep.b, align 1 %ext.b = zext i8 %load.b to i32 %mul = mul i32 %ext.b, %ext.a %add = add i32 %mul, %accum %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body ret i32 %add } define i64 @not_dotp_i8_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %b) #1 { ; CHECK-INTERLEAVE1-LABEL: define i64 @not_dotp_i8_to_i64_has_neon_dotprod( ; CHECK-INTERLEAVE1-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1:[0-9]+]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64> ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64> ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = mul nuw nsw <16 x i64> [[TMP0]], [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP2]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVE1: for.exit: ; CHECK-INTERLEAVE1-NEXT: ret i64 [[TMP4]] ; ; CHECK-INTERLEAVED-LABEL: define i64 @not_dotp_i8_to_i64_has_neon_dotprod( ; CHECK-INTERLEAVED-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1:[0-9]+]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE6:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE13:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE14:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 32 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 48 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP0]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD7:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i64 32 ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[NEXT_GEP1]], i64 48 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load <16 x i8>, ptr [[TMP18]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD11:%.*]] = load <16 x i8>, ptr [[TMP19]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD4]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = mul nuw nsw <16 x i64> [[TMP2]], [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP4]]) ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <16 x i8> [[WIDE_LOAD5]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <16 x i64> [[TMP5]], [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE6]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI1]], <16 x i64> [[TMP7]]) ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = zext <16 x i8> [[WIDE_LOAD10]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext <16 x i8> [[WIDE_LOAD6]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul nuw nsw <16 x i64> [[TMP12]], [[TMP13]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE13]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI2]], <16 x i64> [[TMP14]]) ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext <16 x i8> [[WIDE_LOAD11]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = zext <16 x i8> [[WIDE_LOAD7]] to <16 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = mul nuw nsw <16 x i64> [[TMP15]], [[TMP16]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE14]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI3]], <16 x i64> [[TMP17]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64 ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[PARTIAL_REDUCE6]], [[PARTIAL_REDUCE]] ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX15:%.*]] = add <2 x i64> [[PARTIAL_REDUCE13]], [[BIN_RDX]] ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX16:%.*]] = add <2 x i64> [[PARTIAL_REDUCE14]], [[BIN_RDX15]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX16]]) ; CHECK-INTERLEAVED-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVED: for.exit: ; CHECK-INTERLEAVED-NEXT: ret i64 [[TMP9]] ; ; CHECK-MAXBW-LABEL: define i64 @not_dotp_i8_to_i64_has_neon_dotprod( ; CHECK-MAXBW-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1:[0-9]+]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[NEXT_GEP]], align 1 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[NEXT_GEP1]], align 1 ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i64> ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i64> ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw nsw <16 x i64> [[TMP0]], [[TMP1]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v16i64(<2 x i64> [[VEC_PHI]], <16 x i64> [[TMP2]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-MAXBW-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-MAXBW: for.exit: ; CHECK-MAXBW-NEXT: ret i64 [[TMP4]] ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i64 [ 0, %entry ], [ %add, %for.body ] %gep.a = phi ptr [ %a, %entry ], [ %gep.a.next, %for.body ] %gep.b = phi ptr [ %b, %entry ], [ %gep.b.next, %for.body ] %gep.a.next = getelementptr inbounds nuw i8, ptr %gep.a, i64 1 %load.a = load i8, ptr %gep.a, align 1 %ext.a = zext i8 %load.a to i64 %gep.b.next = getelementptr inbounds nuw i8, ptr %gep.b, i64 1 %load.b = load i8, ptr %gep.b, align 1 %ext.b = zext i8 %load.b to i64 %mul = mul nuw nsw i64 %ext.b, %ext.a %add = add nsw i64 %mul, %accum %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body ret i64 %add } define i64 @not_dotp_i16_to_i64_has_neon_dotprod(ptr readonly %a, ptr readonly %b) #1 { ; CHECK-INTERLEAVE1-LABEL: define i64 @not_dotp_i16_to_i64_has_neon_dotprod( ; CHECK-INTERLEAVE1-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2 ; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]] ; CHECK-INTERLEAVE1-NEXT: [[OFFSET_IDX1:%.*]] = mul i64 [[INDEX]], 2 ; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX1]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i16>, ptr [[NEXT_GEP2]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext <8 x i16> [[WIDE_LOAD3]] to <8 x i64> ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[TMP0]], [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP2]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVE1: for.exit: ; CHECK-INTERLEAVE1-NEXT: ret i64 [[TMP4]] ; ; CHECK-INTERLEAVED-LABEL: define i64 @not_dotp_i16_to_i64_has_neon_dotprod( ; CHECK-INTERLEAVED-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE7:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE14:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE15:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2 ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]] ; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX2:%.*]] = mul i64 [[INDEX]], 2 ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX2]] ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 8 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i16, ptr [[NEXT_GEP]], i64 24 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load <8 x i16>, ptr [[TMP0]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD7:%.*]] = load <8 x i16>, ptr [[TMP10]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD8:%.*]] = load <8 x i16>, ptr [[TMP11]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 8 ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = getelementptr i16, ptr [[NEXT_GEP3]], i64 24 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load <8 x i16>, ptr [[NEXT_GEP3]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD6:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD11:%.*]] = load <8 x i16>, ptr [[TMP18]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD12:%.*]] = load <8 x i16>, ptr [[TMP19]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <8 x i16> [[WIDE_LOAD5]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = mul nuw nsw <8 x i64> [[TMP2]], [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP4]]) ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <8 x i16> [[WIDE_LOAD6]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <8 x i16> [[WIDE_LOAD4]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <8 x i64> [[TMP5]], [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE7]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI1]], <8 x i64> [[TMP7]]) ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = zext <8 x i16> [[WIDE_LOAD11]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext <8 x i16> [[WIDE_LOAD7]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul nuw nsw <8 x i64> [[TMP12]], [[TMP13]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE14]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI2]], <8 x i64> [[TMP14]]) ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext <8 x i16> [[WIDE_LOAD12]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = zext <8 x i16> [[WIDE_LOAD8]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = mul nuw nsw <8 x i64> [[TMP15]], [[TMP16]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE15]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI3]], <8 x i64> [[TMP17]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32 ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <2 x i64> [[PARTIAL_REDUCE7]], [[PARTIAL_REDUCE]] ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX16:%.*]] = add <2 x i64> [[PARTIAL_REDUCE14]], [[BIN_RDX]] ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX17:%.*]] = add <2 x i64> [[PARTIAL_REDUCE15]], [[BIN_RDX16]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[BIN_RDX17]]) ; CHECK-INTERLEAVED-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVED: for.exit: ; CHECK-INTERLEAVED-NEXT: ret i64 [[TMP9]] ; ; CHECK-MAXBW-LABEL: define i64 @not_dotp_i16_to_i64_has_neon_dotprod( ; CHECK-MAXBW-SAME: ptr readonly [[A:%.*]], ptr readonly [[B:%.*]]) #[[ATTR1]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 2 ; CHECK-MAXBW-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]] ; CHECK-MAXBW-NEXT: [[OFFSET_IDX1:%.*]] = mul i64 [[INDEX]], 2 ; CHECK-MAXBW-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[OFFSET_IDX1]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[NEXT_GEP]], align 2 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD3:%.*]] = load <8 x i16>, ptr [[NEXT_GEP2]], align 2 ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = zext <8 x i16> [[WIDE_LOAD3]] to <8 x i64> ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[TMP0]], [[TMP1]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <2 x i64> @llvm.vector.partial.reduce.add.v2i64.v8i64(<2 x i64> [[VEC_PHI]], <8 x i64> [[TMP2]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-MAXBW-NEXT: br i1 [[TMP3]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-MAXBW: for.exit: ; CHECK-MAXBW-NEXT: ret i64 [[TMP4]] ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i64 [ 0, %entry ], [ %add, %for.body ] %gep.a = phi ptr [ %a, %entry ], [ %gep.a.next, %for.body ] %gep.b = phi ptr [ %b, %entry ], [ %gep.b.next, %for.body ] %gep.a.next = getelementptr inbounds nuw i16, ptr %gep.a, i64 1 %load.a = load i16, ptr %gep.a, align 2 %ext.a = zext i16 %load.a to i64 %gep.b.next = getelementptr inbounds nuw i16, ptr %gep.b, i64 1 %load.b = load i16, ptr %gep.b, align 2 %ext.b = zext i16 %load.b to i64 %mul = mul nuw nsw i64 %ext.b, %ext.a %add = add nsw i64 %mul, %accum %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body ret i64 %add } define i32 @not_dotp_different_types(ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @not_dotp_different_types( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP69:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3 ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 4 ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 5 ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 6 ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 7 ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 9 ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 10 ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], 11 ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 12 ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], 13 ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = add i64 [[INDEX]], 14 ; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 15 ; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP0]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP16]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP18:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]] ; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[TMP21:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP2]] ; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP4]] ; CHECK-INTERLEAVE1-NEXT: [[TMP24:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]] ; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP6]] ; CHECK-INTERLEAVE1-NEXT: [[TMP26:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP7]] ; CHECK-INTERLEAVE1-NEXT: [[TMP27:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP8]] ; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP9]] ; CHECK-INTERLEAVE1-NEXT: [[TMP29:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP10]] ; CHECK-INTERLEAVE1-NEXT: [[TMP30:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP11]] ; CHECK-INTERLEAVE1-NEXT: [[TMP31:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP12]] ; CHECK-INTERLEAVE1-NEXT: [[TMP32:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP13]] ; CHECK-INTERLEAVE1-NEXT: [[TMP33:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP14]] ; CHECK-INTERLEAVE1-NEXT: [[TMP34:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP15]] ; CHECK-INTERLEAVE1-NEXT: [[TMP35:%.*]] = load i16, ptr [[TMP19]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP36:%.*]] = load i16, ptr [[TMP20]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP37:%.*]] = load i16, ptr [[TMP21]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP38:%.*]] = load i16, ptr [[TMP22]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP41:%.*]] = load i16, ptr [[TMP23]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP42:%.*]] = load i16, ptr [[TMP24]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP43:%.*]] = load i16, ptr [[TMP25]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP44:%.*]] = load i16, ptr [[TMP26]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP45:%.*]] = load i16, ptr [[TMP27]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP46:%.*]] = load i16, ptr [[TMP28]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP55:%.*]] = load i16, ptr [[TMP29]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP56:%.*]] = load i16, ptr [[TMP30]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP47:%.*]] = load i16, ptr [[TMP31]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP48:%.*]] = load i16, ptr [[TMP32]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP49:%.*]] = load i16, ptr [[TMP33]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP50:%.*]] = load i16, ptr [[TMP34]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP51:%.*]] = insertelement <16 x i16> poison, i16 [[TMP35]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP52:%.*]] = insertelement <16 x i16> [[TMP51]], i16 [[TMP36]], i32 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP53:%.*]] = insertelement <16 x i16> [[TMP52]], i16 [[TMP37]], i32 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP54:%.*]] = insertelement <16 x i16> [[TMP53]], i16 [[TMP38]], i32 3 ; CHECK-INTERLEAVE1-NEXT: [[TMP57:%.*]] = insertelement <16 x i16> [[TMP54]], i16 [[TMP41]], i32 4 ; CHECK-INTERLEAVE1-NEXT: [[TMP58:%.*]] = insertelement <16 x i16> [[TMP57]], i16 [[TMP42]], i32 5 ; CHECK-INTERLEAVE1-NEXT: [[TMP59:%.*]] = insertelement <16 x i16> [[TMP58]], i16 [[TMP43]], i32 6 ; CHECK-INTERLEAVE1-NEXT: [[TMP60:%.*]] = insertelement <16 x i16> [[TMP59]], i16 [[TMP44]], i32 7 ; CHECK-INTERLEAVE1-NEXT: [[TMP61:%.*]] = insertelement <16 x i16> [[TMP60]], i16 [[TMP45]], i32 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP96:%.*]] = insertelement <16 x i16> [[TMP61]], i16 [[TMP46]], i32 9 ; CHECK-INTERLEAVE1-NEXT: [[TMP97:%.*]] = insertelement <16 x i16> [[TMP96]], i16 [[TMP55]], i32 10 ; CHECK-INTERLEAVE1-NEXT: [[TMP62:%.*]] = insertelement <16 x i16> [[TMP97]], i16 [[TMP56]], i32 11 ; CHECK-INTERLEAVE1-NEXT: [[TMP63:%.*]] = insertelement <16 x i16> [[TMP62]], i16 [[TMP47]], i32 12 ; CHECK-INTERLEAVE1-NEXT: [[TMP64:%.