-
Notifications
You must be signed in to change notification settings - Fork 15.3k
[RISCV][GISel] Add fcopysign tests with mismatched types. NFC #115364
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
fcopysign takes two operands, one for the magnitude and exponent and the other for the sign. The result type is determined by the magnitude and exponent operand. The sign operand does not need to be the same type. Note, in IR all 3 types must match. We have isel patterns for all possible combinations of types and the legalizer thinks they are all legal, but we had no tests for it. I was unable to find any combiner rules that would create an instruction with mismatch types.
|
@llvm/pr-subscribers-llvm-globalisel Author: Craig Topper (topperc) Changesfcopysign takes two operands, one for the magnitude and exponent and the other for the sign. The result type is determined by the magnitude and exponent operand. The sign operand does not need to be the same type. Note, in IR all 3 types must match. We have isel patterns for all possible combinations of types and the legalizer thinks they are all legal, but we had no tests for it. I was unable to find any combiner rules that would create an instruction with mismatch types. Patch is 21.09 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/115364.diff 6 Files Affected:
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith-f16.mir b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith-f16.mir
index 96f019948537fd..855764a8b8c09e 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith-f16.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith-f16.mir
@@ -1,7 +1,7 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=riscv32 -mattr=+zfh -run-pass=instruction-select \
+# RUN: llc -mtriple=riscv32 -mattr=+zfh,+d -run-pass=instruction-select \
# RUN: -simplify-mir -verify-machineinstrs %s -o - | FileCheck %s
-# RUN: llc -mtriple=riscv64 -mattr=+zfh -run-pass=instruction-select \
+# RUN: llc -mtriple=riscv64 -mattr=+zfh,+d -run-pass=instruction-select \
# RUN: -simplify-mir -verify-machineinstrs %s -o - | FileCheck %s
---
@@ -241,7 +241,7 @@ body: |
...
---
-name: fcopysign_f16
+name: fcopysign_f16_f16
legalized: true
regBankSelected: true
tracksRegLiveness: true
@@ -249,7 +249,7 @@ body: |
bb.0:
liveins: $f10_h, $f11_h
- ; CHECK-LABEL: name: fcopysign_f16
+ ; CHECK-LABEL: name: fcopysign_f16_f16
; CHECK: liveins: $f10_h, $f11_h
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $f10_h
@@ -264,3 +264,103 @@ body: |
PseudoRET implicit $f10_h
...
+---
+name: fcopysign_f16_f32
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_h, $f11_f
+
+ ; CHECK-LABEL: name: fcopysign_f16_f32
+ ; CHECK: liveins: $f10_h, $f11_f
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $f10_h
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr32 = COPY $f11_f
+ ; CHECK-NEXT: [[FCVT_H_S:%[0-9]+]]:fpr16 = nofpexcept FCVT_H_S [[COPY1]], 7
+ ; CHECK-NEXT: [[FSGNJ_H:%[0-9]+]]:fpr16 = FSGNJ_H [[COPY]], [[FCVT_H_S]]
+ ; CHECK-NEXT: $f10_h = COPY [[FSGNJ_H]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_h
+ %0:fprb(s16) = COPY $f10_h
+ %1:fprb(s32) = COPY $f11_f
+ %2:fprb(s16) = G_FCOPYSIGN %0, %1
+ $f10_h = COPY %2(s16)
+ PseudoRET implicit $f10_h
+
+...
+---
+name: fcopysign_f32_f16
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_f, $f11_h
+
+ ; CHECK-LABEL: name: fcopysign_f32_f16
+ ; CHECK: liveins: $f10_f, $f11_h
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr32 = COPY $f10_f
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $f11_h
+ ; CHECK-NEXT: [[FCVT_S_H:%[0-9]+]]:fpr32 = nofpexcept FCVT_S_H [[COPY1]], 0
+ ; CHECK-NEXT: [[FSGNJ_S:%[0-9]+]]:fpr32 = FSGNJ_S [[COPY]], [[FCVT_S_H]]
+ ; CHECK-NEXT: $f10_f = COPY [[FSGNJ_S]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:fprb(s32) = COPY $f10_f
+ %1:fprb(s16) = COPY $f11_h
+ %2:fprb(s32) = G_FCOPYSIGN %0, %1
+ $f10_f = COPY %2(s32)
+ PseudoRET implicit $f10_f
+
+...
