; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -slp-vectorizer -S %s | FileCheck %s
; RUN: opt -aa-pipeline=basic-aa -passes='slp-vectorizer' -S %s | FileCheck %s
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
target triple = "arm64-apple-ios5.0.0"
define void @select_umin_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_umin_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp ult <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp ult i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp ult i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp ult i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp ult i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp ult i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp ult i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp ult i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp ult i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_umin_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_umin_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp ult <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp ult i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp ult i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp ult i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp ult i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
define void @select_ule_ugt_mix_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_ule_ugt_mix_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[L_0:%.*]] = load i32, i32* [[PTR:%.*]], align 4
; CHECK-NEXT: [[CMP_0:%.*]] = icmp ult i32 [[L_0]], 16383
; CHECK-NEXT: [[S_0:%.*]] = select i1 [[CMP_0]], i32 [[L_0]], i32 16383
; CHECK-NEXT: store i32 [[S_0]], i32* [[PTR]], align 4
; CHECK-NEXT: [[GEP_1:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i32 1
; CHECK-NEXT: [[L_1:%.*]] = load i32, i32* [[GEP_1]], align 4
; CHECK-NEXT: [[CMP_1:%.*]] = icmp ugt i32 [[L_1]], 16383
; CHECK-NEXT: [[S_1:%.*]] = select i1 [[CMP_1]], i32 [[L_1]], i32 16383
; CHECK-NEXT: store i32 [[S_1]], i32* [[GEP_1]], align 4
; CHECK-NEXT: [[GEP_2:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i32 2
; CHECK-NEXT: [[L_2:%.*]] = load i32, i32* [[GEP_2]], align 4
; CHECK-NEXT: [[CMP_2:%.*]] = icmp ult i32 [[L_2]], 16383
; CHECK-NEXT: [[S_2:%.*]] = select i1 [[CMP_2]], i32 [[L_2]], i32 16383
; CHECK-NEXT: store i32 [[S_2]], i32* [[GEP_2]], align 4
; CHECK-NEXT: [[GEP_3:%.*]] = getelementptr inbounds i32, i32* [[PTR]], i32 3
; CHECK-NEXT: [[L_3:%.*]] = load i32, i32* [[GEP_3]], align 4
; CHECK-NEXT: [[CMP_3:%.*]] = icmp ugt i32 [[L_3]], 16383
; CHECK-NEXT: [[S_3:%.*]] = select i1 [[CMP_3]], i32 [[L_3]], i32 16383
; CHECK-NEXT: store i32 [[S_3]], i32* [[GEP_3]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp ult i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp ugt i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp ult i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp ugt i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of umin, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_umin_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_umin_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp ult <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp ult i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp ult i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_umin_ule_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_umin_ule_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp ule i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp ule i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp ule i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp ule i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp ule i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp ule i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp ule i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp ule i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_umin_ule_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_umin_ule_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp ule i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp ule i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp ule i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp ule i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of umin, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_umin_ule_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_umin_ule_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp ule <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp ule i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp ule i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_smin_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_smin_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp slt <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp slt i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp slt i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp slt i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp slt i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp slt i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp slt i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp slt i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp slt i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_smin_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_smin_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp slt <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp slt i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp slt i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp slt i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp slt i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of smin, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_smin_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_smin_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp slt <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp slt i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp slt i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_smin_sle_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_smin_sle_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp sle i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp sle i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp sle i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp sle i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp sle i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp sle i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp sle i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp sle i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_smin_sle_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_smin_sle_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp sle i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp sle i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp sle i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp sle i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of smin, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_smin_sle_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_smin_sle_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp sle <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp sle i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp sle i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_umax_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_umax_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp ugt i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp ugt i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp ugt i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp ugt i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp ugt i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp ugt i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp ugt i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp ugt i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_umax_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_umax_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp