[RISCV] Promote fldexp with Zfh. #117396
Merged
[RISCV] Promote fldexp with Zfh. #117396
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
The default expansion tries to create i16 operations after type legalization. Fixes llvm#117349
Member
|
@llvm/pr-subscribers-backend-risc-v Author: Craig Topper (topperc) ChangesThe default expansion tries to create i16 operations after type legalization. Fixes #117349 Full diff: https://github.com/llvm/llvm-project/pull/117396.diff 4 Files Affected:
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 6eae756b25fb5b..8ca3e714c32c28 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -498,7 +498,7 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
setOperationAction({ISD::FREM, ISD::FPOW, ISD::FPOWI,
ISD::FCOS, ISD::FSIN, ISD::FSINCOS, ISD::FEXP,
ISD::FEXP2, ISD::FEXP10, ISD::FLOG, ISD::FLOG2,
- ISD::FLOG10},
+ ISD::FLOG10, ISD::FLDEXP},
MVT::f16, Promote);
// FIXME: Need to promote f16 STRICT_* to f32 libcalls, but we don't have
diff --git a/llvm/test/CodeGen/RISCV/double-intrinsics.ll b/llvm/test/CodeGen/RISCV/double-intrinsics.ll
index 3ef128ed6d4cdb..e9fca47c87e6be 100644
--- a/llvm/test/CodeGen/RISCV/double-intrinsics.ll
+++ b/llvm/test/CodeGen/RISCV/double-intrinsics.ll
@@ -1637,3 +1637,56 @@ define double @minimumnum_double(double %x, double %y) {
%z = call double @llvm.minimumnum.f64(double %x, double %y)
ret double %z
}
+
+define double @ldexp_double(double %x, i32 signext %y) nounwind {
+; RV32IFD-LABEL: ldexp_double:
+; RV32IFD: # %bb.0:
+; RV32IFD-NEXT: tail ldexp
+;
+; RV64IFD-LABEL: ldexp_double:
+; RV64IFD: # %bb.0:
+; RV64IFD-NEXT: addi sp, sp, -16
+; RV64IFD-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IFD-NEXT: call ldexp
+; RV64IFD-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IFD-NEXT: addi sp, sp, 16
+; RV64IFD-NEXT: ret
+;
+; RV32IZFINXZDINX-LABEL: ldexp_double:
+; RV32IZFINXZDINX: # %bb.0:
+; RV32IZFINXZDINX-NEXT: addi sp, sp, -16
+; RV32IZFINXZDINX-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFINXZDINX-NEXT: call ldexp
+; RV32IZFINXZDINX-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFINXZDINX-NEXT: addi sp, sp, 16
+; RV32IZFINXZDINX-NEXT: ret
+;
+; RV64IZFINXZDINX-LABEL: ldexp_double:
+; RV64IZFINXZDINX: # %bb.0:
+; RV64IZFINXZDINX-NEXT: addi sp, sp, -16
+; RV64IZFINXZDINX-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFINXZDINX-NEXT: call ldexp
+; RV64IZFINXZDINX-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFINXZDINX-NEXT: addi sp, sp, 16
+; RV64IZFINXZDINX-NEXT: ret
+;
+; RV32I-LABEL: ldexp_double:
+; RV32I: # %bb.0:
+; RV32I-NEXT: addi sp, sp, -16
+; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT: call ldexp
+; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT: addi sp, sp, 16
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: ldexp_double:
+; RV64I: # %bb.0:
+; RV64I-NEXT: addi sp, sp, -16
+; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT: call ldexp
+; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT: addi sp, sp, 16
+; RV64I-NEXT: ret
+ %z = call double @llvm.ldexp.f64.i32(double %x, i32 %y)
+ ret double %z
+}
diff --git a/llvm/test/CodeGen/RISCV/float-intrinsics.ll b/llvm/test/CodeGen/RISCV/float-intrinsics.ll
index 37381aeeb2a0f3..ead777647d5327 100644
--- a/llvm/test/CodeGen/RISCV/float-intrinsics.ll
+++ b/llvm/test/CodeGen/RISCV/float-intrinsics.ll
@@ -2242,3 +2242,51 @@ define float @minimumnum_float(float %x, float %y) {
%z = call float @llvm.minimumnum.f32(float %x, float %y)
ret float %z
}
+
+define float @ldexp_float(float %x, i32 signext %y) nounwind {
+; RV32IF-LABEL: ldexp_float:
+; RV32IF: # %bb.0:
+; RV32IF-NEXT: tail ldexpf
+;
+; RV32IZFINX-LABEL: ldexp_float:
+; RV32IZFINX: # %bb.0:
+; RV32IZFINX-NEXT: tail ldexpf
+;
+; RV64IF-LABEL: ldexp_float:
+; RV64IF: # %bb.0:
+; RV64IF-NEXT: addi sp, sp, -16
+; RV64IF-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IF-NEXT: call ldexpf
+; RV64IF-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IF-NEXT: addi sp, sp, 16
+; RV64IF-NEXT: ret
+;
+; RV64IZFINX-LABEL: ldexp_float:
+; RV64IZFINX: # %bb.0:
+; RV64IZFINX-NEXT: addi sp, sp, -16
+; RV64IZFINX-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFINX-NEXT: call ldexpf
+; RV64IZFINX-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFINX-NEXT: addi sp, sp, 16
+; RV64IZFINX-NEXT: ret
+;
+; RV32I-LABEL: ldexp_float:
+; RV32I: # %bb.0:
+; RV32I-NEXT: addi sp, sp, -16
+; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT: call ldexpf
+; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT: addi sp, sp, 16
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: ldexp_float:
+; RV64I: # %bb.0:
+; RV64I-NEXT: addi sp, sp, -16
+; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT: call ldexpf
+; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT: addi sp, sp, 16
+; RV64I-NEXT: ret
+ %z = call float @llvm.ldexp.f32.i32(float %x, i32 %y)
+ ret float %z
+}
diff --git a/llvm/test/CodeGen/RISCV/half-intrinsics.ll b/llvm/test/CodeGen/RISCV/half-intrinsics.ll
index 0d26e660c979bb..f18a7dc1c0c6b0 100644
--- a/llvm/test/CodeGen/RISCV/half-intrinsics.ll
+++ b/llvm/test/CodeGen/RISCV/half-intrinsics.ll
@@ -3186,3 +3186,129 @@ define half @minimumnum_half(half %x, half %y) {
%z = call half @llvm.minimumnum.f16(half %x, half %y)
ret half %z
}
+
+define half @ldexp_half(half %x, i32 signext %y) nounwind {
+; RV32IZFH-LABEL: ldexp_half:
+; RV32IZFH: # %bb.0:
+; RV32IZFH-NEXT: addi sp, sp, -16
+; RV32IZFH-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFH-NEXT: fcvt.s.h fa0, fa0
+; RV32IZFH-NEXT: call ldexpf
+; RV32IZFH-NEXT: fcvt.h.s fa0, fa0
+; RV32IZFH-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFH-NEXT: addi sp, sp, 16
+; RV32IZFH-NEXT: ret
+;
+; RV64IZFH-LABEL: ldexp_half:
+; RV64IZFH: # %bb.0:
+; RV64IZFH-NEXT: addi sp, sp, -16
+; RV64IZFH-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFH-NEXT: fcvt.s.h fa0, fa0