+---
+name: fcopysign_f16_f64
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_h, $f11_d
+
+ ; CHECK-LABEL: name: fcopysign_f16_f64
+ ; CHECK: liveins: $f10_h, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr16 = COPY $f10_h
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr64 = COPY $f11_d
+ ; CHECK-NEXT: [[FCVT_H_D:%[0-9]+]]:fpr16 = nofpexcept FCVT_H_D [[COPY1]], 7
+ ; CHECK-NEXT: [[FSGNJ_H:%[0-9]+]]:fpr16 = FSGNJ_H [[COPY]], [[FCVT_H_D]]
+ ; CHECK-NEXT: $f10_h = COPY [[FSGNJ_H]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_h
+ %0:fprb(s16) = COPY $f10_h
+ %1:fprb(s64) = COPY $f11_d
+ %2:fprb(s16) = G_FCOPYSIGN %0, %1
+ $f10_h = COPY %2(s16)
+ PseudoRET implicit $f10_h
+
+...
+---
+name: fcopysign_f64_f16
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_d, $f11_h
+
+ ; CHECK-LABEL: name: fcopysign_f64_f16
+ ; CHECK: liveins: $f10_d, $f11_h
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr64 = COPY $f10_d
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr16 = COPY $f11_h
+ ; CHECK-NEXT: [[FCVT_D_H:%[0-9]+]]:fpr64 = nofpexcept FCVT_D_H [[COPY1]], 0
+ ; CHECK-NEXT: [[FSGNJ_D:%[0-9]+]]:fpr64 = FSGNJ_D [[COPY]], [[FCVT_D_H]]
+ ; CHECK-NEXT: $f10_d = COPY [[FSGNJ_D]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_d
+ %0:fprb(s64) = COPY $f10_d
+ %1:fprb(s16) = COPY $f11_h
+ %2:fprb(s64) = G_FCOPYSIGN %0, %1
+ $f10_d = COPY %2(s64)
+ PseudoRET implicit $f10_d
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith.mir b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith.mir
index 037647ae688725..45cba58875341e 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/instruction-select/fp-arith.mir
@@ -241,7 +241,7 @@ body: |
...
---
-name: fcopysign_f32
+name: fcopysign_f32_f32
legalized: true
regBankSelected: true
tracksRegLiveness: true
@@ -249,7 +249,7 @@ body: |
bb.0:
liveins: $f10_f, $f11_f
- ; CHECK-LABEL: name: fcopysign_f32
+ ; CHECK-LABEL: name: fcopysign_f32_f32
; CHECK: liveins: $f10_f, $f11_f
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr32 = COPY $f10_f
@@ -263,6 +263,31 @@ body: |
$f10_f = COPY %2(s32)
PseudoRET implicit $f10_f
+...
+---
+name: fcopysign_f32_f64
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_f, $f11_d
+
+ ; CHECK-LABEL: name: fcopysign_f32_f64
+ ; CHECK: liveins: $f10_f, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr32 = COPY $f10_f
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr64 = COPY $f11_d
+ ; CHECK-NEXT: [[FCVT_S_D:%[0-9]+]]:fpr32 = nofpexcept FCVT_S_D [[COPY1]], 7
+ ; CHECK-NEXT: [[FSGNJ_S:%[0-9]+]]:fpr32 = FSGNJ_S [[COPY]], [[FCVT_S_D]]
+ ; CHECK-NEXT: $f10_f = COPY [[FSGNJ_S]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:fprb(s32) = COPY $f10_f
+ %1:fprb(s64) = COPY $f11_d
+ %2:fprb(s32) = G_FCOPYSIGN %0, %1
+ $f10_f = COPY %2(s32)
+ PseudoRET implicit $f10_f
+
...
---
name: fadd_f64
@@ -501,7 +526,7 @@ body: |
...
