[llvm] [AArch64][GlobalISel] Support more types for TRUNC (PR #66927)
via llvm-commits
llvm-commits at lists.llvm.org
Tue Sep 26 07:11:33 PDT 2023
================
@@ -0,0 +1,531 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 3
+# RUN: llc -O0 -mtriple=arm64-unknown-unknown -global-isel -run-pass=legalizer -global-isel-abort=2 %s -o - | FileCheck %s
+
+---
+name: xtn_v2i64_v2i8
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $q0
+
+ ; CHECK-LABEL: name: xtn_v2i64_v2i8
+ ; CHECK: liveins: $q0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $q0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(<2 x s32>) = G_TRUNC [[COPY]](<2 x s64>)
+ ; CHECK-NEXT: $d0 = COPY [[TRUNC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %0:_(<2 x s64>) = COPY $q0
+ %1:_(<2 x s8>) = G_TRUNC %0(<2 x s64>)
+ %2:_(<2 x s32>) = G_ANYEXT %1(<2 x s8>)
+ $d0 = COPY %2(<2 x s32>)
+ RET_ReallyLR implicit $d0
+
+...
+
+---
+name: xtn_v2i128_v2i8
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $x0, $x1, $x2, $x3
+
+ ; CHECK-LABEL: name: xtn_v2i128_v2i8
+ ; CHECK: liveins: $x0, $x1, $x2, $x3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x2
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[COPY]](s64), [[COPY1]](s64)
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(<2 x s32>) = G_TRUNC [[BUILD_VECTOR]](<2 x s64>)
+ ; CHECK-NEXT: $d0 = COPY [[TRUNC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(s64) = COPY $x0
+ %2:_(s64) = COPY $x1
+ %3:_(s64) = COPY $x2
+ %4:_(s64) = COPY $x3
+ %5:_(s128) = G_MERGE_VALUES %1(s64), %2(s64)
+ %6:_(s128) = G_MERGE_VALUES %3(s64), %4(s64)
+ %0:_(<2 x s128>) = G_BUILD_VECTOR %5(s128), %6(s128)
+ %7:_(<2 x s8>) = G_TRUNC %0(<2 x s128>)
+ %8:_(<2 x s32>) = G_ANYEXT %7(<2 x s8>)
+ $d0 = COPY %8(<2 x s32>)
+ RET_ReallyLR implicit $d0
+
+...
+
+---
+name: xtn_v2i64_v2i16
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $q0
+
+ ; CHECK-LABEL: name: xtn_v2i64_v2i16
+ ; CHECK: liveins: $q0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<2 x s64>) = COPY $q0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(<2 x s32>) = G_TRUNC [[COPY]](<2 x s64>)
+ ; CHECK-NEXT: $d0 = COPY [[TRUNC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %0:_(<2 x s64>) = COPY $q0
+ %1:_(<2 x s16>) = G_TRUNC %0(<2 x s64>)
+ %2:_(<2 x s32>) = G_ANYEXT %1(<2 x s16>)
+ $d0 = COPY %2(<2 x s32>)
+ RET_ReallyLR implicit $d0
+
+...
+
+---
+name: xtn_v2i128_v2i16
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $x0, $x1, $x2, $x3
+
+ ; CHECK-LABEL: name: xtn_v2i128_v2i16
+ ; CHECK: liveins: $x0, $x1, $x2, $x3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x2
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[COPY]](s64), [[COPY1]](s64)
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(<2 x s32>) = G_TRUNC [[BUILD_VECTOR]](<2 x s64>)
+ ; CHECK-NEXT: $d0 = COPY [[TRUNC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(s64) = COPY $x0
+ %2:_(s64) = COPY $x1
+ %3:_(s64) = COPY $x2
+ %4:_(s64) = COPY $x3
+ %5:_(s128) = G_MERGE_VALUES %1(s64), %2(s64)
+ %6:_(s128) = G_MERGE_VALUES %3(s64), %4(s64)
+ %0:_(<2 x s128>) = G_BUILD_VECTOR %5(s128), %6(s128)
+ %7:_(<2 x s16>) = G_TRUNC %0(<2 x s128>)
+ %8:_(<2 x s32>) = G_ANYEXT %7(<2 x s16>)
+ $d0 = COPY %8(<2 x s32>)
+ RET_ReallyLR implicit $d0
+
+...
