| ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py |
| ; RUN: opt < %s -passes=instcombine -S -data-layout="E-n64" | FileCheck %s --check-prefixes=ALL,BE |
| ; RUN: opt < %s -passes=instcombine -S -data-layout="e-n64" | FileCheck %s --check-prefixes=ALL,LE |
| |
| declare void @use(<2 x i8>) |
| |
| ; i16 is a common type, so we can convert independently of the data layout. |
| ; Endian determines if a shift is needed (and so the transform is avoided). |
| |
| define i16 @insert0_v2i8(i16 %x, i8 %y) { |
| ; BE-LABEL: @insert0_v2i8( |
| ; BE-NEXT: [[V:%.*]] = bitcast i16 [[X:%.*]] to <2 x i8> |
| ; BE-NEXT: [[I:%.*]] = insertelement <2 x i8> [[V]], i8 [[Y:%.*]], i64 0 |
| ; BE-NEXT: [[R:%.*]] = bitcast <2 x i8> [[I]] to i16 |
| ; BE-NEXT: ret i16 [[R]] |
| ; |
| ; LE-LABEL: @insert0_v2i8( |
| ; LE-NEXT: [[TMP1:%.*]] = and i16 [[X:%.*]], -256 |
| ; LE-NEXT: [[TMP2:%.*]] = zext i8 [[Y:%.*]] to i16 |
| ; LE-NEXT: [[R:%.*]] = or i16 [[TMP1]], [[TMP2]] |
| ; LE-NEXT: ret i16 [[R]] |
| ; |
| %v = bitcast i16 %x to <2 x i8> |
| %i = insertelement <2 x i8> %v, i8 %y, i8 0 |
| %r = bitcast <2 x i8> %i to i16 |
| ret i16 %r |
| } |
| |
| ; i16 is a common type, so we can convert independently of the data layout. |
| ; Endian determines if a shift is needed (and so the transform is avoided). |
| |
| define i16 @insert1_v2i8(i16 %x, i8 %y) { |
| ; BE-LABEL: @insert1_v2i8( |
| ; BE-NEXT: [[TMP1:%.*]] = and i16 [[X:%.*]], -256 |
| ; BE-NEXT: [[TMP2:%.*]] = zext i8 [[Y:%.*]] to i16 |
| ; BE-NEXT: [[R:%.*]] = or i16 [[TMP1]], [[TMP2]] |
| ; BE-NEXT: ret i16 [[R]] |
| ; |
| ; LE-LABEL: @insert1_v2i8( |
| ; LE-NEXT: [[V:%.*]] = bitcast i16 [[X:%.*]] to <2 x i8> |
| ; LE-NEXT: [[I:%.*]] = insertelement <2 x i8> [[V]], i8 [[Y:%.*]], i64 1 |
| ; LE-NEXT: [[R:%.*]] = bitcast <2 x i8> [[I]] to i16 |
| ; LE-NEXT: ret i16 [[R]] |
| ; |
| %v = bitcast i16 %x to <2 x i8> |
| %i = insertelement <2 x i8> %v, i8 %y, i8 1 |
| %r = bitcast <2 x i8> %i to i16 |
| ret i16 %r |
| } |
| |
| ; i32 is a common type, so we can convert independently of the data layout. |
| ; Endian determines if a shift is needed (and so the transform is avoided). |
| |
| define i32 @insert0_v4i8(i32 %x, i8 %y) { |
| ; BE-LABEL: @insert0_v4i8( |
| ; BE-NEXT: [[V:%.*]] = bitcast i32 [[X:%.*]] to <4 x i8> |
| ; BE-NEXT: [[I:%.*]] = insertelement <4 x i8> [[V]], i8 [[Y:%.*]], i64 0 |
| ; BE-NEXT: [[R:%.*]] = bitcast <4 x i8> [[I]] to i32 |
| ; BE-NEXT: ret i32 [[R]] |
| ; |
| ; LE-LABEL: @insert0_v4i8( |
| ; LE-NEXT: [[TMP1:%.*]] = and i32 [[X:%.*]], -256 |