---
-name: fcopysign_f64
+name: fcopysign_f64_f64
legalized: true
regBankSelected: true
tracksRegLiveness: true
@@ -509,7 +534,7 @@ body: |
bb.0:
liveins: $f10_d, $f11_d
- ; CHECK-LABEL: name: fcopysign_f64
+ ; CHECK-LABEL: name: fcopysign_f64_f64
; CHECK: liveins: $f10_d, $f11_d
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr64 = COPY $f10_d
@@ -524,3 +549,28 @@ body: |
PseudoRET implicit $f10_d
...
+---
+name: fcopysign_f64_f32
+legalized: true
+regBankSelected: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_d, $f11_f
+
+ ; CHECK-LABEL: name: fcopysign_f64_f32
+ ; CHECK: liveins: $f10_d, $f11_f
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr64 = COPY $f10_d
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fpr32 = COPY $f11_f
+ ; CHECK-NEXT: [[FCVT_D_S:%[0-9]+]]:fpr64 = nofpexcept FCVT_D_S [[COPY1]], 0
+ ; CHECK-NEXT: [[FSGNJ_D:%[0-9]+]]:fpr64 = FSGNJ_D [[COPY]], [[FCVT_D_S]]
+ ; CHECK-NEXT: $f10_d = COPY [[FSGNJ_D]]
+ ; CHECK-NEXT: PseudoRET implicit $f10_d
+ %0:fprb(s64) = COPY $f10_d
+ %1:fprb(s32) = COPY $f11_f
+ %2:fprb(s64) = G_FCOPYSIGN %0, %1
+ $f10_d = COPY %2(s64)
+ PseudoRET implicit $f10_d
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith-f16.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith-f16.mir
index f1777e945b12db..f9e459271e640d 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith-f16.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith-f16.mir
@@ -1,7 +1,7 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=riscv32 -mattr=+zfh -run-pass=legalizer %s -o - \
+# RUN: llc -mtriple=riscv32 -mattr=+zfh,+d -run-pass=legalizer %s -o - \
# RUN: | FileCheck %s
-# RUN: llc -mtriple=riscv64 -mattr=+zfh -run-pass=legalizer %s -o - \
+# RUN: llc -mtriple=riscv64 -mattr=+zfh,+d -run-pass=legalizer %s -o - \
# RUN: | FileCheck %s
---
@@ -211,12 +211,12 @@ body: |
...
---
-name: fcopysign_f16
+name: fcopysign_f16_f16
body: |
bb.0:
liveins: $f10_h, $f11_h
- ; CHECK-LABEL: name: fcopysign_f16
+ ; CHECK-LABEL: name: fcopysign_f16_f16
; CHECK: liveins: $f10_h, $f11_h
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $f10_h
@@ -231,3 +231,87 @@ body: |
PseudoRET implicit $f10_h
...
+---
+name: fcopysign_f16_f32
+body: |
+ bb.0:
+ liveins: $f10_h, $f11_f
+
+ ; CHECK-LABEL: name: fcopysign_f16_f32
+ ; CHECK: liveins: $f10_h, $f11_f
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $f10_h
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $f11_f
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:_(s16) = G_FCOPYSIGN [[COPY]], [[COPY1]](s32)
+ ; CHECK-NEXT: $f10_h = COPY [[FCOPYSIGN]](s16)
+ ; CHECK-NEXT: PseudoRET implicit $f10_h
+ %0:_(s16) = COPY $f10_h
+ %1:_(s32) = COPY $f11_f
+ %2:_(s16) = G_FCOPYSIGN %0, %1
+ $f10_h = COPY %2(s16)
+ PseudoRET implicit $f10_h
+
+...
+---
+name: fcopysign_f32_f16
+body: |
+ bb.0:
+ liveins: $f10_f, $f11_h
+
+ ; CHECK-LABEL: name: fcopysign_f32_f16
+ ; CHECK: liveins: $f10_f, $f11_h
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $f10_f
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s16) = COPY $f11_h
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:_(s32) = G_FCOPYSIGN [[COPY]], [[COPY1]](s16)
+ ; CHECK-NEXT: $f10_f = COPY [[FCOPYSIGN]](s32)
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:_(s32) = COPY $f10_f
+ %1:_(s16) = COPY $f11_h
+ %2:_(s32) = G_FCOPYSIGN %0, %1
+ $f10_f = COPY %2(s32)
+ PseudoRET implicit $f10_f
+
+...