*]] = insertelement <16 x i16> [[TMP63]], i16 [[TMP48]], i32 13 ; CHECK-INTERLEAVE1-NEXT: [[TMP65:%.*]] = insertelement <16 x i16> [[TMP64]], i16 [[TMP49]], i32 14 ; CHECK-INTERLEAVE1-NEXT: [[TMP66:%.*]] = insertelement <16 x i16> [[TMP65]], i16 [[TMP50]], i32 15 ; CHECK-INTERLEAVE1-NEXT: [[TMP67:%.*]] = zext <16 x i16> [[TMP66]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP68:%.*]] = mul <16 x i32> [[TMP67]], [[TMP18]] ; CHECK-INTERLEAVE1-NEXT: [[TMP69]] = add <16 x i32> [[TMP68]], [[VEC_PHI]] ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVE1-NEXT: [[TMP70:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP70]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP71:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP69]]) ; CHECK-INTERLEAVE1-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVE1: for.exit: ; CHECK-INTERLEAVE1-NEXT: ret i32 [[TMP71]] ; ; CHECK-INTERLEAVED-LABEL: define i32 @not_dotp_different_types( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP137:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP138:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2 ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 5 ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 6 ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 7 ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 8 ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 9 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 10 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], 11 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 12 ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], 13 ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = add i64 [[INDEX]], 14 ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 15 ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = add i64 [[INDEX]], 16 ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = add i64 [[INDEX]], 17 ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = add i64 [[INDEX]], 18 ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = add i64 [[INDEX]], 19 ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = add i64 [[INDEX]], 20 ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = add i64 [[INDEX]], 21 ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = add i64 [[INDEX]], 22 ; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = add i64 [[INDEX]], 23 ; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = add i64 [[INDEX]], 24 ; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = add i64 [[INDEX]], 25 ; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = add i64 [[INDEX]], 26 ; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = add i64 [[INDEX]], 27 ; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = add i64 [[INDEX]], 28 ; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = add i64 [[INDEX]], 29 ; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = add i64 [[INDEX]], 30 ; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = add i64 [[INDEX]], 31 ; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP0]] ; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = getelementptr i8, ptr [[TMP32]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP32]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP34]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]] ; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]] ; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP2]] ; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[TMP44:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]] ; CHECK-INTERLEAVED-NEXT: [[TMP45:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[TMP46:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP7]] ; CHECK-INTERLEAVED-NEXT: [[TMP47:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP8]] ; CHECK-INTERLEAVED-NEXT: [[TMP48:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP9]] ; CHECK-INTERLEAVED-NEXT: [[TMP49:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP10]] ; CHECK-INTERLEAVED-NEXT: [[TMP50:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP11]] ; CHECK-INTERLEAVED-NEXT: [[TMP51:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP12]] ; CHECK-INTERLEAVED-NEXT: [[TMP52:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP13]] ; CHECK-INTERLEAVED-NEXT: [[TMP53:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[TMP54:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP15]] ; CHECK-INTERLEAVED-NEXT: [[TMP55:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP16]] ; CHECK-INTERLEAVED-NEXT: [[TMP56:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP17]] ; CHECK-INTERLEAVED-NEXT: [[TMP57:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP18]] ; CHECK-INTERLEAVED-NEXT: [[TMP58:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP19]] ; CHECK-INTERLEAVED-NEXT: [[TMP59:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP20]] ; CHECK-INTERLEAVED-NEXT: [[TMP60:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP21]] ; CHECK-INTERLEAVED-NEXT: [[TMP61:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP22]] ; CHECK-INTERLEAVED-NEXT: [[TMP62:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP23]] ; CHECK-INTERLEAVED-NEXT: [[TMP63:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP24]] ; CHECK-INTERLEAVED-NEXT: [[TMP64:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP25]] ; CHECK-INTERLEAVED-NEXT: [[TMP65:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP26]] ; CHECK-INTERLEAVED-NEXT: [[TMP66:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP27]] ; CHECK-INTERLEAVED-NEXT: [[TMP67:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP28]] ; CHECK-INTERLEAVED-NEXT: [[TMP68:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP29]] ; CHECK-INTERLEAVED-NEXT: [[TMP139:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP30]] ; CHECK-INTERLEAVED-NEXT: [[TMP140:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP31]] ; CHECK-INTERLEAVED-NEXT: [[TMP69:%.*]] = load i16, ptr [[TMP39]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP70:%.*]] = load i16, ptr [[TMP40]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP71:%.*]] = load i16, ptr [[TMP41]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP72:%.*]] = load i16, ptr [[TMP42]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP73:%.*]] = load i16, ptr [[TMP43]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP74:%.*]] = load i16, ptr [[TMP44]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP75:%.*]] = load i16, ptr [[TMP45]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP76:%.*]] = load i16, ptr [[TMP46]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP77:%.*]] = load i16, ptr [[TMP47]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP78:%.*]] = load i16, ptr [[TMP48]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP79:%.*]] = load i16, ptr [[TMP49]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP80:%.*]] = load i16, ptr [[TMP50]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP81:%.*]] = load i16, ptr [[TMP51]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP82:%.*]] = load i16, ptr [[TMP52]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP83:%.*]] = load i16, ptr [[TMP53]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP84:%.*]] = load i16, ptr [[TMP54]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP85:%.*]] = insertelement <16 x i16> poison, i16 [[TMP69]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP86:%.*]] = insertelement <16 x i16> [[TMP85]], i16 [[TMP70]], i32 1 ; CHECK-INTERLEAVED-NEXT: [[TMP87:%.*]] = insertelement <16 x i16> [[TMP86]], i16 [[TMP71]], i32 2 ; CHECK-INTERLEAVED-NEXT: [[TMP88:%.*]] = insertelement <16 x i16> [[TMP87]], i16 [[TMP72]], i32 3 ; CHECK-INTERLEAVED-NEXT: [[TMP89:%.*]] = insertelement <16 x i16> [[TMP88]], i16 [[TMP73]], i32 4 ; CHECK-INTERLEAVED-NEXT: [[TMP90:%.*]] = insertelement <16 x i16> [[TMP89]], i16 [[TMP74]], i32 5 ; CHECK-INTERLEAVED-NEXT: [[TMP91:%.*]] = insertelement <16 x i16> [[TMP90]], i16 [[TMP75]], i32 6 ; CHECK-INTERLEAVED-NEXT: [[TMP92:%.*]] = insertelement <16 x i16> [[TMP91]], i16 [[TMP76]], i32 7 ; CHECK-INTERLEAVED-NEXT: [[TMP93:%.*]] = insertelement <16 x i16> [[TMP92]], i16 [[TMP77]], i32 8 ; CHECK-INTERLEAVED-NEXT: [[TMP94:%.*]] = insertelement <16 x i16> [[TMP93]], i16 [[TMP78]], i32 9 ; CHECK-INTERLEAVED-NEXT: [[TMP95:%.*]] = insertelement <16 x i16> [[TMP94]], i16 [[TMP79]], i32 10 ; CHECK-INTERLEAVED-NEXT: [[TMP96:%.*]] = insertelement <16 x i16> [[TMP95]], i16 [[TMP80]], i32 11 ; CHECK-INTERLEAVED-NEXT: [[TMP97:%.*]] = insertelement <16 x i16> [[TMP96]], i16 [[TMP81]], i32 12 ; CHECK-INTERLEAVED-NEXT: [[TMP98:%.*]] = insertelement <16 x i16> [[TMP97]], i16 [[TMP82]], i32 13 ; CHECK-INTERLEAVED-NEXT: [[TMP99:%.*]] = insertelement <16 x i16> [[TMP98]], i16 [[TMP83]], i32 14 ; CHECK-INTERLEAVED-NEXT: [[TMP100:%.*]] = insertelement <16 x i16> [[TMP99]], i16 [[TMP84]], i32 15 ; CHECK-INTERLEAVED-NEXT: [[TMP101:%.*]] = load i16, ptr [[TMP55]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP102:%.*]] = load i16, ptr [[TMP56]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP103:%.*]] = load i16, ptr [[TMP57]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP104:%.*]] = load i16, ptr [[TMP58]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP105:%.*]] = load i16, ptr [[TMP59]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP106:%.*]] = load i16, ptr [[TMP60]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP107:%.*]] = load i16, ptr [[TMP61]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP108:%.*]] = load i16, ptr [[TMP62]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP109:%.*]] = load i16, ptr [[TMP63]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP110:%.*]] = load i16, ptr [[TMP64]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP111:%.*]] = load i16, ptr [[TMP65]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP112:%.*]] = load i16, ptr [[TMP66]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP113:%.*]] = load i16, ptr [[TMP67]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP114:%.*]] = load i16, ptr [[TMP68]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP115:%.*]] = load i16, ptr [[TMP139]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP116:%.*]] = load i16, ptr [[TMP140]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP117:%.*]] = insertelement <16 x i16> poison, i16 [[TMP101]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP118:%.*]] = insertelement <16 x i16> [[TMP117]], i16 [[TMP102]], i32 1 ; CHECK-INTERLEAVED-NEXT: [[TMP119:%.*]] = insertelement <16 x i16> [[TMP118]], i16 [[TMP103]], i32 2 ; CHECK-INTERLEAVED-NEXT: [[TMP120:%.*]] = insertelement <16 x i16> [[TMP119]], i16 [[TMP104]], i32 3 ; CHECK-INTERLEAVED-NEXT: [[TMP121:%.*]] = insertelement <16 x i16> [[TMP120]], i16 [[TMP105]], i32 4 ; CHECK-INTERLEAVED-NEXT: [[TMP122:%.*]] = insertelement <16 x i16> [[TMP121]], i16 [[TMP106]], i32 5 ; CHECK-INTERLEAVED-NEXT: [[TMP123:%.*]] = insertelement <16 x i16> [[TMP122]], i16 [[TMP107]], i32 6 ; CHECK-INTERLEAVED-NEXT: [[TMP124:%.*]] = insertelement <16 x i16> [[TMP123]], i16 [[TMP108]], i32 7 ; CHECK-INTERLEAVED-NEXT: [[TMP125:%.*]] = insertelement <16 x i16> [[TMP124]], i16 [[TMP109]], i32 8 ; CHECK-INTERLEAVED-NEXT: [[TMP126:%.*]] = insertelement <16 x i16> [[TMP125]], i16 [[TMP110]], i32 9 ; CHECK-INTERLEAVED-NEXT: [[TMP127:%.*]] = insertelement <16 x i16> [[TMP126]], i16 [[TMP111]], i32 10 ; CHECK-INTERLEAVED-NEXT: [[TMP128:%.*]] = insertelement <16 x i16> [[TMP127]], i16 [[TMP112]], i32 11 ; CHECK-INTERLEAVED-NEXT: [[TMP129:%.*]] = insertelement <16 x i16> [[TMP128]], i16 [[TMP113]], i32 12 ; CHECK-INTERLEAVED-NEXT: [[TMP130:%.*]] = insertelement <16 x i16> [[TMP129]], i16 [[TMP114]], i32 13 ; CHECK-INTERLEAVED-NEXT: [[TMP131:%.*]] = insertelement <16 x i16> [[TMP130]], i16 [[TMP115]], i32 14 ; CHECK-INTERLEAVED-NEXT: [[TMP132:%.*]] = insertelement <16 x i16> [[TMP131]], i16 [[TMP116]], i32 15 ; CHECK-INTERLEAVED-NEXT: [[TMP133:%.*]] = zext <16 x i16> [[TMP100]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP134:%.*]] = zext <16 x i16> [[TMP132]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP135:%.*]] = mul <16 x i32> [[TMP133]], [[TMP35]] ; CHECK-INTERLEAVED-NEXT: [[TMP136:%.*]] = mul <16 x i32> [[TMP134]], [[TMP36]] ; CHECK-INTERLEAVED-NEXT: [[TMP137]] = add <16 x i32> [[TMP135]], [[VEC_PHI]] ; CHECK-INTERLEAVED-NEXT: [[TMP138]] = add <16 x i32> [[TMP136]], [[VEC_PHI1]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32 ; CHECK-INTERLEAVED-NEXT: [[TMP141:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP141]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <16 x i32> [[TMP138]], [[TMP137]] ; CHECK-INTERLEAVED-NEXT: [[TMP142:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVED: for.exit: ; CHECK-INTERLEAVED-NEXT: ret i32 [[TMP142]] ; ; CHECK-MAXBW-LABEL: define i32 @not_dotp_different_types( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI1:%.*]] = phi <16 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP138:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 1 ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 2 ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 3 ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 4 ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = add i64 [[INDEX]], 5 ; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = add i64 [[INDEX]], 6 ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = add i64 [[INDEX]], 7 ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = add i64 [[INDEX]], 8 ; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = add i64 [[INDEX]], 9 ; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = add i64 [[INDEX]], 10 ; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = add i64 [[INDEX]], 11 ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = add i64 [[INDEX]], 12 ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = add i64 [[INDEX]], 13 ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = add i64 [[INDEX]], 14 ; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = add i64 [[INDEX]], 15 ; CHECK-MAXBW-NEXT: [[TMP32:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP0]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP32]], align 1 ; CHECK-MAXBW-NEXT: [[TMP36:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP37:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP0]] ; CHECK-MAXBW-NEXT: [[TMP38:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP1]] ; CHECK-MAXBW-NEXT: [[TMP39:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP2]] ; CHECK-MAXBW-NEXT: [[TMP40:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP3]] ; CHECK-MAXBW-NEXT: [[TMP41:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP4]] ; CHECK-MAXBW-NEXT: [[TMP42:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP5]] ; CHECK-MAXBW-NEXT: [[TMP43:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP6]] ; CHECK-MAXBW-NEXT: [[TMP44:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP7]] ; CHECK-MAXBW-NEXT: [[TMP45:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP8]] ; CHECK-MAXBW-NEXT: [[TMP46:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP9]] ; CHECK-MAXBW-NEXT: [[TMP47:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP10]] ; CHECK-MAXBW-NEXT: [[TMP48:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP11]] ; CHECK-MAXBW-NEXT: [[TMP49:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP12]] ; CHECK-MAXBW-NEXT: [[TMP50:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP13]] ; CHECK-MAXBW-NEXT: [[TMP51:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP14]] ; CHECK-MAXBW-NEXT: [[TMP52:%.*]] = getelementptr i8, ptr [[B]], i64 [[TMP15]] ; CHECK-MAXBW-NEXT: [[TMP101:%.*]] = load i16, ptr [[TMP37]], align 2 ; CHECK-MAXBW-NEXT: [[TMP102:%.