+
+---
+name: xtn_v2i128_v2i32
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $x0, $x1, $x2, $x3
+
+ ; CHECK-LABEL: name: xtn_v2i128_v2i32
+ ; CHECK: liveins: $x0, $x1, $x2, $x3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x2
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[COPY]](s64), [[COPY1]](s64)
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(<2 x s32>) = G_TRUNC [[BUILD_VECTOR]](<2 x s64>)
+ ; CHECK-NEXT: $d0 = COPY [[TRUNC]](<2 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(s64) = COPY $x0
+ %2:_(s64) = COPY $x1
+ %3:_(s64) = COPY $x2
+ %4:_(s64) = COPY $x3
+ %5:_(s128) = G_MERGE_VALUES %1(s64), %2(s64)
+ %6:_(s128) = G_MERGE_VALUES %3(s64), %4(s64)
+ %0:_(<2 x s128>) = G_BUILD_VECTOR %5(s128), %6(s128)
+ %7:_(<2 x s32>) = G_TRUNC %0(<2 x s128>)
+ $d0 = COPY %7(<2 x s32>)
+ RET_ReallyLR implicit $d0
+
+...
+
+---
+name: xtn_v2i128_v2i64
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $x0, $x1, $x2, $x3
+
+ ; CHECK-LABEL: name: xtn_v2i128_v2i64
+ ; CHECK: liveins: $x0, $x1, $x2, $x3
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $x0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $x2
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<2 x s64>) = G_BUILD_VECTOR [[COPY]](s64), [[COPY1]](s64)
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(<2 x s64>) = COPY [[BUILD_VECTOR]](<2 x s64>)
+ ; CHECK-NEXT: $q0 = COPY [[COPY2]](<2 x s64>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $q0
+ %1:_(s64) = COPY $x0
+ %2:_(s64) = COPY $x1
+ %3:_(s64) = COPY $x2
+ %4:_(s64) = COPY $x3
+ %5:_(s128) = G_MERGE_VALUES %1(s64), %2(s64)
+ %6:_(s128) = G_MERGE_VALUES %3(s64), %4(s64)
+ %0:_(<2 x s128>) = G_BUILD_VECTOR %5(s128), %6(s128)
+ %7:_(<2 x s64>) = G_TRUNC %0(<2 x s128>)
+ $q0 = COPY %7(<2 x s64>)
+ RET_ReallyLR implicit $q0
+
+...
+
+---
+name: xtn_v3i16_v3i8
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $d0
+
+ ; CHECK-LABEL: name: xtn_v3i16_v3i8
+ ; CHECK: liveins: $d0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s16>) = COPY $d0
+ ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s16), [[UV1:%[0-9]+]]:_(s16), [[UV2:%[0-9]+]]:_(s16), [[UV3:%[0-9]+]]:_(s16) = G_UNMERGE_VALUES [[COPY]](<4 x s16>)
+ ; CHECK-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[UV]](s16)
+ ; CHECK-NEXT: [[ANYEXT1:%[0-9]+]]:_(s32) = G_ANYEXT [[UV1]](s16)
+ ; CHECK-NEXT: [[ANYEXT2:%[0-9]+]]:_(s32) = G_ANYEXT [[UV2]](s16)
+ ; CHECK-NEXT: $w0 = COPY [[ANYEXT]](s32)
+ ; CHECK-NEXT: $w1 = COPY [[ANYEXT1]](s32)
+ ; CHECK-NEXT: $w2 = COPY [[ANYEXT2]](s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0, implicit $w1, implicit $w2
+ %1:_(<4 x s16>) = COPY $d0
+ %2:_(s16), %3:_(s16), %4:_(s16), %5:_(s16) = G_UNMERGE_VALUES %1(<4 x s16>)
+ %0:_(<3 x s16>) = G_BUILD_VECTOR %2(s16), %3(s16), %4(s16)
+ %6:_(<3 x s8>) = G_TRUNC %0(<3 x s16>)
+ %10:_(s8), %11:_(s8), %12:_(s8) = G_UNMERGE_VALUES %6(<3 x s8>)
+ %7:_(s32) = G_ANYEXT %10(s8)
+ %8:_(s32) = G_ANYEXT %11(s8)
+ %9:_(s32) = G_ANYEXT %12(s8)
+ $w0 = COPY %7(s32)
+ $w1 = COPY %8(s32)
+ $w2 = COPY %9(s32)
+ RET_ReallyLR implicit $w0, implicit $w1, implicit $w2
+
+...