+---
+name: fcopysign_f16_f64
+body: |
+ bb.0:
+ liveins: $f10_h, $f11_d
+
+ ; CHECK-LABEL: name: fcopysign_f16_f64
+ ; CHECK: liveins: $f10_h, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s16) = COPY $f10_h
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $f11_d
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:_(s16) = G_FCOPYSIGN [[COPY]], [[COPY1]](s64)
+ ; CHECK-NEXT: $f10_h = COPY [[FCOPYSIGN]](s16)
+ ; CHECK-NEXT: PseudoRET implicit $f10_h
+ %0:_(s16) = COPY $f10_h
+ %1:_(s64) = COPY $f11_d
+ %2:_(s16) = G_FCOPYSIGN %0, %1
+ $f10_h = COPY %2(s16)
+ PseudoRET implicit $f10_h
+
+...
+---
+name: fcopysign_f64_f16
+body: |
+ bb.0:
+ liveins: $f10_d, $f11_h
+
+ ; CHECK-LABEL: name: fcopysign_f64_f16
+ ; CHECK: liveins: $f10_d, $f11_h
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $f10_d
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s16) = COPY $f11_h
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:_(s64) = G_FCOPYSIGN [[COPY]], [[COPY1]](s16)
+ ; CHECK-NEXT: $f10_d = COPY [[FCOPYSIGN]](s64)
+ ; CHECK-NEXT: PseudoRET implicit $f10_d
+ %0:_(s64) = COPY $f10_d
+ %1:_(s16) = COPY $f11_h
+ %2:_(s64) = G_FCOPYSIGN %0, %1
+ $f10_d = COPY %2(s64)
+ PseudoRET implicit $f10_d
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith.mir b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith.mir
index 599779d09c2c15..af4feac4e62fe3 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/legalizer/legalize-fp-arith.mir
@@ -211,12 +211,12 @@ body: |
...
---
-name: fcopysign_f32
+name: fcopysign_f32_f32
body: |
bb.0:
liveins: $f10_f, $f11_f
- ; CHECK-LABEL: name: fcopysign_f32
+ ; CHECK-LABEL: name: fcopysign_f32_f32
; CHECK: liveins: $f10_f, $f11_f
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $f10_f
@@ -230,6 +230,27 @@ body: |
$f10_f = COPY %2(s32)
PseudoRET implicit $f10_f
+...
+---
+name: fcopysign_f32_f64
+body: |
+ bb.0:
+ liveins: $f10_f, $f11_d
+
+ ; CHECK-LABEL: name: fcopysign_f32_f64
+ ; CHECK: liveins: $f10_f, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $f10_f
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s64) = COPY $f11_d
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:_(s32) = G_FCOPYSIGN [[COPY]], [[COPY1]](s64)
+ ; CHECK-NEXT: $f10_f = COPY [[FCOPYSIGN]](s32)
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:_(s32) = COPY $f10_f
+ %1:_(s64) = COPY $f11_d
+ %2:_(s32) = G_FCOPYSIGN %0, %1
+ $f10_f = COPY %2(s32)
+ PseudoRET implicit $f10_f
+
...
---
name: fadd_f64
@@ -438,12 +459,12 @@ body: |
...
---
-name: fcopysign_f64
+name: fcopysign_f64_f64
body: |
bb.0:
liveins: $f10_d, $f11_d
- ; CHECK-LABEL: name: fcopysign_f64
+ ; CHECK-LABEL: name: fcopysign_f64_f64
; CHECK: liveins: $f10_d, $f11_d
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $f10_d
@@ -458,3 +479,24 @@ body: |
PseudoRET implicit $f10_d
...