*]] = load i16, ptr [[TMP38]], align 2 ; CHECK-MAXBW-NEXT: [[TMP103:%.*]] = load i16, ptr [[TMP39]], align 2 ; CHECK-MAXBW-NEXT: [[TMP104:%.*]] = load i16, ptr [[TMP40]], align 2 ; CHECK-MAXBW-NEXT: [[TMP105:%.*]] = load i16, ptr [[TMP41]], align 2 ; CHECK-MAXBW-NEXT: [[TMP106:%.*]] = load i16, ptr [[TMP42]], align 2 ; CHECK-MAXBW-NEXT: [[TMP107:%.*]] = load i16, ptr [[TMP43]], align 2 ; CHECK-MAXBW-NEXT: [[TMP108:%.*]] = load i16, ptr [[TMP44]], align 2 ; CHECK-MAXBW-NEXT: [[TMP109:%.*]] = load i16, ptr [[TMP45]], align 2 ; CHECK-MAXBW-NEXT: [[TMP110:%.*]] = load i16, ptr [[TMP46]], align 2 ; CHECK-MAXBW-NEXT: [[TMP111:%.*]] = load i16, ptr [[TMP47]], align 2 ; CHECK-MAXBW-NEXT: [[TMP112:%.*]] = load i16, ptr [[TMP48]], align 2 ; CHECK-MAXBW-NEXT: [[TMP113:%.*]] = load i16, ptr [[TMP49]], align 2 ; CHECK-MAXBW-NEXT: [[TMP114:%.*]] = load i16, ptr [[TMP50]], align 2 ; CHECK-MAXBW-NEXT: [[TMP115:%.*]] = load i16, ptr [[TMP51]], align 2 ; CHECK-MAXBW-NEXT: [[TMP116:%.*]] = load i16, ptr [[TMP52]], align 2 ; CHECK-MAXBW-NEXT: [[TMP117:%.*]] = insertelement <16 x i16> poison, i16 [[TMP101]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP118:%.*]] = insertelement <16 x i16> [[TMP117]], i16 [[TMP102]], i32 1 ; CHECK-MAXBW-NEXT: [[TMP119:%.*]] = insertelement <16 x i16> [[TMP118]], i16 [[TMP103]], i32 2 ; CHECK-MAXBW-NEXT: [[TMP120:%.*]] = insertelement <16 x i16> [[TMP119]], i16 [[TMP104]], i32 3 ; CHECK-MAXBW-NEXT: [[TMP121:%.*]] = insertelement <16 x i16> [[TMP120]], i16 [[TMP105]], i32 4 ; CHECK-MAXBW-NEXT: [[TMP122:%.*]] = insertelement <16 x i16> [[TMP121]], i16 [[TMP106]], i32 5 ; CHECK-MAXBW-NEXT: [[TMP123:%.*]] = insertelement <16 x i16> [[TMP122]], i16 [[TMP107]], i32 6 ; CHECK-MAXBW-NEXT: [[TMP124:%.*]] = insertelement <16 x i16> [[TMP123]], i16 [[TMP108]], i32 7 ; CHECK-MAXBW-NEXT: [[TMP125:%.*]] = insertelement <16 x i16> [[TMP124]], i16 [[TMP109]], i32 8 ; CHECK-MAXBW-NEXT: [[TMP126:%.*]] = insertelement <16 x i16> [[TMP125]], i16 [[TMP110]], i32 9 ; CHECK-MAXBW-NEXT: [[TMP127:%.*]] = insertelement <16 x i16> [[TMP126]], i16 [[TMP111]], i32 10 ; CHECK-MAXBW-NEXT: [[TMP128:%.*]] = insertelement <16 x i16> [[TMP127]], i16 [[TMP112]], i32 11 ; CHECK-MAXBW-NEXT: [[TMP129:%.*]] = insertelement <16 x i16> [[TMP128]], i16 [[TMP113]], i32 12 ; CHECK-MAXBW-NEXT: [[TMP130:%.*]] = insertelement <16 x i16> [[TMP129]], i16 [[TMP114]], i32 13 ; CHECK-MAXBW-NEXT: [[TMP131:%.*]] = insertelement <16 x i16> [[TMP130]], i16 [[TMP115]], i32 14 ; CHECK-MAXBW-NEXT: [[TMP132:%.*]] = insertelement <16 x i16> [[TMP131]], i16 [[TMP116]], i32 15 ; CHECK-MAXBW-NEXT: [[TMP134:%.*]] = zext <16 x i16> [[TMP132]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP136:%.*]] = mul <16 x i32> [[TMP134]], [[TMP36]] ; CHECK-MAXBW-NEXT: [[TMP138]] = add <16 x i32> [[TMP136]], [[VEC_PHI1]] ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-MAXBW-NEXT: [[TMP70:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-MAXBW-NEXT: br i1 [[TMP70]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP71:%.*]] = call i32 @llvm.vector.reduce.add.v16i32(<16 x i32> [[TMP138]]) ; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-MAXBW: for.exit: ; CHECK-MAXBW-NEXT: ret i32 [[TMP71]] ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i32 [ 0, %entry ], [ %add, %for.body ] %gep.a = getelementptr i8, ptr %a, i64 %iv %load.a = load i8, ptr %gep.a, align 1 %ext.a = zext i8 %load.a to i32 %gep.b = getelementptr i8, ptr %b, i64 %iv %load.b = load i16, ptr %gep.b, align 2 %ext.b = zext i16 %load.b to i32 %mul = mul i32 %ext.b, %ext.a %add = add i32 %mul, %accum %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body ret i32 %add } define i32 @not_dotp_not_loop_carried(ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @not_dotp_not_loop_carried( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VECTOR_RECUR:%.*]] = phi <16 x i32> [ , [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP13]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP4]] = mul <16 x i32> [[TMP3]], [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = shufflevector <16 x i32> [[VECTOR_RECUR]], <16 x i32> [[TMP4]], <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = add <16 x i32> [[TMP4]], [[TMP5]] ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = extractelement <16 x i32> [[TMP6]], i32 15 ; CHECK-INTERLEAVE1-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVE1: for.exit: ; CHECK-INTERLEAVE1-NEXT: ret i32 [[TMP8]] ; ; CHECK-INTERLEAVED-LABEL: define i32 @not_dotp_not_loop_carried( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr i8, ptr [[TMP10]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP1]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP17]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP17]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP5]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext <16 x i8> [[WIDE_LOAD2]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = zext <16 x i8> [[WIDE_LOAD3]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = mul <16 x i32> [[TMP6]], [[TMP2]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = mul <16 x i32> [[TMP7]], [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = shufflevector <16 x i32> [[TMP8]], <16 x i32> [[TMP9]], <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = add <16 x i32> [[TMP9]], [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = extractelement <16 x i32> [[TMP11]], i32 15 ; CHECK-INTERLEAVED-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-INTERLEAVED: for.exit: ; CHECK-INTERLEAVED-NEXT: ret i32 [[TMP13]] ; ; CHECK-MAXBW-LABEL: define i32 @not_dotp_not_loop_carried( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VECTOR_RECUR:%.*]] = phi <16 x i32> [ , [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1 ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load <16 x i8>, ptr [[TMP17]], align 1 ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = zext <16 x i8> [[WIDE_LOAD1]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP4]] = mul <16 x i32> [[TMP3]], [[TMP1]] ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = shufflevector <16 x i32> [[VECTOR_RECUR]], <16 x i32> [[TMP4]], <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = add <16 x i32> [[TMP4]], [[TMP5]] ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1024 ; CHECK-MAXBW-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = extractelement <16 x i32> [[TMP6]], i32 15 ; CHECK-MAXBW-NEXT: br label [[FOR_EXIT:%.*]] ; CHECK-MAXBW: for.exit: ; CHECK-MAXBW-NEXT: ret i32 [[TMP8]] ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i32 [ 0, %entry ], [ %mul, %for.body ] %gep.a = getelementptr i8, ptr %a, i64 %iv %load.a = load i8, ptr %gep.a, align 1 %ext.a = zext i8 %load.a to i32 %gep.b = getelementptr i8, ptr %b, i64 %iv %load.b = load i8, ptr %gep.b, align 1 %ext.b = zext i8 %load.b to i32 %mul = mul i32 %ext.b, %ext.a %add = add i32 %mul, %accum %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body ret i32 %add } define i32 @not_dotp_not_phi(ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @not_dotp_not_phi( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8 ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP10]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP13]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = mul [[TMP15]], [[TMP12]] ; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = add [[TMP16]], [[TMP15]] ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = call i32 @llvm.vscale.i32() ; CHECK-INTERLEAVE1-NEXT: [[TMP24:%.*]] = mul nuw i32 [[TMP23]], 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = sub i32 [[TMP24]], 1 ; CHECK-INTERLEAVE1-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement [[TMP17]], i32 [[TMP25]] ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i32 @not_dotp_not_phi( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16 ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = shl nuw i64 [[TMP12]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[TMP10]], i64 [[TMP13]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP14]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = shl nuw i64 [[TMP18]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = getelementptr i8, ptr [[TMP16]], i64 [[TMP19]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP20]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = mul [[TMP22]], [[TMP15]] ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = add [[TMP30]], [[TMP22]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP7:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = call i32 @llvm.vscale.i32() ; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = mul nuw i32 [[TMP27]], 8 ; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = sub i32 [[TMP28]], 1 ; CHECK-INTERLEAVED-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement [[TMP21]], i32 [[TMP29]] ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i32 @not_dotp_not_phi( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8 ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP10]], align 1 ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP15]], align 1 ; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = mul [[TMP19]], [[TMP14]] ; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = add [[TMP20]], [[TMP19]] ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP22]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = call i32 @llvm.vscale.i32() ; CHECK-MAXBW-NEXT: [[TMP24:%.*]] = mul nuw i32 [[TMP23]], 8 ; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = sub i32 [[TMP24]], 1 ; CHECK-MAXBW-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement [[TMP21]], i32 [[TMP25]] ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-MAXBW: scalar.ph: ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i32 [ 0, %entry ], [ %add, %for.body ] %gep.a = getelementptr i8, ptr %a, i64 %iv %load.a = load i8, ptr %gep.a, align 1 %ext.a = zext i8 %load.a to i32 %gep.b = getelementptr i8, ptr %b, i64 %iv %load.b = load i8, ptr %gep.b, align 1 %ext.b = zext i8 %load.b to i32 %mul = mul i32 %ext.b, %ext.a %add = add i32 %mul, %ext.b %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body ret i32 %add } define i32 @dotp_unrolled(i32 %num_out, i64 %num_in, ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @dotp_unrolled( ; CHECK-INTERLEAVE1-SAME: i32 [[NUM_OUT:%.*]], i64 [[NUM_IN:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = shl nuw nsw i64 [[TMP0]], 4 ; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[NUM_IN]], [[TMP12]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = mul nuw i64 [[TMP13]], 16 ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[NUM_IN]], [[TMP14]] ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 [[NUM_IN]], [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE13:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE10:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI2:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE7:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI3:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = or disjoint i64 [[INDEX]], 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP6]] ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP6]] ; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 3 ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]] ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP9]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP1]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[TMP2]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = sext [[WIDE_LOAD4]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = sext [[WIDE_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = mul nsw [[TMP15]], [[TMP16]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI3]], [[TMP17]]) ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD5:%.*]] = load , ptr [[TMP4]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD6:%.*]] = load , ptr [[TMP5]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP18:%.*]] = sext [[WIDE_LOAD5]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = sext [[WIDE_LOAD6]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = mul nsw [[TMP18]], [[TMP19]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE7]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI2]], [[TMP20]]) ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD8:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD9:%.*]] = load , ptr [[TMP8]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP21:%.*]] = sext [[WIDE_LOAD8]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = sext [[WIDE_LOAD9]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = mul nsw [[TMP21]], [[TMP22]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE10]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI1]], [[TMP23]]) ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD11:%.*]] = load , ptr [[TMP10]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD12:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP24:%.*]] = sext [[WIDE_LOAD11]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = sext [[WIDE_LOAD12]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP26:%.*]] = mul nsw [[TMP24]], [[TMP25]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE13]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP26]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP14]] ; CHECK-INTERLEAVE1-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE13]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP29:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE10]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP30:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE7]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[NUM_IN]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i32 @dotp_unrolled( ; CHECK-INTERLEAVED-SAME: i32 [[NUM_OUT:%.*]], i64 [[NUM_IN:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = shl nuw nsw i64 [[TMP0]], 5 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[NUM_IN]], [[TMP12]] ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul nuw i64 [[TMP13]], 32 ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[NUM_IN]], [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[NUM_IN]], [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE28:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE29:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE22:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE23:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI4:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE16:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI5:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE17:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI6:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI7:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE11:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = or disjoint i64 [[INDEX]], 1 ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = or disjoint i64 [[INDEX]], 2 ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]] ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP9]] ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = shl nuw i64 [[TMP15]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[TMP1]], i64 [[TMP16]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP1]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD8:%.*]] = load , ptr [[TMP17]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = shl nuw i64 [[TMP18]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = getelementptr inbounds i8, ptr [[TMP2]], i64 [[TMP19]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD9:%.*]] = load , ptr [[TMP2]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD10:%.