[llvm] 9a7c26a - [GISel] Restrict G_BSWAP to multiples of 16 bits. (#70245)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Oct 30 10:28:01 PDT 2023
Author: Craig Topper
Date: 2023-10-30T10:27:57-07:00
New Revision: 9a7c26a399a827a5eb6ec5a2c5895c1bb5b08c54
URL: https://github.com/llvm/llvm-project/commit/9a7c26a399a827a5eb6ec5a2c5895c1bb5b08c54
DIFF: https://github.com/llvm/llvm-project/commit/9a7c26a399a827a5eb6ec5a2c5895c1bb5b08c54.diff
LOG: [GISel] Restrict G_BSWAP to multiples of 16 bits. (#70245)
This is consistent with the IR verifier and SelectionDAG's getNode.
Update tests accordingly. I tried to keep some coverage of non-pow2 when
possible. X86 didn't like a G_UNMERGE_VALUES from s48 to 3 s16 that got
created when I tried s48.
Added:
llvm/test/MachineVerifier/test_g_bswap.mir
Modified:
llvm/lib/CodeGen/MachineVerifier.cpp
llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-bswap.mir
llvm/test/CodeGen/X86/GlobalISel/legalize-bswap.mir
Removed:
################################################################################
diff --git a/llvm/lib/CodeGen/MachineVerifier.cpp b/llvm/lib/CodeGen/MachineVerifier.cpp
index f3e676b3a41a2cb..dadaf60fa09da04 100644
--- a/llvm/lib/CodeGen/MachineVerifier.cpp
+++ b/llvm/lib/CodeGen/MachineVerifier.cpp
@@ -1592,6 +1592,12 @@ void MachineVerifier::verifyPreISelGenericInstruction(const MachineInstr *MI) {
report("G_SEXT_INREG size must be less than source bit width", MI);
break;
}
+ case TargetOpcode::G_BSWAP: {
+ LLT DstTy = MRI->getType(MI->getOperand(0).getReg());
+ if (DstTy.getScalarSizeInBits() % 16 != 0)
+ report("G_BSWAP size must be a multiple of 16 bits", MI);
+ break;
+ }
case TargetOpcode::G_SHUFFLE_VECTOR: {
const MachineOperand &MaskOp = MI->getOperand(3);
if (!MaskOp.isShuffleMask()) {
diff --git a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
index 6111f4966028939..fba0881d4e86f10 100644
--- a/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
+++ b/llvm/test/CodeGen/AArch64/GlobalISel/legalize-bswap.mir
@@ -110,48 +110,27 @@ body: |
RET_ReallyLR implicit $q0
...
---
-name: bswap_s88
+name: bswap_s80
tracksRegLiveness: true
body: |
bb.0:
liveins: $x0
- ; CHECK-LABEL: name: bswap_s88
+ ; CHECK-LABEL: name: bswap_s80
; CHECK: liveins: $x0
- ; CHECK: [[DEF:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
- ; CHECK: [[BSWAP:%[0-9]+]]:_(s64) = G_BSWAP [[DEF]]
- ; CHECK: [[BSWAP1:%[0-9]+]]:_(s64) = G_BSWAP [[DEF]]
- ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 40
- ; CHECK: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[BSWAP]], [[C]](s64)
- ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 24
- ; CHECK: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[BSWAP1]], [[C1]](s64)
- ; CHECK: [[OR:%[0-9]+]]:_(s64) = G_OR [[LSHR]], [[SHL]]
- ; CHECK: $x0 = COPY [[OR]](s64)
- ; CHECK: RET_ReallyLR implicit $x0
- %val:_(s88) = G_IMPLICIT_DEF
- %bswap:_(s88) = G_BSWAP %val
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[DEF:%[0-9]+]]:_(s64) = G_IMPLICIT_DEF
+ ; CHECK-NEXT: [[BSWAP:%[0-9]+]]:_(s64) = G_BSWAP [[DEF]]
+ ; CHECK-NEXT: [[BSWAP1:%[0-9]+]]:_(s64) = G_BSWAP [[DEF]]
+ ; CHECK-NEXT: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 48
+ ; CHECK-NEXT: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[BSWAP]], [[C]](s64)
+ ; CHECK-NEXT: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 16
+ ; CHECK-NEXT: [[SHL:%[0-9]+]]:_(s64) = G_SHL [[BSWAP1]], [[C1]](s64)
+ ; CHECK-NEXT: [[OR:%[0-9]+]]:_(s64) = G_OR [[LSHR]], [[SHL]]
+ ; CHECK-NEXT: $x0 = COPY [[OR]](s64)
+ ; CHECK-NEXT: RET_ReallyLR implicit $x0
+ %val:_(s80) = G_IMPLICIT_DEF
+ %bswap:_(s80) = G_BSWAP %val
%trunc:_(s64) = G_TRUNC %bswap
$x0 = COPY %trunc(s64)
RET_ReallyLR implicit $x0
...