+---
+name: fcopysign_f64_f32
+body: |
+ bb.0:
+ liveins: $f10_d, $f11_f
+
+ ; CHECK-LABEL: name: fcopysign_f64_f32
+ ; CHECK: liveins: $f10_d, $f11_f
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s64) = COPY $f10_d
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $f11_f
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:_(s64) = G_FCOPYSIGN [[COPY]], [[COPY1]](s32)
+ ; CHECK-NEXT: $f10_d = COPY [[FCOPYSIGN]](s64)
+ ; CHECK-NEXT: PseudoRET implicit $f10_d
+ %0:_(s64) = COPY $f10_d
+ %1:_(s32) = COPY $f11_f
+ %2:_(s64) = G_FCOPYSIGN %0, %1
+ $f10_d = COPY %2(s64)
+ PseudoRET implicit $f10_d
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith-f16.mir b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith-f16.mir
index ace4aa542377b3..1ef5c48f0b5b67 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith-f16.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith-f16.mir
@@ -1,8 +1,8 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=riscv32 -mattr=+zfh -run-pass=regbankselect \
+# RUN: llc -mtriple=riscv32 -mattr=+zfh,+d -run-pass=regbankselect \
# RUN: -simplify-mir -verify-machineinstrs %s \
# RUN: -o - | FileCheck %s
-# RUN: llc -mtriple=riscv64 -mattr=+zfh -run-pass=regbankselect \
+# RUN: llc -mtriple=riscv64 -mattr=+zfh,+d -run-pass=regbankselect \
# RUN: -simplify-mir -verify-machineinstrs %s \
# RUN: -o - | FileCheck %s
@@ -233,14 +233,14 @@ body: |
...
---
-name: fcopysign_f16
+name: fcopysign_f16_f16
legalized: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $f10_h, $f11_h
- ; CHECK-LABEL: name: fcopysign_f16
+ ; CHECK-LABEL: name: fcopysign_f16_f16
; CHECK: liveins: $f10_h, $f11_h
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s16) = COPY $f10_h
@@ -255,3 +255,95 @@ body: |
PseudoRET implicit $f10_h
...
+---
+name: fcopysign_f16_f32
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_h, $f11_f
+
+ ; CHECK-LABEL: name: fcopysign_f16_f32
+ ; CHECK: liveins: $f10_h, $f11_f
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s16) = COPY $f10_h
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fprb(s32) = COPY $f11_f
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:fprb(s16) = G_FCOPYSIGN [[COPY]], [[COPY1]](s32)
+ ; CHECK-NEXT: $f10_h = COPY [[FCOPYSIGN]](s16)
+ ; CHECK-NEXT: PseudoRET implicit $f10_h
+ %0:_(s16) = COPY $f10_h
+ %1:_(s32) = COPY $f11_f
+ %2:_(s16) = G_FCOPYSIGN %0, %1
+ $f10_h = COPY %2(s16)
+ PseudoRET implicit $f10_h
+
+...
+---
+name: fcopysign_f32_f16
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_f, $f11_h
+
+ ; CHECK-LABEL: name: fcopysign_f32_f16
+ ; CHECK: liveins: $f10_f, $f11_h
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s32) = COPY $f10_f
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fprb(s16) = COPY $f11_h
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:fprb(s32) = G_FCOPYSIGN [[COPY]], [[COPY1]](s16)
+ ; CHECK-NEXT: $f10_f = COPY [[FCOPYSIGN]](s32)
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:_(s32) = COPY $f10_f
+ %1:_(s16) = COPY $f11_h
+ %2:_(s32) = G_FCOPYSIGN %0, %1
+ $f10_f = COPY %2(s32)
+ PseudoRET implicit $f10_f
+
+...
+---
+name: fcopysign_f16_f64
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_h, $f11_d
+
+ ; CHECK-LABEL: name: fcopysign_f16_f64
+ ; CHECK: liveins: $f10_h, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s16) = COPY $f10_h
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fprb(s64) = COPY $f11_d
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:fprb(s16) = G_FCOPYSIGN [[COPY]], [[COPY1]](s64)
+ ; CHECK-NEXT: $f10_h = COPY [[FCOPYSIGN]](s16)
+ ; CHECK-NEXT: PseudoRET implicit $f10_h
+ %0:_(s16) = COPY $f10_h
+ %1:_(s64) = COPY $f11_d
+ %2:_(s16) = G_FCOPYSIGN %0, %1
+ $f10_h = COPY %2(s16)
+ PseudoRET implicit $f10_h
+
+...