*]] = load , ptr [[TMP20]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = sext [[WIDE_LOAD9]] to ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = sext [[WIDE_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = mul nsw [[TMP21]], [[TMP22]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI6]], [[TMP23]]) ; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = sext [[WIDE_LOAD10]] to ; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = sext [[WIDE_LOAD8]] to ; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = mul nsw [[TMP24]], [[TMP25]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE11]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI7]], [[TMP26]]) ; CHECK-INTERLEAVED-NEXT: [[TMP27:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = shl nuw i64 [[TMP27]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i64 [[TMP28]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD12:%.*]] = load , ptr [[TMP4]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD13:%.*]] = load , ptr [[TMP29]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP30:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = shl nuw i64 [[TMP30]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP63:%.*]] = getelementptr inbounds i8, ptr [[TMP5]], i64 [[TMP31]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD14:%.*]] = load , ptr [[TMP5]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD15:%.*]] = load , ptr [[TMP63]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP33:%.*]] = sext [[WIDE_LOAD12]] to ; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = sext [[WIDE_LOAD14]] to ; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = mul nsw [[TMP33]], [[TMP34]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI4]], [[TMP35]]) ; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = sext [[WIDE_LOAD13]] to ; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = sext [[WIDE_LOAD15]] to ; CHECK-INTERLEAVED-NEXT: [[TMP38:%.*]] = mul nsw [[TMP36]], [[TMP37]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI5]], [[TMP38]]) ; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = shl nuw i64 [[TMP39]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = getelementptr inbounds i8, ptr [[TMP7]], i64 [[TMP40]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD18:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD19:%.*]] = load , ptr [[TMP41]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = shl nuw i64 [[TMP42]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP44:%.*]] = getelementptr inbounds i8, ptr [[TMP8]], i64 [[TMP43]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD20:%.*]] = load , ptr [[TMP8]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD21:%.*]] = load , ptr [[TMP44]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP45:%.*]] = sext [[WIDE_LOAD18]] to ; CHECK-INTERLEAVED-NEXT: [[TMP46:%.*]] = sext [[WIDE_LOAD20]] to ; CHECK-INTERLEAVED-NEXT: [[TMP47:%.*]] = mul nsw [[TMP45]], [[TMP46]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE22]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI2]], [[TMP47]]) ; CHECK-INTERLEAVED-NEXT: [[TMP48:%.*]] = sext [[WIDE_LOAD19]] to ; CHECK-INTERLEAVED-NEXT: [[TMP49:%.*]] = sext [[WIDE_LOAD21]] to ; CHECK-INTERLEAVED-NEXT: [[TMP50:%.*]] = mul nsw [[TMP48]], [[TMP49]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE23]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI3]], [[TMP50]]) ; CHECK-INTERLEAVED-NEXT: [[TMP51:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP52:%.*]] = shl nuw i64 [[TMP51]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP53:%.*]] = getelementptr inbounds i8, ptr [[TMP10]], i64 [[TMP52]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD24:%.*]] = load , ptr [[TMP10]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD25:%.*]] = load , ptr [[TMP53]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP54:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP55:%.*]] = shl nuw i64 [[TMP54]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP56:%.*]] = getelementptr inbounds i8, ptr [[TMP11]], i64 [[TMP55]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD26:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD27:%.*]] = load , ptr [[TMP56]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP57:%.*]] = sext [[WIDE_LOAD24]] to ; CHECK-INTERLEAVED-NEXT: [[TMP58:%.*]] = sext [[WIDE_LOAD26]] to ; CHECK-INTERLEAVED-NEXT: [[TMP59:%.*]] = mul nsw [[TMP57]], [[TMP58]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE28]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP59]]) ; CHECK-INTERLEAVED-NEXT: [[TMP60:%.*]] = sext [[WIDE_LOAD25]] to ; CHECK-INTERLEAVED-NEXT: [[TMP61:%.*]] = sext [[WIDE_LOAD27]] to ; CHECK-INTERLEAVED-NEXT: [[TMP62:%.*]] = mul nsw [[TMP60]], [[TMP61]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE29]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI1]], [[TMP62]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add [[PARTIAL_REDUCE29]], [[PARTIAL_REDUCE28]] ; CHECK-INTERLEAVED-NEXT: [[TMP64:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX30:%.*]] = add [[PARTIAL_REDUCE23]], [[PARTIAL_REDUCE22]] ; CHECK-INTERLEAVED-NEXT: [[TMP65:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[BIN_RDX30]]) ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX31:%.*]] = add [[PARTIAL_REDUCE17]], [[PARTIAL_REDUCE16]] ; CHECK-INTERLEAVED-NEXT: [[TMP66:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[BIN_RDX31]]) ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX32:%.*]] = add [[PARTIAL_REDUCE11]], [[PARTIAL_REDUCE]] ; CHECK-INTERLEAVED-NEXT: [[TMP67:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[BIN_RDX32]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[NUM_IN]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i32 @dotp_unrolled( ; CHECK-MAXBW-SAME: i32 [[NUM_OUT:%.*]], i64 [[NUM_IN:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 4 ; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[NUM_IN]], [[TMP1]] ; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16 ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[NUM_IN]], [[TMP3]] ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[NUM_IN]], [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE13:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE10:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI2:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE7:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI3:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = or disjoint i64 [[INDEX]], 1 ; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP9]] ; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP9]] ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = or disjoint i64 [[INDEX]], 2 ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP12]] ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP12]] ; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = or disjoint i64 [[INDEX]], 3 ; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP15]] ; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP15]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[TMP8]], align 1 ; CHECK-MAXBW-NEXT: [[TMP27:%.*]] = sext [[WIDE_LOAD4]] to ; CHECK-MAXBW-NEXT: [[TMP32:%.*]] = sext [[WIDE_LOAD]] to ; CHECK-MAXBW-NEXT: [[TMP33:%.*]] = mul nsw [[TMP27]], [[TMP32]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI3]], [[TMP33]]) ; CHECK-MAXBW-NEXT: [[WIDE_LOAD5:%.*]] = load , ptr [[TMP10]], align 1 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD6:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = sext [[WIDE_LOAD5]] to ; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = sext [[WIDE_LOAD6]] to ; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = mul nsw [[TMP18]], [[TMP19]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE7]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI2]], [[TMP20]]) ; CHECK-MAXBW-NEXT: [[WIDE_LOAD8:%.*]] = load , ptr [[TMP13]], align 1 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD9:%.*]] = load , ptr [[TMP14]], align 1 ; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = sext [[WIDE_LOAD8]] to ; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = sext [[WIDE_LOAD9]] to ; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = mul nsw [[TMP21]], [[TMP22]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE10]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI1]], [[TMP23]]) ; CHECK-MAXBW-NEXT: [[WIDE_LOAD11:%.*]] = load , ptr [[TMP16]], align 1 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD12:%.*]] = load , ptr [[TMP17]], align 1 ; CHECK-MAXBW-NEXT: [[TMP24:%.*]] = sext [[WIDE_LOAD11]] to ; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = sext [[WIDE_LOAD12]] to ; CHECK-MAXBW-NEXT: [[TMP26:%.*]] = mul nsw [[TMP24]], [[TMP25]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE13]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP26]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-MAXBW-NEXT: [[TMP74:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP74]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP10:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE13]]) ; CHECK-MAXBW-NEXT: [[TMP29:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE10]]) ; CHECK-MAXBW-NEXT: [[TMP30:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE7]]) ; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[NUM_IN]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-MAXBW: scalar.ph: ; entry: br label %for.body for.body: ; preds = %entry, %for.body %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum3 = phi i32 [ 0, %entry ], [ %add.a3, %for.body ] %accum2 = phi i32 [ 0, %entry ], [ %add.a2, %for.body ] %accum1 = phi i32 [ 0, %entry ], [ %add.a1, %for.body ] %accum0 = phi i32 [ 0, %entry ], [ %add.a0, %for.body ] %gep.a0 = getelementptr inbounds i8, ptr %a, i64 %iv %gep.b0 = getelementptr inbounds i8, ptr %b, i64 %iv %offset.1 = or disjoint i64 %iv, 1 %gep.a1 = getelementptr inbounds i8, ptr %a, i64 %offset.1 %gep.b1 = getelementptr inbounds i8, ptr %b, i64 %offset.1 %offset.2 = or disjoint i64 %iv, 2 %gep.a2 = getelementptr inbounds i8, ptr %a, i64 %offset.2 %gep.b2 = getelementptr inbounds i8, ptr %b, i64 %offset.2 %offset.3 = or disjoint i64 %iv, 3 %gep.a3 = getelementptr inbounds i8, ptr %a, i64 %offset.3 %gep.b3 = getelementptr inbounds i8, ptr %b, i64 %offset.3 %load.a0 = load i8, ptr %gep.a0, align 1 %ext.a0 = sext i8 %load.a0 to i32 %load.b0 = load i8, ptr %gep.b0, align 1 %ext.b0 = sext i8 %load.b0 to i32 %mul.a0 = mul nsw i32 %ext.b0, %ext.a0 %add.a0 = add nsw i32 %mul.a0, %accum0 %load.a1 = load i8, ptr %gep.a1, align 1 %ext.a1 = sext i8 %load.a1 to i32 %load.b1 = load i8, ptr %gep.b1, align 1 %ext.b1 = sext i8 %load.b1 to i32 %mul.a1 = mul nsw i32 %ext.a1, %ext.b1 %add.a1 = add nsw i32 %mul.a1, %accum1 %load.a2 = load i8, ptr %gep.a2, align 1 %ext.a2 = sext i8 %load.a2 to i32 %load.b2 = load i8, ptr %gep.b2, align 1 %ext.b2 = sext i8 %load.b2 to i32 %mul.a2 = mul nsw i32 %ext.a2, %ext.b2 %add.a2 = add nsw i32 %mul.a2, %accum2 %load.a3 = load i8, ptr %gep.a3, align 1 %ext.a3 = sext i8 %load.a3 to i32 %load.b3 = load i8, ptr %gep.b3, align 1 %ext.b3 = sext i8 %load.b3 to i32 %mul.a3 = mul nsw i32 %ext.a3, %ext.b3 %add.a3 = add nsw i32 %mul.a3, %accum3 %iv.next = add nuw nsw i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, %num_in br i1 %exitcond.not, label %exit, label %for.body exit: ; preds = %for.body %result0 = add nsw i32 %add.a0, %add.a1 %result1 = add nsw i32 %add.a2, %add.a3 %result = add nsw i32 %result0, %result1 ret i32 %result } define i32 @dotp_predicated(i64 %N, ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @dotp_predicated( ; CHECK-INTERLEAVE1-SAME: i64 [[N:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 16 ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 4 ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = sub i64 [[N]], [[TMP6]] ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = icmp ugt i64 [[N]], [[TMP6]] ; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i64 [[TMP7]], i64 0 ; CHECK-INTERLEAVE1-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call @llvm.get.active.lane.mask.nxv16i1.i64(i64 0, i64 [[N]]) ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi [ [[ACTIVE_LANE_MASK_ENTRY]], [[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], [[ACTIVE_LANE_MASK]], poison) ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP14]], [[ACTIVE_LANE_MASK]], poison) ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = sext [[WIDE_MASKED_LOAD1]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = sext [[WIDE_MASKED_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP15:%.*]] = mul nsw [[TMP10]], [[TMP13]] ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = select [[ACTIVE_LANE_MASK]], [[TMP15]], zeroinitializer ; CHECK-INTERLEAVE1-NEXT: [[TMP19]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP12]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP4]] ; CHECK-INTERLEAVE1-NEXT: [[ACTIVE_LANE_MASK_NEXT]] = call @llvm.get.active.lane.mask.nxv16i1.i64(i64 [[INDEX]], i64 [[TMP9]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = extractelement [[ACTIVE_LANE_MASK_NEXT]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP21:%.*]] = xor i1 [[TMP20]], true ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[TMP19]]) ; CHECK-INTERLEAVE1-NEXT: br label [[EXIT:%.*]] ; CHECK-INTERLEAVE1: exit: ; CHECK-INTERLEAVE1-NEXT: ret i32 [[TMP22]] ; ; CHECK-INTERLEAVED-LABEL: define i32 @dotp_predicated( ; CHECK-INTERLEAVED-SAME: i64 [[N:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 16 ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = sub i64 [[N]], [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = icmp ugt i64 [[N]], [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i64 [[TMP7]], i64 0 ; CHECK-INTERLEAVED-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call @llvm.get.active.lane.mask.nxv16i1.i64(i64 0, i64 [[N]]) ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi [ [[ACTIVE_LANE_MASK_ENTRY]], [[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP19:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], [[ACTIVE_LANE_MASK]], poison) ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP14]], [[ACTIVE_LANE_MASK]], poison) ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = sext [[WIDE_MASKED_LOAD1]] to ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = sext [[WIDE_MASKED_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = mul nsw [[TMP10]], [[TMP13]] ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = select [[ACTIVE_LANE_MASK]], [[TMP15]], zeroinitializer ; CHECK-INTERLEAVED-NEXT: [[TMP19]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP12]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[ACTIVE_LANE_MASK_NEXT]] = call @llvm.get.active.lane.mask.nxv16i1.i64(i64 [[INDEX]], i64 [[TMP9]]) ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = extractelement [[ACTIVE_LANE_MASK_NEXT]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = xor i1 [[TMP20]], true ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP21]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[TMP19]]) ; CHECK-INTERLEAVED-NEXT: br label [[EXIT:%.*]] ; CHECK-INTERLEAVED: exit: ; CHECK-INTERLEAVED-NEXT: ret i32 [[TMP22]] ; ; CHECK-MAXBW-LABEL: define i32 @dotp_predicated( ; CHECK-MAXBW-SAME: i64 [[N:%.*]], ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 16 ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 4 ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = sub i64 [[N]], [[TMP6]] ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = icmp ugt i64 [[N]], [[TMP6]] ; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = select i1 [[TMP8]], i64 [[TMP7]], i64 0 ; CHECK-MAXBW-NEXT: [[ACTIVE_LANE_MASK_ENTRY:%.