----
-name: bswap_s4
-tracksRegLiveness: true
-body: |
- bb.0:
- liveins: $x0
- ; CHECK-LABEL: name: bswap_s4
- ; CHECK: liveins: $x0
- ; CHECK: [[DEF:%[0-9]+]]:_(s32) = G_IMPLICIT_DEF
- ; CHECK: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[DEF]]
- ; CHECK: [[C:%[0-9]+]]:_(s64) = G_CONSTANT i64 28
- ; CHECK: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BSWAP]], [[C]](s64)
- ; CHECK: [[C1:%[0-9]+]]:_(s64) = G_CONSTANT i64 15
- ; CHECK: [[ANYEXT:%[0-9]+]]:_(s64) = G_ANYEXT [[LSHR]](s32)
- ; CHECK: %ext:_(s64) = G_AND [[ANYEXT]], [[C1]]
- ; CHECK: $x0 = COPY %ext(s64)
- ; CHECK: RET_ReallyLR implicit $x0
- %val:_(s4) = G_IMPLICIT_DEF
- %bswap:_(s4) = G_BSWAP %val
- %ext:_(s64) = G_ZEXT %bswap
- $x0 = COPY %ext(s64)
- RET_ReallyLR implicit $x0
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-bswap.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-bswap.mir
index 2b855e33e96d4d5..63235842de57bf2 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-bswap.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/legalize-bswap.mir
@@ -2,42 +2,6 @@
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=hawaii -O0 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX7 %s
# RUN: llc -mtriple=amdgcn-mesa-mesa3d -mcpu=fiji -O0 -run-pass=legalizer %s -o - | FileCheck -check-prefix=GFX8 %s
----
-name: bswap_s8
-
-body: |
- bb.0:
- liveins: $vgpr0
- ; GFX7-LABEL: name: bswap_s8
- ; GFX7: liveins: $vgpr0
- ; GFX7-NEXT: {{ $}}
- ; GFX7-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
- ; GFX7-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
- ; GFX7-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 255
- ; GFX7-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
- ; GFX7-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[COPY1]](s32)
- ; GFX7-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C]](s32)
- ; GFX7-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C1]]
- ; GFX7-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[AND]], [[COPY2]](s32)
- ; GFX7-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[LSHR]], [[SHL]]
- ; GFX7-NEXT: $vgpr0 = COPY [[OR]](s32)
- ; GFX8-LABEL: name: bswap_s8
- ; GFX8: liveins: $vgpr0
- ; GFX8-NEXT: {{ $}}
- ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
- ; GFX8-NEXT: [[TRUNC:%[0-9]+]]:_(s16) = G_TRUNC [[COPY]](s32)
- ; GFX8-NEXT: [[BSWAP:%[0-9]+]]:_(s16) = G_BSWAP [[TRUNC]]
- ; GFX8-NEXT: [[C:%[0-9]+]]:_(s16) = G_CONSTANT i16 8
- ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s16) = G_LSHR [[BSWAP]], [[C]](s16)
- ; GFX8-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[LSHR]](s16)
- ; GFX8-NEXT: $vgpr0 = COPY [[ANYEXT]](s32)
- %0:_(s32) = COPY $vgpr0
- %1:_(s8) = G_TRUNC %0
- %2:_(s8) = G_BSWAP %1
- %3:_(s32) = G_ANYEXT %2
- $vgpr0 = COPY %3
-...