| ; LE-NEXT: [[TMP2:%.*]] = zext i8 [[Y:%.*]] to i32 |
| ; LE-NEXT: [[R:%.*]] = or i32 [[TMP1]], [[TMP2]] |
| ; LE-NEXT: ret i32 [[R]] |
| ; |
| %v = bitcast i32 %x to <4 x i8> |
| %i = insertelement <4 x i8> %v, i8 %y, i8 0 |
| %r = bitcast <4 x i8> %i to i32 |
| ret i32 %r |
| } |
| |
| ; i32 is a common type, so we can convert independently of the data layout. |
| ; Endian determines if a shift is needed (and so the transform is avoided). |
| ; half type can not be used in zext instruction (and so the transform is avoided). |
| |
| define i32 @insert0_v2half(i32 %x, half %y) { |
| ; ALL-LABEL: @insert0_v2half( |
| ; ALL-NEXT: [[V:%.*]] = bitcast i32 [[X:%.*]] to <2 x half> |
| ; ALL-NEXT: [[I:%.*]] = insertelement <2 x half> [[V]], half [[Y:%.*]], i64 0 |
| ; ALL-NEXT: [[R:%.*]] = bitcast <2 x half> [[I]] to i32 |
| ; ALL-NEXT: ret i32 [[R]] |
| ; |
| %v = bitcast i32 %x to <2 x half> |
| %i = insertelement <2 x half> %v, half %y, i8 0 |
| %r = bitcast <2 x half> %i to i32 |
| ret i32 %r |
| } |
| |
| ; i64 is a legal type, so we can convert based on the data layout. |
| ; Endian determines if a shift is needed (and so the transform is avoided). |
| |
| define i64 @insert0_v4i16(i64 %x, i16 %y) { |
| ; BE-LABEL: @insert0_v4i16( |
| ; BE-NEXT: [[V:%.*]] = bitcast i64 [[X:%.*]] to <4 x i16> |
| ; BE-NEXT: [[I:%.*]] = insertelement <4 x i16> [[V]], i16 [[Y:%.*]], i64 0 |
| ; BE-NEXT: [[R:%.*]] = bitcast <4 x i16> [[I]] to i64 |
| ; BE-NEXT: ret i64 [[R]] |
| ; |
| ; LE-LABEL: @insert0_v4i16( |
| ; LE-NEXT: [[TMP1:%.*]] = and i64 [[X:%.*]], -65536 |
| ; LE-NEXT: [[TMP2:%.*]] = zext i16 [[Y:%.*]] to i64 |
| ; LE-NEXT: [[R:%.*]] = or i64 [[TMP1]], [[TMP2]] |
| ; LE-NEXT: ret i64 [[R]] |
| ; |
| %v = bitcast i64 %x to <4 x i16> |
| %i = insertelement <4 x i16> %v, i16 %y, i8 0 |
| %r = bitcast <4 x i16> %i to i64 |
| ret i64 %r |
| } |
| |
| ; Negative test - shifts needed for both endians. |
| |
| define i64 @insert1_v4i16(i64 %x, i16 %y) { |
| ; ALL-LABEL: @insert1_v4i16( |
| ; ALL-NEXT: [[V:%.*]] = bitcast i64 [[X:%.*]] to <4 x i16> |
| ; ALL-NEXT: [[I:%.*]] = insertelement <4 x i16> [[V]], i16 [[Y:%.*]], i64 1 |
| ; ALL-NEXT: [[R:%.*]] = bitcast <4 x i16> [[I]] to i64 |
| ; ALL-NEXT: ret i64 [[R]] |
| ; |
| %v = bitcast i64 %x to <4 x i16> |
| %i = insertelement <4 x i16> %v, i16 %y, i8 1 |
| %r = bitcast <4 x i16> %i to i64 |
| ret i64 %r |
| } |
| |
| ; i64 is a legal type, so we can convert based on the data layout. |
| ; Endian determines if a shift is needed (and so the transform is avoided). |