*]] = call @llvm.get.active.lane.mask.nxv16i1.i64(i64 0, i64 [[N]]) ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[ACTIVE_LANE_MASK:%.*]] = phi [ [[ACTIVE_LANE_MASK_ENTRY]], [[VECTOR_PH]] ], [ [[ACTIVE_LANE_MASK_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP11]], [[ACTIVE_LANE_MASK]], poison) ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_MASKED_LOAD1:%.*]] = call @llvm.masked.load.nxv16i8.p0(ptr align 1 [[TMP14]], [[ACTIVE_LANE_MASK]], poison) ; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = sext [[WIDE_MASKED_LOAD1]] to ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = sext [[WIDE_MASKED_LOAD]] to ; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = mul nsw [[TMP16]], [[TMP13]] ; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = select [[ACTIVE_LANE_MASK]], [[TMP17]], zeroinitializer ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv4i32.nxv16i32( [[VEC_PHI]], [[TMP18]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], [[TMP4]] ; CHECK-MAXBW-NEXT: [[ACTIVE_LANE_MASK_NEXT]] = call @llvm.get.active.lane.mask.nxv16i1.i64(i64 [[INDEX]], i64 [[TMP9]]) ; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = extractelement [[ACTIVE_LANE_MASK_NEXT]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = xor i1 [[TMP19]], true ; CHECK-MAXBW-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP12:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = call i32 @llvm.vector.reduce.add.nxv4i32( [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: br label [[EXIT:%.*]] ; CHECK-MAXBW: exit: ; CHECK-MAXBW-NEXT: ret i32 [[TMP21]] ; entry: br label %for.body for.body: ; preds = %entry, %for.body %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i32 [ 0, %entry ], [ %add, %for.body ] %gep.a = getelementptr inbounds i8, ptr %a, i64 %iv %load.a = load i8, ptr %gep.a, align 1 %ext.a = sext i8 %load.a to i32 %gep.b = getelementptr inbounds i8, ptr %b, i64 %iv %load.b = load i8, ptr %gep.b, align 1 %ext.b = sext i8 %load.b to i32 %mul = mul nsw i32 %ext.b, %ext.a %add = add nsw i32 %mul, %accum %iv.next = add nuw nsw i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, %N br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !7 exit: ; preds = %for.body ret i32 %add } define i32 @not_dotp_extend_user(ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i32 @not_dotp_extend_user( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 8 ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP7:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP3]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP6]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = mul [[TMP5]], [[TMP4]] ; CHECK-INTERLEAVE1-NEXT: [[TMP7]] = add [[TMP14]], [[VEC_PHI]] ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32( [[TMP7]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = call i32 @llvm.vscale.i32() ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = mul nuw i32 [[TMP10]], 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = sub i32 [[TMP11]], 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = extractelement [[TMP5]], i32 [[TMP12]] ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i32 @not_dotp_extend_user( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = mul nuw i64 [[TMP0]], 16 ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP1]] ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = shl nuw i64 [[TMP24]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP3]], i64 [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP3]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load , ptr [[TMP5]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = zext [[WIDE_LOAD2]] to ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = shl nuw i64 [[TMP9]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[TMP8]], i64 [[TMP10]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load , ptr [[TMP8]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = zext [[WIDE_LOAD3]] to ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext [[WIDE_LOAD4]] to ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = mul [[TMP12]], [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = mul [[TMP13]], [[TMP7]] ; CHECK-INTERLEAVED-NEXT: [[TMP16]] = add [[TMP14]], [[VEC_PHI]] ; CHECK-INTERLEAVED-NEXT: [[TMP17]] = add [[TMP15]], [[VEC_PHI1]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP1]] ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add [[TMP17]], [[TMP16]] ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32( [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = call i32 @llvm.vscale.i32() ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = mul nuw i32 [[TMP20]], 8 ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = sub i32 [[TMP21]], 1 ; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = extractelement [[TMP13]], i32 [[TMP22]] ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i32 @not_dotp_extend_user( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: br label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 8 ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 1024, [[TMP3]] ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 1024, [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP24:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = zext [[WIDE_LOAD2]] to ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[TMP14]], align 1 ; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = zext [[WIDE_LOAD4]] to ; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = mul [[TMP20]], [[TMP13]] ; CHECK-MAXBW-NEXT: [[TMP24]] = add [[TMP22]], [[VEC_PHI1]] ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP25]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP14:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = call i32 @llvm.vector.reduce.add.nxv8i32( [[TMP24]]) ; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = call i32 @llvm.vscale.i32() ; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = mul nuw i32 [[TMP17]], 8 ; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = sub i32 [[TMP18]], 1 ; CHECK-MAXBW-NEXT: [[TMP21:%.*]] = extractelement [[TMP20]], i32 [[TMP19]] ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 1024, [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH:%.*]] ; CHECK-MAXBW: scalar.ph: ; entry: br label %for.body for.body: ; preds = %for.body, %entry %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] %accum = phi i32 [ 0, %entry ], [ %add, %for.body ] %gep.a = getelementptr i8, ptr %a, i64 %iv %load.a = load i8, ptr %gep.a, align 1 %ext.a = zext i8 %load.a to i32 %gep.b = getelementptr i8, ptr %b, i64 %iv %load.b = load i8, ptr %gep.b, align 1 %ext.b = zext i8 %load.b to i32 %mul = mul i32 %ext.b, %ext.a %add = add i32 %mul, %accum %iv.next = add i64 %iv, 1 %exitcond.not = icmp eq i64 %iv.next, 1024 br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body %result = add i32 %add, %ext.b ret i32 %result } define i64 @dotp_cost_disagreement(ptr %a, ptr %b) #0 { ; CHECK-INTERLEAVE1-LABEL: define i64 @dotp_cost_disagreement( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 4 ; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 41, [[TMP1]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16 ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 41, [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 41, [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = add nuw nsw i64 [[INDEX]], 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP10]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = mul nuw nsw [[TMP12]], [[TMP8]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI]], [[TMP9]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP13]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64( [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 41, [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i64 @dotp_cost_disagreement( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 5 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 41, [[TMP1]] ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 32 ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 41, [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 41, [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE5:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = shl nuw i64 [[TMP5]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP7]], i64 [[TMP6]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load , ptr [[TMP8]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = add nuw nsw i64 [[INDEX]], 1 ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = shl nuw i64 [[TMP10]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[TMP15]], i64 [[TMP11]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load , ptr [[TMP15]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[TMP12]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = zext [[WIDE_LOAD3]] to ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = mul nuw nsw [[TMP13]], [[TMP21]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI]], [[TMP22]]) ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = zext [[WIDE_LOAD4]] to ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = zext [[WIDE_LOAD2]] to ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = mul nuw nsw [[TMP16]], [[TMP17]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE5]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI1]], [[TMP18]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add [[PARTIAL_REDUCE5]], [[PARTIAL_REDUCE]] ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64( [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 41, [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i64 @dotp_cost_disagreement( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = shl nuw nsw i64 [[TMP0]], 4 ; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 41, [[TMP1]] ; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = mul nuw i64 [[TMP2]], 16 ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 41, [[TMP3]] ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 41, [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[TMP7]], align 1 ; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = add nuw nsw i64 [[INDEX]], 1 ; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP10]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD1:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = zext [[WIDE_LOAD1]] to ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = mul nuw nsw [[TMP12]], [[TMP8]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI]], [[TMP9]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP3]] ; CHECK-MAXBW-NEXT: [[TMP15:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP15]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64( [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 41, [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-MAXBW: scalar.ph: ; entry: br label %for.body for.body: ; preds = %entry, %for.body %i.iv = phi i64 [ 0, %entry ], [ %i.iv.next, %for.body ] %sum = phi i64 [ 0, %entry ], [ %add, %for.body ] %arrayidx = getelementptr inbounds nuw i8, ptr %a, i64 %i.iv %0 = load i8, ptr %arrayidx, align 1 %conv = zext i8 %0 to i64 %i.iv.next = add nuw nsw i64 %i.iv, 1 %arrayidx2 = getelementptr inbounds nuw i8, ptr %b, i64 %i.iv.next %1 = load i8, ptr %arrayidx2, align 1 %conv3 = zext i8 %1 to i64 %mul = mul nuw nsw i64 %conv3, %conv %add = add i64 %sum, %mul %exitcond.not = icmp eq i64 %i.iv.next, 41 br i1 %exitcond.not, label %exit, label %for.body exit: ; preds = %for.body ret i64 %add } define void @not_dotp_not_phi2(ptr %matrix, i32 %n) #0 { ; CHECK-INTERLEAVE1-LABEL: define void @not_dotp_not_phi2( ; CHECK-INTERLEAVE1-SAME: ptr [[MATRIX:%.*]], i32 [[N:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP]], label [[FOR_PREHEADER:%.*]], label [[EXIT:%.*]] ; CHECK-INTERLEAVE1: for.preheader: ; CHECK-INTERLEAVE1-NEXT: [[LOAD_A:%.*]] = load i8, ptr null, align 1 ; CHECK-INTERLEAVE1-NEXT: [[LOAD_A1:%.*]] = load i8, ptr inttoptr (i64 1 to ptr), align 1 ; CHECK-INTERLEAVE1-NEXT: [[A_EXT:%.*]] = sext i8 [[LOAD_A]] to i32 ; CHECK-INTERLEAVE1-NEXT: [[A_EXT1:%.*]] = sext i8 [[LOAD_A1]] to i32 ; CHECK-INTERLEAVE1-NEXT: br label [[FOR_BODY:%.*]] ; CHECK-INTERLEAVE1: for.body: ; CHECK-INTERLEAVE1-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_PREHEADER]] ] ; CHECK-INTERLEAVE1-NEXT: [[PTR:%.*]] = phi ptr [ [[SCEVGEP:%.*]], [[FOR_BODY]] ], [ [[MATRIX]], [[FOR_PREHEADER]] ] ; CHECK-INTERLEAVE1-NEXT: [[ACCUM:%.*]] = phi i32 [ [[ADD_1:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_PREHEADER]] ] ; CHECK-INTERLEAVE1-NEXT: [[GEP_B:%.*]] = getelementptr i8, ptr [[PTR]], i64 1 ; CHECK-INTERLEAVE1-NEXT: [[GEP_B1:%.*]] = getelementptr i8, ptr [[PTR]], i64 2 ; CHECK-INTERLEAVE1-NEXT: [[LOAD_B:%.*]] = load i8, ptr [[GEP_B]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[B_EXT:%.*]] = sext i8 [[LOAD_B]] to i32 ; CHECK-INTERLEAVE1-NEXT: [[MUL:%.*]] = mul nsw i32 [[A_EXT]], [[B_EXT]] ; CHECK-INTERLEAVE1-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[ACCUM]] ; CHECK-INTERLEAVE1-NEXT: [[LOAD_B1:%.*]] = load i8, ptr [[GEP_B1]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[B_EXT1:%.*]] = sext i8 [[LOAD_B1]] to i32 ; CHECK-INTERLEAVE1-NEXT: [[MUL_1:%.*]] = mul nsw i32 [[A_EXT1]], [[B_EXT1]] ; CHECK-INTERLEAVE1-NEXT: [[ADD_1]] = add i32 [[MUL_1]], [[ADD]] ; CHECK-INTERLEAVE1-NEXT: [[SCEVGEP]] = getelementptr i8, ptr [[PTR]], i64 16 ; CHECK-INTERLEAVE1-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 ; CHECK-INTERLEAVE1-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[IV_NEXT]], [[N]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_EXIT:%.*]], label [[FOR_BODY]] ; CHECK-INTERLEAVE1: for.exit: ; CHECK-INTERLEAVE1-NEXT: [[ADD_1_LCSSA:%.*]] = phi i32 [ [[ADD_1]], [[FOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[ADD_FLOAT:%.*]] = sitofp i32 [[ADD_1_LCSSA]] to float ; CHECK-INTERLEAVE1-NEXT: br label [[EXIT]] ; CHECK-INTERLEAVE1: exit: ; CHECK-INTERLEAVE1-NEXT: [[RESULT:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD_FLOAT]], [[FOR_EXIT]] ] ; CHECK-INTERLEAVE1-NEXT: store float [[RESULT]], ptr [[MATRIX]], align 4 ; CHECK-INTERLEAVE1-NEXT: ret void ; ; CHECK-INTERLEAVED-LABEL: define void @not_dotp_not_phi2( ; CHECK-INTERLEAVED-SAME: ptr [[MATRIX:%.*]], i32 [[N:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP]], label [[FOR_PREHEADER:%.*]], label [[EXIT:%.*]] ; CHECK-INTERLEAVED: for.preheader: ; CHECK-INTERLEAVED-NEXT: [[LOAD_A:%.*]] = load i8, ptr null, align 1 ; CHECK-INTERLEAVED-NEXT: [[LOAD_A1:%.*]] = load i8, ptr inttoptr (i64 1 to ptr), align 1 ; CHECK-INTERLEAVED-NEXT: [[A_EXT:%.*]] = sext i8 [[LOAD_A]] to i32 ; CHECK-INTERLEAVED-NEXT: [[A_EXT1:%.*]] = sext i8 [[LOAD_A1]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], 2 ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 2 ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: [[IND_END:%.*]] = trunc i64 [[N_VEC]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = mul i64 [[N_VEC]], 16 ; CHECK-INTERLEAVED-NEXT: [[IND_END1:%.