[GlobalISel] combine G_TRUNC with G_MERGE_VALUES
Summary:
Truncating the result of a merge means that most likely we could have done without merge in the first place and just used the input merge inputs directly. This can be done in three cases:
1. If the truncation result is smaller than the merge source, we can use the source in the trunc directly
2. If the sizes are the same, we can replace the register or use a copy
3. If the truncation size is a multiple of the merge source size, we can build a smaller merge
This gets rid of most of the larger, hard-to-legalize merges.
Reviewers: qcolombet, aditya_nandakumar, aemerson, paquette, arsenm, Petar.Avramovic
Reviewed By: arsenm
Subscribers: sdardis, jvesely, wdng, nhaehnle, rovka, jrtc27, atanasyan, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75915
2020-03-10 00:00:38 +08:00
|
|
|
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
|
|
|
|
# RUN: llc -O0 -mtriple=amdgcn-mesa-mesa3d -mcpu=tahiti -run-pass=legalizer %s -o - | FileCheck %s
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s16_merge_s64_s32
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(merge) with trunc-size < merge-source-size creates a trunc
|
|
|
|
; of the merge source
|
|
|
|
; CHECK-LABEL: name: trunc_s16_merge_s64_s32
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
|
|
|
|
; CHECK: $vgpr0 = COPY [[C]](s32)
|
|
|
|
%0:_(s32) = G_CONSTANT i32 0
|
|
|
|
%1:_(s32) = G_CONSTANT i32 1
|
|
|
|
%2:_(s64) = G_MERGE_VALUES %0, %1
|
|
|
|
%3:_(s16) = G_TRUNC %2
|
|
|
|
%4:_(s32) = G_ANYEXT %3
|
|
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s32_merge_s64_s32
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(merge) with trunc-size == merge-source-size is eliminated
|
|
|
|
; CHECK-LABEL: name: trunc_s32_merge_s64_s32
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
|
|
|
|
; CHECK: $vgpr0 = COPY [[C]](s32)
|
|
|
|
%0:_(s32) = G_CONSTANT i32 0
|
|
|
|
%1:_(s32) = G_CONSTANT i32 1
|
|
|
|
%2:_(s64) = G_MERGE_VALUES %0, %1
|
|
|
|
%3:_(s32) = G_TRUNC %2
|
|
|
|
$vgpr0 = COPY %3
|
|
|
|
...
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s64_merge_s128_s32
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(merge) with trunc-size > merge-source-size combines to a
|
|
|
|
; smaller merge
|
|
|
|
; CHECK-LABEL: name: trunc_s64_merge_s128_s32
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
|
|
|
|
; CHECK: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[C]](s32), [[C1]](s32)
|
|
|
|
; CHECK: $vgpr0_vgpr1 = COPY [[MV]](s64)
|
|
|
|
%0:_(s32) = G_CONSTANT i32 0
|
|
|
|
%1:_(s32) = G_CONSTANT i32 1
|
|
|
|
%2:_(s128) = G_MERGE_VALUES %0, %1, %0, %1
|
|
|
|
%3:_(s64) = G_TRUNC %2
|
|
|
|
$vgpr0_vgpr1 = COPY %3
|
|
|
|
...
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s32_merge_s128_p0
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(merge) with a non-scalar merge source is not combined
|
|
|
|
; CHECK-LABEL: name: trunc_s32_merge_s128_p0
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(p0) = G_CONSTANT i64 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(p0) = G_CONSTANT i64 1
|
|
|
|
; CHECK: [[MV:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[C]](p0), [[C1]](p0)
|
|
|
|
; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[MV]](s128)
|
|
|
|
; CHECK: $vgpr0 = COPY [[TRUNC]](s32)
|
|
|
|
%0:_(p0) = G_CONSTANT i64 0
|
|
|
|
%1:_(p0) = G_CONSTANT i64 1
|
|
|
|
%2:_(s128) = G_MERGE_VALUES %0, %1
|
|
|
|
%3:_(s32) = G_TRUNC %2
|
|
|
|
$vgpr0 = COPY %3
|
|
|
|
...
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s64_merge_s128_p0
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(merge) with a non-scalar merge source is not combined
|
|
|
|
; CHECK-LABEL: name: trunc_s64_merge_s128_p0
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(p0) = G_CONSTANT i64 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(p0) = G_CONSTANT i64 1
|
|
|
|
; CHECK: [[MV:%[0-9]+]]:_(s128) = G_MERGE_VALUES [[C]](p0), [[C1]](p0)
|
|
|
|
; CHECK: [[TRUNC:%[0-9]+]]:_(s64) = G_TRUNC [[MV]](s128)
|
|
|
|
; CHECK: $vgpr0_vgpr1 = COPY [[TRUNC]](s64)
|
|
|
|
%0:_(p0) = G_CONSTANT i64 0
|
|
|
|
%1:_(p0) = G_CONSTANT i64 1
|
|
|
|
%2:_(s128) = G_MERGE_VALUES %0, %1
|
|
|
|
%3:_(s64) = G_TRUNC %2
|
|
|
|
$vgpr0_vgpr1 = COPY %3
|
|
|
|
...