-
---
name: bswap_s16
@@ -74,40 +38,6 @@ body: |
$vgpr0 = COPY %3
...
----
-name: bswap_s24
-
-body: |
- bb.0:
- liveins: $vgpr0
- ; GFX7-LABEL: name: bswap_s24
- ; GFX7: liveins: $vgpr0
- ; GFX7-NEXT: {{ $}}
- ; GFX7-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
- ; GFX7-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
- ; GFX7-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 16777215
- ; GFX7-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY [[C]](s32)
- ; GFX7-NEXT: [[SHL:%[0-9]+]]:_(s32) = G_SHL [[COPY]], [[COPY1]](s32)
- ; GFX7-NEXT: [[COPY2:%[0-9]+]]:_(s32) = COPY [[C]](s32)
- ; GFX7-NEXT: [[AND:%[0-9]+]]:_(s32) = G_AND [[COPY]], [[C1]]
- ; GFX7-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[AND]], [[COPY2]](s32)
- ; GFX7-NEXT: [[OR:%[0-9]+]]:_(s32) = G_OR [[LSHR]], [[SHL]]
- ; GFX7-NEXT: $vgpr0 = COPY [[OR]](s32)
- ; GFX8-LABEL: name: bswap_s24
- ; GFX8: liveins: $vgpr0
- ; GFX8-NEXT: {{ $}}
- ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $vgpr0
- ; GFX8-NEXT: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[COPY]]
- ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 8
- ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BSWAP]], [[C]](s32)
- ; GFX8-NEXT: $vgpr0 = COPY [[LSHR]](s32)
- %0:_(s32) = COPY $vgpr0
- %1:_(s24) = G_TRUNC %0
- %2:_(s24) = G_BSWAP %1
- %3:_(s32) = G_ANYEXT %2
- $vgpr0 = COPY %3
-...
-
---
name: bswap_s32
@@ -438,3 +368,39 @@ body: |
%1:_(<2 x s64>) = G_BSWAP %0
$vgpr0_vgpr1_vgpr2_vgpr3 = COPY %1
...
+
+---
+name: bswap_s48
+
+body: |
+ bb.0:
+ liveins: $vgpr0_vgpr1
+ ; GFX7-LABEL: name: bswap_s48
+ ; GFX7: liveins: $vgpr0_vgpr1
+ ; GFX7-NEXT: {{ $}}
+ ; GFX7-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+ ; GFX7-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
+ ; GFX7-NEXT: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[UV1]]
+ ; GFX7-NEXT: [[BSWAP1:%[0-9]+]]:_(s32) = G_BSWAP [[UV]]
+ ; GFX7-NEXT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[BSWAP]](s32), [[BSWAP1]](s32)
+ ; GFX7-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+ ; GFX7-NEXT: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[MV]], [[C]](s32)
+ ; GFX7-NEXT: $vgpr0_vgpr1 = COPY [[LSHR]](s64)
+ ;
+ ; GFX8-LABEL: name: bswap_s48
+ ; GFX8: liveins: $vgpr0_vgpr1
+ ; GFX8-NEXT: {{ $}}
+ ; GFX8-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $vgpr0_vgpr1
+ ; GFX8-NEXT: [[UV:%[0-9]+]]:_(s32), [[UV1:%[0-9]+]]:_(s32) = G_UNMERGE_VALUES [[COPY]](s64)
+ ; GFX8-NEXT: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[UV1]]
+ ; GFX8-NEXT: [[BSWAP1:%[0-9]+]]:_(s32) = G_BSWAP [[UV]]
+ ; GFX8-NEXT: [[MV:%[0-9]+]]:_(s64) = G_MERGE_VALUES [[BSWAP]](s32), [[BSWAP1]](s32)
+ ; GFX8-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 16
+ ; GFX8-NEXT: [[LSHR:%[0-9]+]]:_(s64) = G_LSHR [[MV]], [[C]](s32)
+ ; GFX8-NEXT: $vgpr0_vgpr1 = COPY [[LSHR]](s64)
+ %0:_(s64) = COPY $vgpr0_vgpr1
+ %1:_(s48) = G_TRUNC %0
+ %2:_(s48) = G_BSWAP %1
+ %3:_(s64) = G_ANYEXT %2
+ $vgpr0_vgpr1 = COPY %3
+...