*]] = getelementptr i8, ptr [[MATRIX]], i64 [[TMP1]] ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP22:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[TMP23:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 16 ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = add i64 [[OFFSET_IDX]], 16 ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[MATRIX]], i64 [[OFFSET_IDX]] ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[MATRIX]], i64 [[TMP3]] ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 1 ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[NEXT_GEP3]], i64 1 ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 2 ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = getelementptr i8, ptr [[NEXT_GEP3]], i64 2 ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = load i8, ptr [[TMP4]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = load i8, ptr [[TMP5]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = sext i8 [[TMP8]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = sext i8 [[TMP9]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = mul nsw i32 [[A_EXT]], [[TMP10]] ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = mul nsw i32 [[A_EXT]], [[TMP11]] ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = add i32 [[TMP12]], [[VEC_PHI]] ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = add i32 [[TMP13]], [[VEC_PHI2]] ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = load i8, ptr [[TMP6]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = load i8, ptr [[TMP7]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = sext i8 [[TMP16]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = sext i8 [[TMP17]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = mul nsw i32 [[A_EXT1]], [[TMP18]] ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = mul nsw i32 [[A_EXT1]], [[TMP19]] ; CHECK-INTERLEAVED-NEXT: [[TMP22]] = add i32 [[TMP20]], [[TMP14]] ; CHECK-INTERLEAVED-NEXT: [[TMP23]] = add i32 [[TMP21]], [[TMP15]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2 ; CHECK-INTERLEAVED-NEXT: [[TMP24:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP24]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add i32 [[TMP23]], [[TMP22]] ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[FOR_EXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define void @not_dotp_not_phi2( ; CHECK-MAXBW-SAME: ptr [[MATRIX:%.*]], i32 [[N:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-MAXBW-NEXT: br i1 [[CMP]], label [[FOR_PREHEADER:%.*]], label [[EXIT:%.*]] ; CHECK-MAXBW: for.preheader: ; CHECK-MAXBW-NEXT: [[LOAD_A:%.*]] = load i8, ptr null, align 1 ; CHECK-MAXBW-NEXT: [[LOAD_A1:%.*]] = load i8, ptr inttoptr (i64 1 to ptr), align 1 ; CHECK-MAXBW-NEXT: [[A_EXT:%.*]] = sext i8 [[LOAD_A]] to i32 ; CHECK-MAXBW-NEXT: [[A_EXT1:%.*]] = sext i8 [[LOAD_A1]] to i32 ; CHECK-MAXBW-NEXT: br label [[FOR_BODY:%.*]] ; CHECK-MAXBW: for.body: ; CHECK-MAXBW-NEXT: [[IV:%.*]] = phi i32 [ [[IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_PREHEADER]] ] ; CHECK-MAXBW-NEXT: [[PTR:%.*]] = phi ptr [ [[SCEVGEP:%.*]], [[FOR_BODY]] ], [ [[MATRIX]], [[FOR_PREHEADER]] ] ; CHECK-MAXBW-NEXT: [[ACCUM:%.*]] = phi i32 [ [[ADD_1:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_PREHEADER]] ] ; CHECK-MAXBW-NEXT: [[GEP_B:%.*]] = getelementptr i8, ptr [[PTR]], i64 1 ; CHECK-MAXBW-NEXT: [[GEP_B1:%.*]] = getelementptr i8, ptr [[PTR]], i64 2 ; CHECK-MAXBW-NEXT: [[LOAD_B:%.*]] = load i8, ptr [[GEP_B]], align 1 ; CHECK-MAXBW-NEXT: [[B_EXT:%.*]] = sext i8 [[LOAD_B]] to i32 ; CHECK-MAXBW-NEXT: [[MUL:%.*]] = mul nsw i32 [[A_EXT]], [[B_EXT]] ; CHECK-MAXBW-NEXT: [[ADD:%.*]] = add i32 [[MUL]], [[ACCUM]] ; CHECK-MAXBW-NEXT: [[LOAD_B1:%.*]] = load i8, ptr [[GEP_B1]], align 1 ; CHECK-MAXBW-NEXT: [[B_EXT1:%.*]] = sext i8 [[LOAD_B1]] to i32 ; CHECK-MAXBW-NEXT: [[MUL_1:%.*]] = mul nsw i32 [[A_EXT1]], [[B_EXT1]] ; CHECK-MAXBW-NEXT: [[ADD_1]] = add i32 [[MUL_1]], [[ADD]] ; CHECK-MAXBW-NEXT: [[SCEVGEP]] = getelementptr i8, ptr [[PTR]], i64 16 ; CHECK-MAXBW-NEXT: [[IV_NEXT]] = add nuw nsw i32 [[IV]], 1 ; CHECK-MAXBW-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[IV_NEXT]], [[N]] ; CHECK-MAXBW-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_EXIT:%.*]], label [[FOR_BODY]] ; CHECK-MAXBW: for.exit: ; CHECK-MAXBW-NEXT: [[ADD_1_LCSSA:%.*]] = phi i32 [ [[ADD_1]], [[FOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[ADD_FLOAT:%.*]] = sitofp i32 [[ADD_1_LCSSA]] to float ; CHECK-MAXBW-NEXT: br label [[EXIT]] ; CHECK-MAXBW: exit: ; CHECK-MAXBW-NEXT: [[RESULT:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ADD_FLOAT]], [[FOR_EXIT]] ] ; CHECK-MAXBW-NEXT: store float [[RESULT]], ptr [[MATRIX]], align 4 ; CHECK-MAXBW-NEXT: ret void ; entry: %cmp = icmp sgt i32 %n, 0 br i1 %cmp, label %for.preheader, label %exit for.preheader: ; preds = %entry %load.a = load i8, ptr inttoptr (i64 0 to ptr), align 1 %load.a1 = load i8, ptr inttoptr (i64 1 to ptr), align 1 %a.ext = sext i8 %load.a to i32 %a.ext1 = sext i8 %load.a1 to i32 br label %for.body for.body: ; preds = %for.preheader, %for.body %iv = phi i32 [ %iv.next, %for.body ], [ 0, %for.preheader ] %ptr = phi ptr [ %scevgep, %for.body ], [ %matrix, %for.preheader ] %accum = phi i32 [ %add.1, %for.body ], [ 0, %for.preheader ] %gep.b = getelementptr i8, ptr %ptr, i64 1 %gep.b1 = getelementptr i8, ptr %ptr, i64 2 %load.b = load i8, ptr %gep.b, align 1 %b.ext = sext i8 %load.b to i32 %mul = mul nsw i32 %a.ext, %b.ext %add = add i32 %mul, %accum %load.b1 = load i8, ptr %gep.b1, align 1 %b.ext1 = sext i8 %load.b1 to i32 %mul.1 = mul nsw i32 %a.ext1, %b.ext1 %add.1 = add i32 %mul.1, %add %scevgep = getelementptr i8, ptr %ptr, i64 16 %iv.next = add nuw nsw i32 %iv, 1 %exitcond.not = icmp eq i32 %iv.next, %n br i1 %exitcond.not, label %for.exit, label %for.body for.exit: ; preds = %for.body %add.1.lcssa = phi i32 [ %add.1, %for.body ] %add.float = sitofp i32 %add.1.lcssa to float br label %exit exit: ; preds = %for.exit, %entry %result = phi float [ 0.000000e+00, %entry ], [ %add.float, %for.exit ] store float %result, ptr %matrix, align 4 ret void } define i64 @not_dotp_ext_outside_plan(ptr %a, i16 %b, i64 %n) #0 { ; CHECK-INTERLEAVE1-LABEL: define i64 @not_dotp_ext_outside_plan( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], i16 [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 0 ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_PH:%.*]] ; CHECK-INTERLEAVE1: for.ph: ; CHECK-INTERLEAVE1-NEXT: [[EXT_B:%.*]] = zext i16 [[B]] to i64 ; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8 ; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 8 ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[EXT_B]], i64 0 ; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul nuw nsw <8 x i64> [[TMP3]], [[BROADCAST_SPLAT]] ; CHECK-INTERLEAVE1-NEXT: [[TMP5]] = add <8 x i64> [[TMP4]], [[VEC_PHI]] ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP16:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[TMP5]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i64 @not_dotp_ext_outside_plan( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], i16 [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 0 ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_PH:%.*]] ; CHECK-INTERLEAVED: for.ph: ; CHECK-INTERLEAVED-NEXT: [[EXT_B:%.*]] = zext i16 [[B]] to i64 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 16 ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 16 ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[EXT_B]], i64 0 ; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i16, ptr [[TMP1]], i64 8 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i16>, ptr [[TMP3]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <8 x i16> [[WIDE_LOAD2]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = mul nuw nsw <8 x i64> [[TMP4]], [[BROADCAST_SPLAT]] ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <8 x i64> [[TMP5]], [[BROADCAST_SPLAT]] ; CHECK-INTERLEAVED-NEXT: [[TMP8]] = add <8 x i64> [[TMP6]], [[VEC_PHI]] ; CHECK-INTERLEAVED-NEXT: [[TMP9]] = add <8 x i64> [[TMP7]], [[VEC_PHI1]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <8 x i64> [[TMP9]], [[TMP8]] ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i64 @not_dotp_ext_outside_plan( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], i16 [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 0 ; CHECK-MAXBW-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_PH:%.*]] ; CHECK-MAXBW: for.ph: ; CHECK-MAXBW-NEXT: [[EXT_B:%.*]] = zext i16 [[B]] to i64 ; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8 ; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 8 ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[EXT_B]], i64 0 ; CHECK-MAXBW-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP7]], align 2 ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[TMP1]], [[BROADCAST_SPLAT]] ; CHECK-MAXBW-NEXT: [[TMP3]] = add <8 x i64> [[TMP2]], [[VEC_PHI]] ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[TMP3]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-MAXBW: scalar.ph: ; entry: %cmp = icmp eq i64 %n, 0 br i1 %cmp, label %exit, label %for.ph for.ph: ; preds = %entry %ext.b = zext i16 %b to i64 br label %for.body for.body: ; preds = %for.body.lr.ph, %for.body %iv = phi i64 [ 0, %for.ph ], [ %iv.next, %for.body ] %accum = phi i64 [ 0, %for.ph ], [ %add, %for.body ] %gep.a = getelementptr inbounds nuw i16, ptr %a, i64 %iv %load.a = load i16, ptr %gep.a, align 2 %ext.a = zext i16 %load.a to i64 %mul = mul nuw nsw i64 %ext.a, %ext.b %add = add i64 %mul, %accum %iv.next = add nuw nsw i64 %iv, 1 %cmp.1 = icmp eq i64 %iv.next, %n br i1 %cmp.1, label %exit, label %for.body exit: ; preds = %for.cond.cleanup.loopexit, %entry %result = phi i64 [ 0, %entry ], [ %add, %for.body ] ret i64 %result } define i64 @not_dotp_ext_outside_plan2(ptr %a, i16 %b, i64 %n) #0 { ; CHECK-INTERLEAVE1-LABEL: define i64 @not_dotp_ext_outside_plan2( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], i16 [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 0 ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_PH:%.*]] ; CHECK-INTERLEAVE1: for.ph: ; CHECK-INTERLEAVE1-NEXT: [[EXT_B:%.*]] = zext i16 [[B]] to i64 ; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8 ; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 8 ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[EXT_B]], i64 0 ; CHECK-INTERLEAVE1-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul nuw nsw <8 x i64> [[BROADCAST_SPLAT]], [[TMP3]] ; CHECK-INTERLEAVE1-NEXT: [[TMP5]] = add <8 x i64> [[TMP4]], [[VEC_PHI]] ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP18:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[TMP5]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define i64 @not_dotp_ext_outside_plan2( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], i16 [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 0 ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_PH:%.*]] ; CHECK-INTERLEAVED: for.ph: ; CHECK-INTERLEAVED-NEXT: [[EXT_B:%.*]] = zext i16 [[B]] to i64 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 16 ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 16 ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[EXT_B]], i64 0 ; CHECK-INTERLEAVED-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP8:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = getelementptr inbounds nuw i16, ptr [[TMP1]], i64 8 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP1]], align 2 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD2:%.*]] = load <8 x i16>, ptr [[TMP3]], align 2 ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = zext <8 x i16> [[WIDE_LOAD2]] to <8 x i64> ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = mul nuw nsw <8 x i64> [[BROADCAST_SPLAT]], [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = mul nuw nsw <8 x i64> [[BROADCAST_SPLAT]], [[TMP5]] ; CHECK-INTERLEAVED-NEXT: [[TMP8]] = add <8 x i64> [[TMP6]], [[VEC_PHI]] ; CHECK-INTERLEAVED-NEXT: [[TMP9]] = add <8 x i64> [[TMP7]], [[VEC_PHI1]] ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add <8 x i64> [[TMP9]], [[TMP8]] ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define i64 @not_dotp_ext_outside_plan2( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], i16 [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[CMP:%.*]] = icmp eq i64 [[N]], 0 ; CHECK-MAXBW-NEXT: br i1 [[CMP]], label [[EXIT:%.*]], label [[FOR_PH:%.*]] ; CHECK-MAXBW: for.ph: ; CHECK-MAXBW-NEXT: [[EXT_B:%.*]] = zext i16 [[B]] to i64 ; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8 ; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 8 ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <8 x i64> poison, i64 [[EXT_B]], i64 0 ; CHECK-MAXBW-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <8 x i64> [[BROADCAST_SPLATINSERT]], <8 x i64> poison, <8 x i32> zeroinitializer ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <8 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = getelementptr inbounds nuw i16, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <8 x i16>, ptr [[TMP7]], align 2 ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = zext <8 x i16> [[WIDE_LOAD]] to <8 x i64> ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = mul nuw nsw <8 x i64> [[BROADCAST_SPLAT]], [[TMP1]] ; CHECK-MAXBW-NEXT: [[TMP3]] = add <8 x i64> [[TMP2]], [[VEC_PHI]] ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP20:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = call i64 @llvm.vector.reduce.add.v8i64(<8 x i64> [[TMP3]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-MAXBW: scalar.ph: ; entry: %cmp = icmp eq i64 %n, 0 br i1 %cmp, label %exit, label %for.ph for.ph: ; preds = %entry %ext.b = zext i16 %b to i64 br label %for.body for.body: ; preds = %for.body.lr.ph, %for.body %iv = phi i64 [ 0, %for.ph ], [ %iv.next, %for.body ] %accum = phi i64 [ 0, %for.ph ], [ %add, %for.body ] %gep.a = getelementptr inbounds nuw i16, ptr %a, i64 %iv %load.a = load i16, ptr %gep.a, align 2 %ext.a = zext i16 %load.a to i64 %mul = mul nuw nsw i64 %ext.b, %ext.a %add = add i64 %mul, %accum %iv.next = add nuw nsw i64 %iv, 1 %cmp.1 = icmp eq i64 %iv.next, %n br i1 %cmp.1, label %exit, label %for.body exit: ; preds = %for.cond.cleanup.loopexit, %entry %result = phi i64 [ 0, %entry ], [ %add, %for.body ] ret i64 %result } define dso_local i32 @not_dotp_vscale1(ptr %a, ptr %b, i32 %n, i64 %cost) #0 { ; CHECK-INTERLEAVE1-LABEL: define dso_local i32 @not_dotp_vscale1( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i32 [[N:%.*]], i64 [[COST:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP]], label [[FOR_BODY_PREHEADER:%.*]], label [[EXIT:%.*]] ; CHECK-INTERLEAVE1: for.body.preheader: ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64 ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 4 ; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 16 ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP4]] ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = trunc i64 [[N_VEC]] to i32 ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[A]], i64 [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = insertelement zeroinitializer, i64 [[COST]], i32 0 ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi [ [[TMP12]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[NEXT_GEP]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD2:%.*]] = load , ptr [[NEXT_GEP1]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = zext [[WIDE_LOAD2]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP10:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = mul nuw nsw [[TMP14]], [[TMP10]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI]], [[TMP11]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]] ; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP19]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64( [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define dso_local i32 @not_dotp_vscale1( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i32 [[N:%.*]], i64 [[COST:%.*]]) #[[ATTR0]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP]], label [[FOR_BODY_PREHEADER:%.*]], label [[EXIT:%.*]] ; CHECK-INTERLEAVED: for.body.preheader: ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64 ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 5 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]] ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 32 ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = trunc i64 [[N_VEC]] to i32 ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[A]], i64 [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: [[TMP21:%.