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s128_merge_s192_p0
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(merge) with a non-scalar merge source is not combined
|
|
|
|
; CHECK-LABEL: name: trunc_s128_merge_s192_p0
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(p0) = G_CONSTANT i64 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(p0) = G_CONSTANT i64 1
|
|
|
|
; CHECK: [[MV:%[0-9]+]]:_(s192) = G_MERGE_VALUES [[C]](p0), [[C1]](p0), [[C]](p0)
|
|
|
|
; CHECK: [[TRUNC:%[0-9]+]]:_(s128) = G_TRUNC [[MV]](s192)
|
|
|
|
; CHECK: $vgpr0_vgpr1_vgpr2_vgpr3 = COPY [[TRUNC]](s128)
|
|
|
|
%0:_(p0) = G_CONSTANT i64 0
|
|
|
|
%1:_(p0) = G_CONSTANT i64 1
|
|
|
|
%2:_(s192) = G_MERGE_VALUES %0, %1, %0
|
|
|
|
%3:_(s128) = G_TRUNC %2
|
|
|
|
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %3
|
|
|
|
...
|
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_s68_merge_s128_s32
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
2020-03-23 19:58:55 +08:00
|
|
|
; Test that trunc(merge) with trunc-size > merge-source-size and
|
|
|
|
; trunc-size % merge-source-size != 0 can be combined after the G_TRUNCs
|
|
|
|
; have been combined.
|
[GlobalISel] combine G_TRUNC with G_MERGE_VALUES
Summary:
Truncating the result of a merge means that most likely we could have done without merge in the first place and just used the input merge inputs directly. This can be done in three cases:
1. If the truncation result is smaller than the merge source, we can use the source in the trunc directly
2. If the sizes are the same, we can replace the register or use a copy
3. If the truncation size is a multiple of the merge source size, we can build a smaller merge
This gets rid of most of the larger, hard-to-legalize merges.
Reviewers: qcolombet, aditya_nandakumar, aemerson, paquette, arsenm, Petar.Avramovic
Reviewed By: arsenm
Subscribers: sdardis, jvesely, wdng, nhaehnle, rovka, jrtc27, atanasyan, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75915
2020-03-10 00:00:38 +08:00
|
|
|
; CHECK-LABEL: name: trunc_s68_merge_s128_s32
|
|
|
|
; CHECK: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
|
|
|
|
; CHECK: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 1
|
2020-03-23 19:58:55 +08:00
|
|
|
; CHECK: $vgpr0 = COPY [[C]](s32)
|
[GlobalISel] combine G_TRUNC with G_MERGE_VALUES
Summary:
Truncating the result of a merge means that most likely we could have done without merge in the first place and just used the input merge inputs directly. This can be done in three cases:
1. If the truncation result is smaller than the merge source, we can use the source in the trunc directly
2. If the sizes are the same, we can replace the register or use a copy
3. If the truncation size is a multiple of the merge source size, we can build a smaller merge
This gets rid of most of the larger, hard-to-legalize merges.
Reviewers: qcolombet, aditya_nandakumar, aemerson, paquette, arsenm, Petar.Avramovic
Reviewed By: arsenm
Subscribers: sdardis, jvesely, wdng, nhaehnle, rovka, jrtc27, atanasyan, kerbowa, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D75915
2020-03-10 00:00:38 +08:00
|
|
|
%0:_(s32) = G_CONSTANT i32 0
|
|
|
|
%1:_(s32) = G_CONSTANT i32 1
|
|
|
|
%2:_(s128) = G_MERGE_VALUES %0, %1, %0, %1
|
|
|
|
%3:_(s68) = G_TRUNC %2
|
|
|
|
%4:_(s32) = G_TRUNC %3
|
|
|
|
$vgpr0 = COPY %4
|
|
|
|
...
|
2020-03-23 19:58:55 +08:00
|
|
|
|
|
|
|
---
|
|
|
|
name: trunc_trunc
|
|
|
|
|
|
|
|
body: |
|
|
|
|
bb.0:
|
|
|
|
; Test that trunc(trunc) is combined to a single trunc
|
|
|
|
; CHECK-LABEL: name: trunc_trunc
|
|
|
|
; CHECK: [[DEF:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
|
|
|
|
; CHECK: [[TRUNC:%[0-9]+]]:_(s32) = G_TRUNC [[DEF]](s64)
|
|
|
|
; CHECK: $vgpr0 = COPY [[TRUNC]](s32)
|
|
|
|
%0:_(s64) = G_IMPLICIT_DEF
|
|
|
|
%1:_(s48) = G_TRUNC %0
|
|
|
|
%2:_(s32) = G_TRUNC %1
|
|
|
|
$vgpr0 = COPY %2
|
|
|
|
...
|