diff --git a/llvm/test/CodeGen/X86/GlobalISel/legalize-bswap.mir b/llvm/test/CodeGen/X86/GlobalISel/legalize-bswap.mir
index bdac19b090d2235..5aa78f33aa98bd8 100644
--- a/llvm/test/CodeGen/X86/GlobalISel/legalize-bswap.mir
+++ b/llvm/test/CodeGen/X86/GlobalISel/legalize-bswap.mir
@@ -2,37 +2,8 @@
# RUN: llc -mtriple=i386-linux-gnu -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=X86-32
# RUN: llc -mtriple=x86_64-linux-gnu -run-pass=legalizer %s -o - | FileCheck %s --check-prefix=X86-64
-# test bswap for s16, s17, s32, and s64
+# test bswap for s16, s32, and s64
-...
----
-name: test_bswap17
-body: |
- bb.1:
- ; X86-32-LABEL: name: test_bswap17
- ; X86-32: [[DEF:%[0-9]+]]:_(s17) = IMPLICIT_DEF
- ; X86-32-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[DEF]](s17)
- ; X86-32-NEXT: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[ANYEXT]]
- ; X86-32-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 15
- ; X86-32-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BSWAP]], [[C]](s8)
- ; X86-32-NEXT: [[TRUNC:%[0-9]+]]:_(s17) = G_TRUNC [[LSHR]](s32)
- ; X86-32-NEXT: [[COPY:%[0-9]+]]:_(s17) = COPY [[TRUNC]](s17)
- ; X86-32-NEXT: RET 0, implicit [[COPY]](s17)
- ; X86-64-LABEL: name: test_bswap17
- ; X86-64: [[DEF:%[0-9]+]]:_(s17) = IMPLICIT_DEF
- ; X86-64-NEXT: [[ANYEXT:%[0-9]+]]:_(s32) = G_ANYEXT [[DEF]](s17)
- ; X86-64-NEXT: [[BSWAP:%[0-9]+]]:_(s32) = G_BSWAP [[ANYEXT]]
- ; X86-64-NEXT: [[C:%[0-9]+]]:_(s8) = G_CONSTANT i8 15
- ; X86-64-NEXT: [[LSHR:%[0-9]+]]:_(s32) = G_LSHR [[BSWAP]], [[C]](s8)
- ; X86-64-NEXT: [[TRUNC:%[0-9]+]]:_(s17) = G_TRUNC [[LSHR]](s32)
- ; X86-64-NEXT: [[COPY:%[0-9]+]]:_(s17) = COPY [[TRUNC]](s17)
- ; X86-64-NEXT: RET 0, implicit [[COPY]](s17)
- %0:_(s17) = IMPLICIT_DEF
- %1:_(s17) = G_BSWAP %0
- %2:_(s17) = COPY %1(s17)
- RET 0, implicit %2
-
-...
---
name: test_bswap64
body: |
diff --git a/llvm/test/MachineVerifier/test_g_bswap.mir b/llvm/test/MachineVerifier/test_g_bswap.mir
new file mode 100644
index 000000000000000..679114f06a00e0c
--- /dev/null
+++ b/llvm/test/MachineVerifier/test_g_bswap.mir
@@ -0,0 +1,19 @@
+#RUN: not --crash llc -mtriple=aarch64 -o - -global-isel -run-pass=none -verify-machineinstrs %s 2>&1 | FileCheck %s
+# REQUIRES: aarch64-registered-target
+
+---
+name: test_bswap
+legalized: true
+regBankSelected: false
+selected: false
+tracksRegLiveness: true
+liveins:
+body: |
+ bb.0:
+
+ %0:_(s17) = G_CONSTANT i32 17
+
+ ; CHECK: Bad machine code: G_BSWAP size must be a multiple of 16 bits
+ %1:_(s17) = G_BSWAP %0
+
+...
More information about the llvm-commits
mailing list