*]] = insertelement zeroinitializer, i64 [[COST]], i32 0 ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi [ [[TMP21]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi [ zeroinitializer, [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE6:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP10:%.*]] = shl nuw i64 [[TMP23]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = getelementptr i8, ptr [[NEXT_GEP]], i64 [[TMP10]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[NEXT_GEP]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD3:%.*]] = load , ptr [[TMP11]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = shl nuw i64 [[TMP12]], 4 ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = getelementptr i8, ptr [[NEXT_GEP2]], i64 [[TMP13]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD4:%.*]] = load , ptr [[NEXT_GEP2]], align 1 ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD5:%.*]] = load , ptr [[TMP14]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP15:%.*]] = zext [[WIDE_LOAD4]] to ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = mul nuw nsw [[TMP15]], [[TMP16]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI]], [[TMP17]]) ; CHECK-INTERLEAVED-NEXT: [[TMP18:%.*]] = zext [[WIDE_LOAD5]] to ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = zext [[WIDE_LOAD3]] to ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = mul nuw nsw [[TMP18]], [[TMP19]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE6]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI1]], [[TMP20]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]] ; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[BIN_RDX:%.*]] = add [[PARTIAL_REDUCE6]], [[PARTIAL_REDUCE]] ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64( [[BIN_RDX]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define dso_local i32 @not_dotp_vscale1( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]], i32 [[N:%.*]], i64 [[COST:%.*]]) #[[ATTR0]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[CMP:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-MAXBW-NEXT: br i1 [[CMP]], label [[FOR_BODY_PREHEADER:%.*]], label [[EXIT:%.*]] ; CHECK-MAXBW: for.body.preheader: ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = zext i32 [[N]] to i64 ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = shl nuw nsw i64 [[TMP1]], 4 ; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP0]], [[TMP2]] ; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = call i64 @llvm.vscale.i64() ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = mul nuw i64 [[TMP3]], 16 ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], [[TMP4]] ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = trunc i64 [[N_VEC]] to i32 ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = getelementptr i8, ptr [[A]], i64 [[N_VEC]] ; CHECK-MAXBW-NEXT: [[TMP9:%.*]] = getelementptr i8, ptr [[B]], i64 [[N_VEC]] ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = insertelement zeroinitializer, i64 [[COST]], i32 0 ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi [ [[TMP12]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[NEXT_GEP1:%.*]] = getelementptr i8, ptr [[B]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load , ptr [[NEXT_GEP]], align 1 ; CHECK-MAXBW-NEXT: [[WIDE_LOAD2:%.*]] = load , ptr [[NEXT_GEP1]], align 1 ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = zext [[WIDE_LOAD2]] to ; CHECK-MAXBW-NEXT: [[TMP10:%.*]] = zext [[WIDE_LOAD]] to ; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = mul nuw nsw [[TMP14]], [[TMP10]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call @llvm.vector.partial.reduce.add.nxv2i64.nxv16i64( [[VEC_PHI]], [[TMP11]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP4]] ; CHECK-MAXBW-NEXT: [[TMP18:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP18]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64( [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP0]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[EXIT_LOOPEXIT:%.*]], label [[SCALAR_PH]] ; CHECK-MAXBW: scalar.ph: ; entry: %cmp = icmp sgt i32 %n, 0 br i1 %cmp, label %for.body, label %exit for.body: ; preds = %for.body.preheader, %for.body %iv = phi i32 [ %iv.next, %for.body ], [ 0, %entry ] %accum = phi i64 [ %add, %for.body ], [ %cost, %entry ] %a.ptr = phi ptr [ %a.gep, %for.body ], [ %a, %entry ] %b.ptr = phi ptr [ %b.gep, %for.body ], [ %b, %entry ] %a.load = load i8, ptr %a.ptr, align 1 %a.ext = zext i8 %a.load to i64 %b.load = load i8, ptr %b.ptr, align 1 %b.ext = zext i8 %b.load to i64 %mul = mul nuw nsw i64 %b.ext, %a.ext %add = add nsw i64 %mul, %accum %a.gep = getelementptr inbounds nuw i8, ptr %a.ptr, i64 1 %b.gep = getelementptr inbounds nuw i8, ptr %b.ptr, i64 1 %iv.next = add nuw nsw i32 %iv, 1 %cmp.2 = icmp eq i32 %iv.next, %n br i1 %cmp.2, label %exit, label %for.body exit: ; preds = %for.cond.cleanup.loopexit, %entry %cost.result = phi i64 [ %cost, %entry ], [ %add, %for.body ] %result = trunc i64 %cost.result to i32 ret i32 %result } define dso_local void @not_dotp_high_register_pressure(ptr %a, ptr %b, ptr %sum, i32 %n) #1 { ; CHECK-INTERLEAVE1-LABEL: define dso_local void @not_dotp_high_register_pressure( ; CHECK-INTERLEAVE1-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[SUM:%.*]], i32 [[N:%.*]]) #[[ATTR1]] { ; CHECK-INTERLEAVE1-NEXT: entry: ; CHECK-INTERLEAVE1-NEXT: [[CMP100:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP100]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_COND_CLEANUP:%.*]] ; CHECK-INTERLEAVE1: for.body.lr.ph: ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 4 ; CHECK-INTERLEAVE1-NEXT: [[GEP_B_12:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 8 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX31:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 12 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX40:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 16 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX49:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 20 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX58:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 24 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX67:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 28 ; CHECK-INTERLEAVE1-NEXT: [[SUM_PROMOTED:%.*]] = load i32, ptr [[SUM]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX13_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX13]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[GEP_B_12_PROMOTED:%.*]] = load i32, ptr [[GEP_B_12]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX31_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX31]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX40_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX40]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX49_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX49]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX58_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX58]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[ARRAYIDX67_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX67]], align 4 ; CHECK-INTERLEAVE1-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[N]] to i64 ; CHECK-INTERLEAVE1-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16 ; CHECK-INTERLEAVE1-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVE1: vector.ph: ; CHECK-INTERLEAVE1-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16 ; CHECK-INTERLEAVE1-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]] ; CHECK-INTERLEAVE1-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX67_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX58_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX49_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX40_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX31_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[GEP_B_12_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX13_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[SUM_PROMOTED]], i32 0 ; CHECK-INTERLEAVE1-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVE1: vector.body: ; CHECK-INTERLEAVE1-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE21:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ [[TMP1]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE20:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE19:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ [[TMP3]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE18:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI4:%.*]] = phi <4 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE17:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI5:%.*]] = phi <4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE16:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI6:%.*]] = phi <4 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE15:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVE1-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[TMP11:%.*]] = shl nsw i64 [[INDEX]], 3 ; CHECK-INTERLEAVE1-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP11]] ; CHECK-INTERLEAVE1-NEXT: [[WIDE_VEC:%.*]] = load <128 x i8>, ptr [[TMP12]], align 1 ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC8:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC9:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC10:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC11:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC12:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC13:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP13]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[STRIDED_VEC9]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP19]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP22]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP25]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP26:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP26]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP28]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP29:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP29]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP31]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32> ; CHECK-INTERLEAVE1-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP44]] ; CHECK-INTERLEAVE1-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP34]]) ; CHECK-INTERLEAVE1-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVE1-NEXT: [[TMP37:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[TMP37]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP22:![0-9]+]] ; CHECK-INTERLEAVE1: middle.block: ; CHECK-INTERLEAVE1-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP45:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]]) ; CHECK-INTERLEAVE1-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVE1-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] ; CHECK-INTERLEAVE1-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVE1: scalar.ph: ; ; CHECK-INTERLEAVED-LABEL: define dso_local void @not_dotp_high_register_pressure( ; CHECK-INTERLEAVED-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[SUM:%.*]], i32 [[N:%.*]]) #[[ATTR1]] { ; CHECK-INTERLEAVED-NEXT: entry: ; CHECK-INTERLEAVED-NEXT: [[CMP100:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP100]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_COND_CLEANUP:%.*]] ; CHECK-INTERLEAVED: for.body.lr.ph: ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 4 ; CHECK-INTERLEAVED-NEXT: [[GEP_B_12:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 8 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX31:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 12 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX40:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 16 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX49:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 20 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX58:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 24 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX67:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 28 ; CHECK-INTERLEAVED-NEXT: [[SUM_PROMOTED:%.*]] = load i32, ptr [[SUM]], align 4 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX13_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX13]], align 4 ; CHECK-INTERLEAVED-NEXT: [[GEP_B_12_PROMOTED:%.*]] = load i32, ptr [[GEP_B_12]], align 4 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX31_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX31]], align 4 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX40_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX40]], align 4 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX49_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX49]], align 4 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX58_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX58]], align 4 ; CHECK-INTERLEAVED-NEXT: [[ARRAYIDX67_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX67]], align 4 ; CHECK-INTERLEAVED-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[N]] to i64 ; CHECK-INTERLEAVED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16 ; CHECK-INTERLEAVED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-INTERLEAVED: vector.ph: ; CHECK-INTERLEAVED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16 ; CHECK-INTERLEAVED-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]] ; CHECK-INTERLEAVED-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX67_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX58_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX49_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX40_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX31_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[GEP_B_12_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX13_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[SUM_PROMOTED]], i32 0 ; CHECK-INTERLEAVED-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-INTERLEAVED: vector.body: ; CHECK-INTERLEAVED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE21:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ [[TMP1]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE20:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE19:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ [[TMP3]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE18:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI4:%.*]] = phi <4 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE17:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI5:%.*]] = phi <4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE16:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI6:%.*]] = phi <4 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE15:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-INTERLEAVED-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1 ; CHECK-INTERLEAVED-NEXT: [[TMP11:%.*]] = shl nsw i64 [[INDEX]], 3 ; CHECK-INTERLEAVED-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP11]] ; CHECK-INTERLEAVED-NEXT: [[WIDE_VEC:%.