; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6 ; RUN: opt -passes=vector-combine -mtriple=arm64-apple-darwinos -S %s | FileCheck %s define i32 @load_v4i8_bitcast_to_i32(ptr %x) { ; CHECK-LABEL: define i32 @load_v4i8_bitcast_to_i32( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[R_SCALAR:%.*]] = load i32, ptr [[X]], align 4 ; CHECK-NEXT: ret i32 [[R_SCALAR]] ; %lv = load <4 x i8>, ptr %x %r = bitcast <4 x i8> %lv to i32 ret i32 %r } define i64 @load_v2i32_bitcast_to_i64(ptr %x) { ; CHECK-LABEL: define i64 @load_v2i32_bitcast_to_i64( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[R_SCALAR:%.*]] = load i64, ptr [[X]], align 8 ; CHECK-NEXT: ret i64 [[R_SCALAR]] ; %lv = load <2 x i32>, ptr %x %r = bitcast <2 x i32> %lv to i64 ret i64 %r } define float @load_v4i8_bitcast_to_float(ptr %x) { ; CHECK-LABEL: define float @load_v4i8_bitcast_to_float( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[R_SCALAR:%.*]] = load float, ptr [[X]], align 4 ; CHECK-NEXT: ret float [[R_SCALAR]] ; %lv = load <4 x i8>, ptr %x %r = bitcast <4 x i8> %lv to float ret float %r } define float @load_v2i16_bitcast_to_float(ptr %x) { ; CHECK-LABEL: define float @load_v2i16_bitcast_to_float( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[R_SCALAR:%.*]] = load float, ptr [[X]], align 4 ; CHECK-NEXT: ret float [[R_SCALAR]] ; %lv = load <2 x i16>, ptr %x %r = bitcast <2 x i16> %lv to float ret float %r } define double @load_v4i16_bitcast_to_double(ptr %x) { ; CHECK-LABEL: define double @load_v4i16_bitcast_to_double( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[LV:%.*]] = load <4 x i16>, ptr [[X]], align 8 ; CHECK-NEXT: [[R_SCALAR:%.*]] = bitcast <4 x i16> [[LV]] to double ; CHECK-NEXT: ret double [[R_SCALAR]] ; %lv = load <4 x i16>, ptr %x %r = bitcast <4 x i16> %lv to double ret double %r } define double @load_v2i32_bitcast_to_double(ptr %x) { ; CHECK-LABEL: define double @load_v2i32_bitcast_to_double( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[LV:%.*]] = load <2 x i32>, ptr [[X]], align 8 ; CHECK-NEXT: [[R_SCALAR:%.*]] = bitcast <2 x i32> [[LV]] to double ; CHECK-NEXT: ret double [[R_SCALAR]] ; %lv = load <2 x i32>, ptr %x %r = bitcast <2 x i32> %lv to double ret double %r } ; Multiple users with the same bitcast type should be scalarized. define i32 @load_v4i8_bitcast_multiple_users_same_type(ptr %x) { ; CHECK-LABEL: define i32 @load_v4i8_bitcast_multiple_users_same_type( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[LV_SCALAR:%.*]] = load i32, ptr [[X]], align 4 ; CHECK-NEXT: [[ADD:%.*]] = add i32 [[LV_SCALAR]], [[LV_SCALAR]] ; CHECK-NEXT: ret i32 [[ADD]] ; %lv = load <4 x i8>, ptr %x %r1 = bitcast <4 x i8> %lv to i32 %r2 = bitcast <4 x i8> %lv to i32 %add = add i32 %r1, %r2 ret i32 %add } ; Different bitcast types should not be scalarized. define i32 @load_v4i8_bitcast_multiple_users_different_types(ptr %x) { ; CHECK-LABEL: define i32 @load_v4i8_bitcast_multiple_users_different_types( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[LV:%.*]] = load <4 x i8>, ptr [[X]], align 4 ; CHECK-NEXT: [[R1:%.*]] = bitcast <4 x i8> [[LV]] to i32 ; CHECK-NEXT: [[R2:%.*]] = bitcast <4 x i8> [[LV]] to float ; CHECK-NEXT: [[R2_INT:%.*]] = bitcast float [[R2]] to i32 ; CHECK-NEXT: [[ADD:%.*]] = add i32 [[R1]], [[R2_INT]] ; CHECK-NEXT: ret i32 [[ADD]] ; %lv = load <4 x i8>, ptr %x %r1 = bitcast <4 x i8> %lv to i32 %r2 = bitcast <4 x i8> %lv to float %r2.int = bitcast float %r2 to i32 %add = add i32 %r1, %r2.int ret i32 %add } ; Bitcast to vector should not be scalarized. define <2 x i16> @load_v4i8_bitcast_to_vector(ptr %x) { ; CHECK-LABEL: define <2 x i16> @load_v4i8_bitcast_to_vector( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[LV:%.*]] = load <4 x i8>, ptr [[X]], align 4 ; CHECK-NEXT: [[R:%.*]] = bitcast <4 x i8> [[LV]] to <2 x i16> ; CHECK-NEXT: ret <2 x i16> [[R]] ; %lv = load <4 x i8>, ptr %x %r = bitcast <4 x i8> %lv to <2 x i16> ret <2 x i16> %r } ; Load with both bitcast users and other users should not be scalarized. define i32 @load_v4i8_mixed_users(ptr %x) { ; CHECK-LABEL: define i32 @load_v4i8_mixed_users( ; CHECK-SAME: ptr [[X:%.*]]) { ; CHECK-NEXT: [[LV:%.*]] = load <4 x i8>, ptr [[X]], align 4 ; CHECK-NEXT: [[R1:%.*]] = bitcast <4 x i8> [[LV]] to i32 ; CHECK-NEXT: [[R2:%.*]] = extractelement <4 x i8> [[LV]], i32 0 ; CHECK-NEXT: [[R2_EXT:%.*]] = zext i8 [[R2]] to i32 ; CHECK-NEXT: [[ADD:%.*]] = add i32 [[R1]], [[R2_EXT]] ; CHECK-NEXT: ret i32 [[ADD]] ; %lv = load <4 x i8>, ptr %x %r1 = bitcast <4 x i8> %lv to i32 %r2 = extractelement <4 x i8> %lv, i32 0 %r2.ext = zext i8 %r2 to i32 %add = add i32 %r1, %r2.ext ret i32 %add }