+
+---
+name: xtn_v3i32_v3i8
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $q0
+
+ ; CHECK-LABEL: name: xtn_v3i32_v3i8
+ ; CHECK: liveins: $q0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $q0
+ ; CHECK-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32), [[UV2:%[0-9]+]]:_(s32), [[UV3:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](<4 x s32>)
+ ; CHECK-NEXT: $w0 = COPY [[UV]](s32)
+ ; CHECK-NEXT: $w1 = COPY [[UV1]](s32)
+ ; CHECK-NEXT: $w2 = COPY [[UV2]](s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0, implicit $w1, implicit $w2
+ %1:_(<4 x s32>) = COPY $q0
+ %2:_(s32), %3:_(s32), %4:_(s32), %5:_(s32) = G_UNMERGE_VALUES %1(<4 x s32>)
+ %0:_(<3 x s32>) = G_BUILD_VECTOR %2(s32), %3(s32), %4(s32)
+ %6:_(<3 x s8>) = G_TRUNC %0(<3 x s32>)
+ %10:_(s8), %11:_(s8), %12:_(s8) = G_UNMERGE_VALUES %6(<3 x s8>)
+ %7:_(s32) = G_ANYEXT %10(s8)
+ %8:_(s32) = G_ANYEXT %11(s8)
+ %9:_(s32) = G_ANYEXT %12(s8)
+ $w0 = COPY %7(s32)
+ $w1 = COPY %8(s32)
+ $w2 = COPY %9(s32)
+ RET_ReallyLR implicit $w0, implicit $w1, implicit $w2
+
+...
+
+---
+name: xtn_v3i64_v3i8
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $d0, $d1, $d2
+
+ ; CHECK-LABEL: name: xtn_v3i64_v3i8
+ ; CHECK: liveins: $d0, $d1, $d2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $d0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $d1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $d2
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[COPY1]](s64)
+ ; CHECK-NEXT: [[TRUNC2:%[0-9]+]]:_(s32) = G_TRUNC [[COPY2]](s64)
+ ; CHECK-NEXT: $w0 = COPY [[TRUNC]](s32)
+ ; CHECK-NEXT: $w1 = COPY [[TRUNC1]](s32)
+ ; CHECK-NEXT: $w2 = COPY [[TRUNC2]](s32)
+ ; CHECK-NEXT: RET_ReallyLR implicit $w0, implicit $w1, implicit $w2
+ %1:_(s64) = COPY $d0
+ %2:_(s64) = COPY $d1
+ %3:_(s64) = COPY $d2
+ %0:_(<3 x s64>) = G_BUILD_VECTOR %1(s64), %2(s64), %3(s64)
+ %4:_(<3 x s8>) = G_TRUNC %0(<3 x s64>)
+ %8:_(s8), %9:_(s8), %10:_(s8) = G_UNMERGE_VALUES %4(<3 x s8>)
+ %5:_(s32) = G_ANYEXT %8(s8)
+ %6:_(s32) = G_ANYEXT %9(s8)
+ %7:_(s32) = G_ANYEXT %10(s8)
+ $w0 = COPY %5(s32)
+ $w1 = COPY %6(s32)
+ $w2 = COPY %7(s32)
+ RET_ReallyLR implicit $w0, implicit $w1, implicit $w2
+
+...
