When extracting a smaller integer from a scalar_to_vector source, we were limited to only folding/truncating the lowest bits of the scalar source. This patch extends the fold to handle extraction of any other element, by right shifting the source before truncation. Fixes a regression from #117884
122 lines
3.4 KiB
LLVM
122 lines
3.4 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple=aarch64_be-- < %s | FileCheck %s --check-prefix=BE
|
|
; RUN: llc -mtriple=aarch64-- < %s | FileCheck %s --check-prefix=LE
|
|
|
|
define i32 @trunc_i64_to_i32_le(i64 %x) {
|
|
; BE-LABEL: trunc_i64_to_i32_le:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: lsr x0, x0, #32
|
|
; BE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i64_to_i32_le:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <2 x i64> undef, i64 %x, i32 0
|
|
%bc = bitcast <2 x i64> %ins to <4 x i32>
|
|
%ext = extractelement <4 x i32> %bc, i32 0
|
|
ret i32 %ext
|
|
}
|
|
|
|
define i32 @trunc_i64_to_i32_be(i64 %x) {
|
|
; BE-LABEL: trunc_i64_to_i32_be:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i64_to_i32_be:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: lsr x0, x0, #32
|
|
; LE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <2 x i64> undef, i64 %x, i32 0
|
|
%bc = bitcast <2 x i64> %ins to <4 x i32>
|
|
%ext = extractelement <4 x i32> %bc, i32 1
|
|
ret i32 %ext
|
|
}
|
|
|
|
define i16 @trunc_i64_to_i16_le(i64 %x) {
|
|
; BE-LABEL: trunc_i64_to_i16_le:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: lsr x0, x0, #48
|
|
; BE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i64_to_i16_le:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <2 x i64> undef, i64 %x, i32 0
|
|
%bc = bitcast <2 x i64> %ins to <8 x i16>
|
|
%ext = extractelement <8 x i16> %bc, i32 0
|
|
ret i16 %ext
|
|
}
|
|
|
|
define i16 @trunc_i64_to_i16_be(i64 %x) {
|
|
; BE-LABEL: trunc_i64_to_i16_be:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i64_to_i16_be:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: lsr x0, x0, #48
|
|
; LE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <2 x i64> undef, i64 %x, i32 0
|
|
%bc = bitcast <2 x i64> %ins to <8 x i16>
|
|
%ext = extractelement <8 x i16> %bc, i32 3
|
|
ret i16 %ext
|
|
}
|
|
|
|
define i8 @trunc_i32_to_i8_le(i32 %x) {
|
|
; BE-LABEL: trunc_i32_to_i8_le:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: lsr w0, w0, #24
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i32_to_i8_le:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <4 x i32> undef, i32 %x, i32 0
|
|
%bc = bitcast <4 x i32> %ins to <16 x i8>
|
|
%ext = extractelement <16 x i8> %bc, i32 0
|
|
ret i8 %ext
|
|
}
|
|
|
|
define i8 @trunc_i32_to_i8_be(i32 %x) {
|
|
; BE-LABEL: trunc_i32_to_i8_be:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i32_to_i8_be:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: lsr w0, w0, #24
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <4 x i32> undef, i32 %x, i32 0
|
|
%bc = bitcast <4 x i32> %ins to <16 x i8>
|
|
%ext = extractelement <16 x i8> %bc, i32 3
|
|
ret i8 %ext
|
|
}
|
|
|
|
; Weird type (non-power-of-2 vector) is ok.
|
|
|
|
define i8 @trunc_i64_to_i8_be(i64 %x) {
|
|
; BE-LABEL: trunc_i64_to_i8_be:
|
|
; BE: // %bb.0:
|
|
; BE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; BE-NEXT: ret
|
|
;
|
|
; LE-LABEL: trunc_i64_to_i8_be:
|
|
; LE: // %bb.0:
|
|
; LE-NEXT: lsr x0, x0, #56
|
|
; LE-NEXT: // kill: def $w0 killed $w0 killed $x0
|
|
; LE-NEXT: ret
|
|
%ins = insertelement <3 x i64> undef, i64 %x, i32 0
|
|
%bc = bitcast <3 x i64> %ins to <24 x i8>
|
|
%ext = extractelement <24 x i8> %bc, i32 7
|
|
ret i8 %ext
|
|
}
|
|
|