ugt i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp ugt i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp ugt i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp ugt i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of umax, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_umax_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_umax_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp ugt <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp ugt i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp ugt i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_umax_uge_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_umax_uge_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp uge <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp uge i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp uge i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp uge i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp uge i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp uge i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp uge i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp uge i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp uge i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_umax_uge_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_umax_uge_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp uge <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp uge i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp uge i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp uge i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp uge i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of umax, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_umax_uge_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_umax_uge_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp uge <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp uge i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp uge i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_smax_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_smax_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp sgt i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp sgt i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp sgt i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp sgt i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp sgt i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp sgt i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp sgt i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp sgt i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_smax_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_smax_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp sgt i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp sgt i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp sgt i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp sgt i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of smax, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_smax_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_smax_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp sgt <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp sgt i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp sgt i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}
define void @select_smax_sge_8xi16(i16* %ptr, i16 %x) {
; CHECK-LABEL: @select_smax_sge_8xi16(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i16* [[PTR:%.*]] to <8 x i16>*
; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i16>, <8 x i16>* [[TMP0]], align 2
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <8 x i16> [[TMP1]], <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <8 x i1> [[TMP2]], <8 x i16> [[TMP1]], <8 x i16> <i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383, i16 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i16* [[PTR]] to <8 x i16>*
; CHECK-NEXT: store <8 x i16> [[TMP3]], <8 x i16>* [[TMP4]], align 2
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i16, i16* %ptr
%cmp.0 = icmp sge i16 %l.0, 16383
%s.0 = select i1 %cmp.0, i16 %l.0, i16 16383
store i16 %s.0, i16* %ptr, align 2
%gep.1 = getelementptr inbounds i16, i16* %ptr, i16 1
%l.1 = load i16, i16* %gep.1
%cmp.1 = icmp sge i16 %l.1, 16383
%s.1 = select i1 %cmp.1, i16 %l.1, i16 16383
store i16 %s.1, i16* %gep.1, align 2
%gep.2 = getelementptr inbounds i16, i16* %ptr, i16 2
%l.2 = load i16, i16* %gep.2
%cmp.2 = icmp sge i16 %l.2, 16383
%s.2 = select i1 %cmp.2, i16 %l.2, i16 16383
store i16 %s.2, i16* %gep.2, align 2
%gep.3 = getelementptr inbounds i16, i16* %ptr, i16 3
%l.3 = load i16, i16* %gep.3
%cmp.3 = icmp sge i16 %l.3, 16383
%s.3 = select i1 %cmp.3, i16 %l.3, i16 16383
store i16 %s.3, i16* %gep.3, align 2
%gep.4 = getelementptr inbounds i16, i16* %ptr, i16 4
%l.4 = load i16, i16* %gep.4
%cmp.4 = icmp sge i16 %l.4, 16383
%s.4 = select i1 %cmp.4, i16 %l.4, i16 16383
store i16 %s.4, i16* %gep.4, align 2
%gep.5 = getelementptr inbounds i16, i16* %ptr, i16 5
%l.5 = load i16, i16* %gep.5
%cmp.5 = icmp sge i16 %l.5, 16383
%s.5 = select i1 %cmp.5, i16 %l.5, i16 16383
store i16 %s.5, i16* %gep.5, align 2
%gep.6 = getelementptr inbounds i16, i16* %ptr, i16 6
%l.6 = load i16, i16* %gep.6
%cmp.6 = icmp sge i16 %l.6, 16383
%s.6 = select i1 %cmp.6, i16 %l.6, i16 16383
store i16 %s.6, i16* %gep.6, align 2
%gep.7 = getelementptr inbounds i16, i16* %ptr, i16 7
%l.7 = load i16, i16* %gep.7
%cmp.7 = icmp sge i16 %l.7, 16383
%s.7 = select i1 %cmp.7, i16 %l.7, i16 16383
store i16 %s.7, i16* %gep.7, align 2
ret void
}
define void @select_smax_sge_4xi32(i32* %ptr, i32 %x) {
; CHECK-LABEL: @select_smax_sge_4xi32(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[PTR:%.*]] to <4 x i32>*
; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <4 x i32> [[TMP1]], <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <4 x i1> [[TMP2]], <4 x i32> [[TMP1]], <4 x i32> <i32 16383, i32 16383, i32 16383, i32 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i32* [[PTR]] to <4 x i32>*
; CHECK-NEXT: store <4 x i32> [[TMP3]], <4 x i32>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i32, i32* %ptr
%cmp.0 = icmp sge i32 %l.0, 16383
%s.0 = select i1 %cmp.0, i32 %l.0, i32 16383
store i32 %s.0, i32* %ptr, align 4
%gep.1 = getelementptr inbounds i32, i32* %ptr, i32 1
%l.1 = load i32, i32* %gep.1
%cmp.1 = icmp sge i32 %l.1, 16383
%s.1 = select i1 %cmp.1, i32 %l.1, i32 16383
store i32 %s.1, i32* %gep.1, align 4
%gep.2 = getelementptr inbounds i32, i32* %ptr, i32 2
%l.2 = load i32, i32* %gep.2
%cmp.2 = icmp sge i32 %l.2, 16383
%s.2 = select i1 %cmp.2, i32 %l.2, i32 16383
store i32 %s.2, i32* %gep.2, align 4
%gep.3 = getelementptr inbounds i32, i32* %ptr, i32 3
%l.3 = load i32, i32* %gep.3
%cmp.3 = icmp sge i32 %l.3, 16383
%s.3 = select i1 %cmp.3, i32 %l.3, i32 16383
store i32 %s.3, i32* %gep.3, align 4
ret void
}
; There is no <2 x i64> version of smax, but we can efficiently lower
; compare/select pairs with uniform predicates.
define void @select_smax_sge_2xi64(i64* %ptr, i64 %x) {
; CHECK-LABEL: @select_smax_sge_2xi64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[TMP0:%.*]] = bitcast i64* [[PTR:%.*]] to <2 x i64>*
; CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, <2 x i64>* [[TMP0]], align 8
; CHECK-NEXT: [[TMP2:%.*]] = icmp sge <2 x i64> [[TMP1]], <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP3:%.*]] = select <2 x i1> [[TMP2]], <2 x i64> [[TMP1]], <2 x i64> <i64 16383, i64 16383>
; CHECK-NEXT: [[TMP4:%.*]] = bitcast i64* [[PTR]] to <2 x i64>*
; CHECK-NEXT: store <2 x i64> [[TMP3]], <2 x i64>* [[TMP4]], align 4
; CHECK-NEXT: ret void
;
entry:
%l.0 = load i64, i64* %ptr
%cmp.0 = icmp sge i64 %l.0, 16383
%s.0 = select i1 %cmp.0, i64 %l.0, i64 16383
store i64 %s.0, i64* %ptr, align 4
%gep.1 = getelementptr inbounds i64, i64* %ptr, i64 1
%l.1 = load i64, i64* %gep.1
%cmp.1 = icmp sge i64 %l.1, 16383
%s.1 = select i1 %cmp.1, i64 %l.1, i64 16383
store i64 %s.1, i64* %gep.1, align 4
ret void
}