+
+---
+name: xtn_v3i64_v3i16
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $d0, $d1, $d2
+
+ ; CHECK-LABEL: name: xtn_v3i64_v3i16
+ ; CHECK: liveins: $d0, $d1, $d2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $d0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $d1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $d2
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s64)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:_(s16) = G_TRUNC [[COPY1]](s64)
+ ; CHECK-NEXT: [[TRUNC2:%[0-9]+]]:_(s16) = G_TRUNC [[COPY2]](s64)
+ ; CHECK-NEXT: [[DEF:%[0-9]+]]:_(s16) = G_IMPLICIT_DEF
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s16>) = G_BUILD_VECTOR [[TRUNC]](s16), [[TRUNC1]](s16), [[TRUNC2]](s16), [[DEF]](s16)
+ ; CHECK-NEXT: $d0 = COPY [[BUILD_VECTOR]](<4 x s16>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %1:_(s64) = COPY $d0
+ %2:_(s64) = COPY $d1
+ %3:_(s64) = COPY $d2
+ %0:_(<3 x s64>) = G_BUILD_VECTOR %1(s64), %2(s64), %3(s64)
+ %4:_(<3 x s16>) = G_TRUNC %0(<3 x s64>)
+ %5:_(s16), %6:_(s16), %7:_(s16) = G_UNMERGE_VALUES %4(<3 x s16>)
+ %8:_(s16) = G_IMPLICIT_DEF
+ %9:_(<4 x s16>) = G_BUILD_VECTOR %5(s16), %6(s16), %7(s16), %8(s16)
+ $d0 = COPY %9(<4 x s16>)
+ RET_ReallyLR implicit $d0
+
+...
+
+---
+name: xtn_v3i64_v3i32
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $d0, $d1, $d2
+
+ ; CHECK-LABEL: name: xtn_v3i64_v3i32
+ ; CHECK: liveins: $d0, $d1, $d2
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $d0
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $d1
+ ; CHECK-NEXT: [[COPY2:%[0-9]+]]:_(s64) = COPY $d2
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[COPY]](s64)
+ ; CHECK-NEXT: [[TRUNC1:%[0-9]+]]:_(s32) = G_TRUNC [[COPY1]](s64)
+ ; CHECK-NEXT: [[TRUNC2:%[0-9]+]]:_(s32) = G_TRUNC [[COPY2]](s64)
+ ; CHECK-NEXT: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
+ ; CHECK-NEXT: [[BUILD_VECTOR:%[0-9]+]]:_(<4 x s32>) = G_BUILD_VECTOR [[TRUNC]](s32), [[TRUNC1]](s32), [[TRUNC2]](s32), [[DEF]](s32)
+ ; CHECK-NEXT: $q0 = COPY [[BUILD_VECTOR]](<4 x s32>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $q0
+ %1:_(s64) = COPY $d0
+ %2:_(s64) = COPY $d1
+ %3:_(s64) = COPY $d2
+ %0:_(<3 x s64>) = G_BUILD_VECTOR %1(s64), %2(s64), %3(s64)
+ %4:_(<3 x s32>) = G_TRUNC %0(<3 x s64>)
+ %5:_(s32), %6:_(s32), %7:_(s32) = G_UNMERGE_VALUES %4(<3 x s32>)
+ %8:_(s32) = G_IMPLICIT_DEF
+ %9:_(<4 x s32>) = G_BUILD_VECTOR %5(s32), %6(s32), %7(s32), %8(s32)
+ $q0 = COPY %9(<4 x s32>)
+ RET_ReallyLR implicit $q0
+
+...
+
+---
+name: xtn_v4i32_v4i8
+alignment: 4
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $q0
+
+ ; CHECK-LABEL: name: xtn_v4i32_v4i8
+ ; CHECK: liveins: $q0
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(<4 x s32>) = COPY $q0
+ ; CHECK-NEXT: [[TRUNC:%[0-9]+]]:_(<4 x s16>) = G_TRUNC [[COPY]](<4 x s32>)
+ ; CHECK-NEXT: $d0 = COPY [[TRUNC]](<4 x s16>)
+ ; CHECK-NEXT: RET_ReallyLR implicit $d0
+ %0:_(<4 x s32>) = COPY $q0
+ %1:_(<4 x s8>) = G_TRUNC %0(<4 x s32>)
+ %2:_(<4 x s16>) = G_ANYEXT %1(<4 x s8>)
----------------
chuongg3 wrote:
I believe this the removal of G_ANYEXT was done by the artifact combiner, where it recognises `anyext (trunc)` pattern and combines them into one instruction, either a `trunc` if the resulting element size is smaller than the source element size, or an `anyext` if the resulting element size is larger than the source element size.
https://github.com/llvm/llvm-project/pull/66927
More information about the llvm-commits
mailing list