+---
+name: fcopysign_f64_f16
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_d, $f11_h
+
+ ; CHECK-LABEL: name: fcopysign_f64_f16
+ ; CHECK: liveins: $f10_d, $f11_h
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s64) = COPY $f10_d
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fprb(s16) = COPY $f11_h
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:fprb(s64) = G_FCOPYSIGN [[COPY]], [[COPY1]](s16)
+ ; CHECK-NEXT: $f10_d = COPY [[FCOPYSIGN]](s64)
+ ; CHECK-NEXT: PseudoRET implicit $f10_d
+ %0:_(s64) = COPY $f10_d
+ %1:_(s16) = COPY $f11_h
+ %2:_(s64) = G_FCOPYSIGN %0, %1
+ $f10_d = COPY %2(s64)
+ PseudoRET implicit $f10_d
+
+...
diff --git a/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith.mir b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith.mir
index 1ac6837a77bdb9..2801faf76820b2 100644
--- a/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith.mir
+++ b/llvm/test/CodeGen/RISCV/GlobalISel/regbankselect/fp-arith.mir
@@ -233,14 +233,14 @@ body: |
...
---
-name: fcopysign_f32
+name: fcopysign_f32_f32
legalized: true
tracksRegLiveness: true
body: |
bb.0:
liveins: $f10_f, $f11_f
- ; CHECK-LABEL: name: fcopysign_f32
+ ; CHECK-LABEL: name: fcopysign_f32_f32
; CHECK: liveins: $f10_f, $f11_f
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s32) = COPY $f10_f
@@ -254,6 +254,29 @@ body: |
$f10_f = COPY %2(s32)
PseudoRET implicit $f10_f
+...
+---
+name: fcopysign_f32_f64
+legalized: true
+tracksRegLiveness: true
+body: |
+ bb.0:
+ liveins: $f10_f, $f11_d
+
+ ; CHECK-LABEL: name: fcopysign_f32_f64
+ ; CHECK: liveins: $f10_f, $f11_d
+ ; CHECK-NEXT: {{ $}}
+ ; CHECK-NEXT: [[COPY:%[0-9]+]]:fprb(s32) = COPY $f10_f
+ ; CHECK-NEXT: [[COPY1:%[0-9]+]]:fprb(s64) = COPY $f11_d
+ ; CHECK-NEXT: [[FCOPYSIGN:%[0-9]+]]:fprb(s32) = G_FCOPYSIGN [[COPY]], [[COPY1]](s64)
+ ; CHECK-NEXT: $f10_f = COPY [[FCOPYSIGN]](s32)
+ ; CHECK-NEXT: PseudoRET implicit $f10_f
+ %0:_(s32) = COPY $f10_f
+ %1:_(s64) = COPY $f11_d
+ %2:_(s32) = G_FCOPYSIGN %0, %1
+ $f10_f = COPY %2(s32)
+ PseudoRET implicit $f10_f
+
...
---
name: fadd_f64
@@ -482,14 +505,14 @@ body: |
...
---
-name: fcopysign_f64
+name: fcopysign_f64_f64
legalized...
[truncated]
|
…15364) fcopysign takes two operands, one for the magnitude and exponent and the other for the sign. The result type is determined by the magnitude and exponent operand. The sign operand does not need to be the same type. Note, in IR all 3 types must match. We have isel patterns for all possible combinations of types and the legalizer thinks they are all legal, but we had no tests for it. I was unable to find any combiner rules that would create an instruction with mismatch types.
fcopysign takes two operands, one for the magnitude and exponent and the other for the sign. The result type is determined by the magnitude and exponent operand. The sign operand does not need to be the same type. Note, in IR all 3 types must match.
We have isel patterns for all possible combinations of types and the legalizer thinks they are all legal, but we had no tests for it. I was unable to find any combiner rules that would create an instruction with mismatch types.