| |
| define i64 @insert3_v4i16(i64 %x, i16 %y) { |
| ; BE-LABEL: @insert3_v4i16( |
| ; BE-NEXT: [[TMP1:%.*]] = and i64 [[X:%.*]], -65536 |
| ; BE-NEXT: [[TMP2:%.*]] = zext i16 [[Y:%.*]] to i64 |
| ; BE-NEXT: [[R:%.*]] = or i64 [[TMP1]], [[TMP2]] |
| ; BE-NEXT: ret i64 [[R]] |
| ; |
| ; LE-LABEL: @insert3_v4i16( |
| ; LE-NEXT: [[V:%.*]] = bitcast i64 [[X:%.*]] to <4 x i16> |
| ; LE-NEXT: [[I:%.*]] = insertelement <4 x i16> [[V]], i16 [[Y:%.*]], i64 3 |
| ; LE-NEXT: [[R:%.*]] = bitcast <4 x i16> [[I]] to i64 |
| ; LE-NEXT: ret i64 [[R]] |
| ; |
| %v = bitcast i64 %x to <4 x i16> |
| %i = insertelement <4 x i16> %v, i16 %y, i8 3 |
| %r = bitcast <4 x i16> %i to i64 |
| ret i64 %r |
| } |
| |
| ; Negative test - i128 is not a legal type, so we do not convert based on the data layout. |
| |
| define i128 @insert0_v4i32(i128 %x, i32 %y) { |
| ; ALL-LABEL: @insert0_v4i32( |
| ; ALL-NEXT: [[V:%.*]] = bitcast i128 [[X:%.*]] to <4 x i32> |
| ; ALL-NEXT: [[I:%.*]] = insertelement <4 x i32> [[V]], i32 [[Y:%.*]], i64 0 |
| ; ALL-NEXT: [[R:%.*]] = bitcast <4 x i32> [[I]] to i128 |
| ; ALL-NEXT: ret i128 [[R]] |
| ; |
| %v = bitcast i128 %x to <4 x i32> |
| %i = insertelement <4 x i32> %v, i32 %y, i8 0 |
| %r = bitcast <4 x i32> %i to i128 |
| ret i128 %r |
| } |
| |
| ; Negative test - extra use requires more instructions. |
| |
| define i16 @insert0_v2i8_use1(i16 %x, i8 %y) { |
| ; ALL-LABEL: @insert0_v2i8_use1( |
| ; ALL-NEXT: [[V:%.*]] = bitcast i16 [[X:%.*]] to <2 x i8> |
| ; ALL-NEXT: call void @use(<2 x i8> [[V]]) |
| ; ALL-NEXT: [[I:%.*]] = insertelement <2 x i8> [[V]], i8 [[Y:%.*]], i64 0 |
| ; ALL-NEXT: [[R:%.*]] = bitcast <2 x i8> [[I]] to i16 |
| ; ALL-NEXT: ret i16 [[R]] |
| ; |
| %v = bitcast i16 %x to <2 x i8> |
| call void @use(<2 x i8> %v) |
| %i = insertelement <2 x i8> %v, i8 %y, i8 0 |
| %r = bitcast <2 x i8> %i to i16 |
| ret i16 %r |
| } |
| |
| ; Negative test - extra use requires more instructions. |
| |
| define i16 @insert0_v2i8_use2(i16 %x, i8 %y) { |
| ; ALL-LABEL: @insert0_v2i8_use2( |
| ; ALL-NEXT: [[V:%.*]] = bitcast i16 [[X:%.*]] to <2 x i8> |
| ; ALL-NEXT: [[I:%.*]] = insertelement <2 x i8> [[V]], i8 [[Y:%.*]], i64 0 |
| ; ALL-NEXT: call void @use(<2 x i8> [[I]]) |
| ; ALL-NEXT: [[R:%.*]] = bitcast <2 x i8> [[I]] to i16 |
| ; ALL-NEXT: ret i16 [[R]] |
| ; |
| %v = bitcast i16 %x to <2 x i8> |
| %i = insertelement <2 x i8> %v, i8 %y, i8 0 |
| call void @use(<2 x i8> %i) |
| %r = bitcast <2 x i8> %i to i16 |
| ret i16 %r |
| } |