*]] = load <128 x i8>, ptr [[TMP12]], align 1 ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC8:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC9:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC10:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC11:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC12:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC13:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP13]]) ; CHECK-INTERLEAVED-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]]) ; CHECK-INTERLEAVED-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[STRIDED_VEC9]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP19]]) ; CHECK-INTERLEAVED-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP22]]) ; CHECK-INTERLEAVED-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP25]]) ; CHECK-INTERLEAVED-NEXT: [[TMP26:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP26]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP28]]) ; CHECK-INTERLEAVED-NEXT: [[TMP45:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP45]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP31]]) ; CHECK-INTERLEAVED-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32> ; CHECK-INTERLEAVED-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP44]] ; CHECK-INTERLEAVED-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP34]]) ; CHECK-INTERLEAVED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-INTERLEAVED-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] ; CHECK-INTERLEAVED: middle.block: ; CHECK-INTERLEAVED-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]]) ; CHECK-INTERLEAVED-NEXT: [[TMP37:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]]) ; CHECK-INTERLEAVED-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]]) ; CHECK-INTERLEAVED-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]]) ; CHECK-INTERLEAVED-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]]) ; CHECK-INTERLEAVED-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]]) ; CHECK-INTERLEAVED-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]]) ; CHECK-INTERLEAVED-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]]) ; CHECK-INTERLEAVED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] ; CHECK-INTERLEAVED-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]] ; CHECK-INTERLEAVED: scalar.ph: ; ; CHECK-MAXBW-LABEL: define dso_local void @not_dotp_high_register_pressure( ; CHECK-MAXBW-SAME: ptr [[A:%.*]], ptr [[B:%.*]], ptr [[SUM:%.*]], i32 [[N:%.*]]) #[[ATTR1]] { ; CHECK-MAXBW-NEXT: entry: ; CHECK-MAXBW-NEXT: [[CMP100:%.*]] = icmp sgt i32 [[N]], 0 ; CHECK-MAXBW-NEXT: br i1 [[CMP100]], label [[FOR_BODY_LR_PH:%.*]], label [[FOR_COND_CLEANUP:%.*]] ; CHECK-MAXBW: for.body.lr.ph: ; CHECK-MAXBW-NEXT: [[ARRAYIDX13:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 4 ; CHECK-MAXBW-NEXT: [[GEP_B_12:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 8 ; CHECK-MAXBW-NEXT: [[ARRAYIDX31:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 12 ; CHECK-MAXBW-NEXT: [[ARRAYIDX40:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 16 ; CHECK-MAXBW-NEXT: [[ARRAYIDX49:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 20 ; CHECK-MAXBW-NEXT: [[ARRAYIDX58:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 24 ; CHECK-MAXBW-NEXT: [[ARRAYIDX67:%.*]] = getelementptr inbounds nuw i8, ptr [[SUM]], i64 28 ; CHECK-MAXBW-NEXT: [[SUM_PROMOTED:%.*]] = load i32, ptr [[SUM]], align 4 ; CHECK-MAXBW-NEXT: [[ARRAYIDX13_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX13]], align 4 ; CHECK-MAXBW-NEXT: [[GEP_B_12_PROMOTED:%.*]] = load i32, ptr [[GEP_B_12]], align 4 ; CHECK-MAXBW-NEXT: [[ARRAYIDX31_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX31]], align 4 ; CHECK-MAXBW-NEXT: [[ARRAYIDX40_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX40]], align 4 ; CHECK-MAXBW-NEXT: [[ARRAYIDX49_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX49]], align 4 ; CHECK-MAXBW-NEXT: [[ARRAYIDX58_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX58]], align 4 ; CHECK-MAXBW-NEXT: [[ARRAYIDX67_PROMOTED:%.*]] = load i32, ptr [[ARRAYIDX67]], align 4 ; CHECK-MAXBW-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[N]] to i64 ; CHECK-MAXBW-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 16 ; CHECK-MAXBW-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] ; CHECK-MAXBW: vector.ph: ; CHECK-MAXBW-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 16 ; CHECK-MAXBW-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]] ; CHECK-MAXBW-NEXT: [[TMP0:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX67_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP1:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX58_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP2:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX49_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP3:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX40_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP4:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX31_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP5:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[GEP_B_12_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP6:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[ARRAYIDX13_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: [[TMP7:%.*]] = insertelement <4 x i32> zeroinitializer, i32 [[SUM_PROMOTED]], i32 0 ; CHECK-MAXBW-NEXT: br label [[VECTOR_BODY:%.*]] ; CHECK-MAXBW: vector.body: ; CHECK-MAXBW-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI:%.*]] = phi <4 x i32> [ [[TMP0]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE21:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI1:%.*]] = phi <4 x i32> [ [[TMP1]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE20:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI2:%.*]] = phi <4 x i32> [ [[TMP2]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE19:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI3:%.*]] = phi <4 x i32> [ [[TMP3]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE18:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI4:%.*]] = phi <4 x i32> [ [[TMP4]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE17:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI5:%.*]] = phi <4 x i32> [ [[TMP5]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE16:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI6:%.*]] = phi <4 x i32> [ [[TMP6]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE15:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[VEC_PHI7:%.*]] = phi <4 x i32> [ [[TMP7]], [[VECTOR_PH]] ], [ [[PARTIAL_REDUCE:%.*]], [[VECTOR_BODY]] ] ; CHECK-MAXBW-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw i8, ptr [[A]], i64 [[INDEX]] ; CHECK-MAXBW-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1 ; CHECK-MAXBW-NEXT: [[TMP11:%.*]] = shl nsw i64 [[INDEX]], 3 ; CHECK-MAXBW-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw i8, ptr [[B]], i64 [[TMP11]] ; CHECK-MAXBW-NEXT: [[WIDE_VEC:%.*]] = load <128 x i8>, ptr [[TMP12]], align 1 ; CHECK-MAXBW-NEXT: [[STRIDED_VEC:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC8:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC9:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC10:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC11:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC12:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC13:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[STRIDED_VEC14:%.*]] = shufflevector <128 x i8> [[WIDE_VEC]], <128 x i8> poison, <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP35:%.*]] = sext <16 x i8> [[STRIDED_VEC]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP44:%.*]] = sext <16 x i8> [[WIDE_LOAD]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP13:%.*]] = mul nsw <16 x i32> [[TMP35]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI7]], <16 x i32> [[TMP13]]) ; CHECK-MAXBW-NEXT: [[TMP14:%.*]] = sext <16 x i8> [[STRIDED_VEC8]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP16:%.*]] = mul nsw <16 x i32> [[TMP14]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE15]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI6]], <16 x i32> [[TMP16]]) ; CHECK-MAXBW-NEXT: [[TMP17:%.*]] = sext <16 x i8> [[STRIDED_VEC9]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP19:%.*]] = mul nsw <16 x i32> [[TMP17]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE16]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI5]], <16 x i32> [[TMP19]]) ; CHECK-MAXBW-NEXT: [[TMP20:%.*]] = sext <16 x i8> [[STRIDED_VEC10]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP22:%.*]] = mul nsw <16 x i32> [[TMP20]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE17]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI4]], <16 x i32> [[TMP22]]) ; CHECK-MAXBW-NEXT: [[TMP23:%.*]] = sext <16 x i8> [[STRIDED_VEC11]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP25:%.*]] = mul nsw <16 x i32> [[TMP23]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE18]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI3]], <16 x i32> [[TMP25]]) ; CHECK-MAXBW-NEXT: [[TMP26:%.*]] = sext <16 x i8> [[STRIDED_VEC12]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP28:%.*]] = mul nsw <16 x i32> [[TMP26]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE19]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI2]], <16 x i32> [[TMP28]]) ; CHECK-MAXBW-NEXT: [[TMP45:%.*]] = sext <16 x i8> [[STRIDED_VEC13]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP31:%.*]] = mul nsw <16 x i32> [[TMP45]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE20]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI1]], <16 x i32> [[TMP31]]) ; CHECK-MAXBW-NEXT: [[TMP32:%.*]] = sext <16 x i8> [[STRIDED_VEC14]] to <16 x i32> ; CHECK-MAXBW-NEXT: [[TMP34:%.*]] = mul nsw <16 x i32> [[TMP32]], [[TMP44]] ; CHECK-MAXBW-NEXT: [[PARTIAL_REDUCE21]] = call <4 x i32> @llvm.vector.partial.reduce.add.v4i32.v16i32(<4 x i32> [[VEC_PHI]], <16 x i32> [[TMP34]]) ; CHECK-MAXBW-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16 ; CHECK-MAXBW-NEXT: [[TMP29:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[TMP29]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP24:![0-9]+]] ; CHECK-MAXBW: middle.block: ; CHECK-MAXBW-NEXT: [[TMP36:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE21]]) ; CHECK-MAXBW-NEXT: [[TMP37:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE20]]) ; CHECK-MAXBW-NEXT: [[TMP38:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE19]]) ; CHECK-MAXBW-NEXT: [[TMP39:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE18]]) ; CHECK-MAXBW-NEXT: [[TMP40:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE17]]) ; CHECK-MAXBW-NEXT: [[TMP41:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE16]]) ; CHECK-MAXBW-NEXT: [[TMP42:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE15]]) ; CHECK-MAXBW-NEXT: [[TMP43:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[PARTIAL_REDUCE]]) ; CHECK-MAXBW-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]] ; CHECK-MAXBW-NEXT: br i1 [[CMP_N]], label [[FOR_COND_FOR_COND_CLEANUP_CRIT_EDGE:%.*]], label [[SCALAR_PH]] ; CHECK-MAXBW: scalar.ph: ; entry: %cmp100 = icmp sgt i32 %n, 0 br i1 %cmp100, label %for.body.lr.ph, label %for.cond.cleanup for.body.lr.ph: ; preds = %entry %arrayidx13 = getelementptr inbounds nuw i8, ptr %sum, i64 4 %gep.b.12 = getelementptr inbounds nuw i8, ptr %sum, i64 8 %arrayidx31 = getelementptr inbounds nuw i8, ptr %sum, i64 12 %arrayidx40 = getelementptr inbounds nuw i8, ptr %sum, i64 16 %arrayidx49 = getelementptr inbounds nuw i8, ptr %sum, i64 20 %arrayidx58 = getelementptr inbounds nuw i8, ptr %sum, i64 24 %arrayidx67 = getelementptr inbounds nuw i8, ptr %sum, i64 28 %sum.promoted = load i32, ptr %sum, align 4 %arrayidx13.promoted = load i32, ptr %arrayidx13, align 4 %gep.b.12.promoted = load i32, ptr %gep.b.12, align 4 %arrayidx31.promoted = load i32, ptr %arrayidx31, align 4 %arrayidx40.promoted = load i32, ptr %arrayidx40, align 4 %arrayidx49.promoted = load i32, ptr %arrayidx49, align 4 %arrayidx58.promoted = load i32, ptr %arrayidx58, align 4 %arrayidx67.promoted = load i32, ptr %arrayidx67, align 4 %wide.trip.count = zext nneg i32 %n to i64 br label %for.body for.cond.for.cond.cleanup_crit_edge: ; preds = %for.body %add.lcssa = phi i32 [ %add.1, %for.body ] %add.2.lcssa = phi i32 [ %add.2, %for.body ] %add.3.lcssa = phi i32 [ %add.3, %for.body ] %add.4.lcssa = phi i32 [ %add.4, %for.body ] %add.5.lcssa = phi i32 [ %add.5, %for.body ] %add.6.lcssa = phi i32 [ %add.6, %for.body ] %add.7.lcssa = phi i32 [ %add.7, %for.body ] %add.8.lcssa = phi i32 [ %add.8, %for.body ] store i32 %add.lcssa, ptr %sum, align 4 store i32 %add.2.lcssa, ptr %arrayidx13, align 4 store i32 %add.3.lcssa, ptr %gep.b.12, align 4 store i32 %add.4.lcssa, ptr %arrayidx31, align 4 store i32 %add.5.lcssa, ptr %arrayidx40, align 4 store i32 %add.6.lcssa, ptr %arrayidx49, align 4 store i32 %add.7.lcssa, ptr %arrayidx58, align 4 store i32 %add.8.lcssa, ptr %arrayidx67, align 4 br label %for.cond.cleanup for.cond.cleanup: ; preds = %for.cond.for.cond.cleanup_crit_edge, %entry ret void for.body: ; preds = %for.body.lr.ph, %for.body %indvars.iv = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next, %for.body ] %0 = phi i32 [ %arrayidx67.promoted, %for.body.lr.ph ], [ %add.8, %for.body ] %1 = phi i32 [ %arrayidx58.promoted, %for.body.lr.ph ], [ %add.7, %for.body ] %2 = phi i32 [ %arrayidx49.promoted, %for.body.lr.ph ], [ %add.6, %for.body ] %3 = phi i32 [ %arrayidx40.promoted, %for.body.lr.ph ], [ %add.5, %for.body ] %4 = phi i32 [ %arrayidx31.promoted, %for.body.lr.ph ], [ %add.4, %for.body ] %5 = phi i32 [ %gep.b.12.promoted, %for.body.lr.ph ], [ %add.3, %for.body ] %6 = phi i32 [ %arrayidx13.promoted, %for.body.lr.ph ], [ %add.2, %for.body ] %7 = phi i32 [ %sum.promoted, %for.body.lr.ph ], [ %add.1, %for.body ] %arrayidx = getelementptr inbounds nuw i8, ptr %a, i64 %indvars.iv %load.a = load i8, ptr %arrayidx, align 1 %ext.a = sext i8 %load.a to i32 %9 = shl nsw i64 %indvars.iv, 3 %gep.b.1 = getelementptr inbounds nuw i8, ptr %b, i64 %9 %load.b.1 = load i8, ptr %gep.b.1, align 1 %ext.b.1 = sext i8 %load.b.1 to i32 %mul.1 = mul nsw i32 %ext.b.1, %ext.a %add.1 = add nsw i32 %mul.1, %7 %11 = or disjoint i64 %9, 1 %gep.b.2 = getelementptr inbounds nuw i8, ptr %b, i64 %11 %load.b.2 = load i8, ptr %gep.b.2, align 1 %ext.b.2 = sext i8 %load.b.2 to i32 %mul.2 = mul nsw i32 %ext.b.2, %ext.a %add.2 = add nsw i32 %mul.2, %6 %13 = or disjoint i64 %9, 2 %gep.b.3 = getelementptr inbounds nuw i8, ptr %b, i64 %13 %load.b.3 = load i8, ptr %gep.b.3, align 1 %ext.b.3 = sext i8 %load.b.3 to i32 %mul.3 = mul nsw i32 %ext.b.3, %ext.a %add.3 = add nsw i32 %mul.3, %5 %15 = or disjoint i64 %9, 3 %gep.b.4 = getelementptr inbounds nuw i8, ptr %b, i64 %15 %load.b.4 = load i8, ptr %gep.b.4, align 1 %ext.b.4 = sext i8 %load.b.4 to i32 %mul.4 = mul nsw i32 %ext.b.4, %ext.a %add.4 = add nsw i32 %mul.4, %4 %17 = or disjoint i64 %9, 4 %gep.b.5 = getelementptr inbounds nuw i8, ptr %b, i64 %17 %load.b.5 = load i8, ptr %gep.b.5, align 1 %ext.b.5 = sext i8 %load.b.5 to i32 %mul.5 = mul nsw i32 %ext.b.5, %ext.a %add.5 = add nsw i32 %mul.5, %3 %19 = or disjoint i64 %9, 5 %gep.b.6 = getelementptr inbounds nuw i8, ptr %b, i64 %19 %load.b.6 = load i8, ptr %gep.b.6, align 1 %ext.b.6 = sext i8 %load.b.6 to i32 %mul.6 = mul nsw i32 %ext.b.6, %ext.a %add.6 = add nsw i32 %mul.6, %2 %21 = or disjoint i64 %9, 6 %gep.b.7 = getelementptr inbounds nuw i8, ptr %b, i64 %21 %load.b.7 = load i8, ptr %gep.b.7, align 1 %ext.b.7 = sext i8 %load.b.7 to i32 %mul.7 = mul nsw i32 %ext.b.7, %ext.a %add.7 = add nsw i32 %mul.7, %1 %23 = or disjoint i64 %9, 7 %gep.b.8 = getelementptr inbounds nuw i8, ptr %b, i64 %23 %load.b.8 = load i8, ptr %gep.b.8, align 1 %ext.b.8 = sext i8 %load.b.8 to i32 %mul.8 = mul nsw i32 %ext.b.8, %ext.a %add.8 = add nsw i32 %mul.8, %0 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count br i1 %exitcond.not, label %for.cond.for.cond.cleanup_crit_edge, label %for.body, !llvm.loop !8 } !7 = distinct !{!7, !8, !9, !10} !8 = !{!"llvm.loop.mustprogress"} !9 = !{!"llvm.loop.vectorize.predicate.enable", i1 true} !10 = !{!"llvm.loop.vectorize.enable", i1 true} attributes #0 = { vscale_range(1,16) "target-features"="+sve" } attributes #1 = { vscale_range(1,16) "target-features"="+neon,+dotprod,+sve" "target-cpu"="neoverse-v2" }