+; RV64IZFH-NEXT: call ldexpf
+; RV64IZFH-NEXT: fcvt.h.s fa0, fa0
+; RV64IZFH-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFH-NEXT: addi sp, sp, 16
+; RV64IZFH-NEXT: ret
+;
+; RV32IZHINX-LABEL: ldexp_half:
+; RV32IZHINX: # %bb.0:
+; RV32IZHINX-NEXT: addi sp, sp, -16
+; RV32IZHINX-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZHINX-NEXT: fcvt.s.h a0, a0
+; RV32IZHINX-NEXT: call ldexpf
+; RV32IZHINX-NEXT: fcvt.h.s a0, a0
+; RV32IZHINX-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZHINX-NEXT: addi sp, sp, 16
+; RV32IZHINX-NEXT: ret
+;
+; RV64IZHINX-LABEL: ldexp_half:
+; RV64IZHINX: # %bb.0:
+; RV64IZHINX-NEXT: addi sp, sp, -16
+; RV64IZHINX-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZHINX-NEXT: fcvt.s.h a0, a0
+; RV64IZHINX-NEXT: call ldexpf
+; RV64IZHINX-NEXT: fcvt.h.s a0, a0
+; RV64IZHINX-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZHINX-NEXT: addi sp, sp, 16
+; RV64IZHINX-NEXT: ret
+;
+; RV32I-LABEL: ldexp_half:
+; RV32I: # %bb.0:
+; RV32I-NEXT: addi sp, sp, -16
+; RV32I-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32I-NEXT: sw s0, 8(sp) # 4-byte Folded Spill
+; RV32I-NEXT: mv s0, a1
+; RV32I-NEXT: slli a0, a0, 16
+; RV32I-NEXT: srli a0, a0, 16
+; RV32I-NEXT: call __extendhfsf2
+; RV32I-NEXT: mv a1, s0
+; RV32I-NEXT: call ldexpf
+; RV32I-NEXT: call __truncsfhf2
+; RV32I-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32I-NEXT: lw s0, 8(sp) # 4-byte Folded Reload
+; RV32I-NEXT: addi sp, sp, 16
+; RV32I-NEXT: ret
+;
+; RV64I-LABEL: ldexp_half:
+; RV64I: # %bb.0:
+; RV64I-NEXT: addi sp, sp, -16
+; RV64I-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64I-NEXT: sd s0, 0(sp) # 8-byte Folded Spill
+; RV64I-NEXT: mv s0, a1
+; RV64I-NEXT: slli a0, a0, 48
+; RV64I-NEXT: srli a0, a0, 48
+; RV64I-NEXT: call __extendhfsf2
+; RV64I-NEXT: mv a1, s0
+; RV64I-NEXT: call ldexpf
+; RV64I-NEXT: call __truncsfhf2
+; RV64I-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64I-NEXT: ld s0, 0(sp) # 8-byte Folded Reload
+; RV64I-NEXT: addi sp, sp, 16
+; RV64I-NEXT: ret
+;
+; RV32IZFHMIN-LABEL: ldexp_half:
+; RV32IZFHMIN: # %bb.0:
+; RV32IZFHMIN-NEXT: addi sp, sp, -16
+; RV32IZFHMIN-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZFHMIN-NEXT: fcvt.s.h fa0, fa0
+; RV32IZFHMIN-NEXT: call ldexpf
+; RV32IZFHMIN-NEXT: fcvt.h.s fa0, fa0
+; RV32IZFHMIN-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZFHMIN-NEXT: addi sp, sp, 16
+; RV32IZFHMIN-NEXT: ret
+;
+; RV64IZFHMIN-LABEL: ldexp_half:
+; RV64IZFHMIN: # %bb.0:
+; RV64IZFHMIN-NEXT: addi sp, sp, -16
+; RV64IZFHMIN-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZFHMIN-NEXT: fcvt.s.h fa0, fa0
+; RV64IZFHMIN-NEXT: call ldexpf
+; RV64IZFHMIN-NEXT: fcvt.h.s fa0, fa0
+; RV64IZFHMIN-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZFHMIN-NEXT: addi sp, sp, 16
+; RV64IZFHMIN-NEXT: ret
+;
+; RV32IZHINXMIN-LABEL: ldexp_half:
+; RV32IZHINXMIN: # %bb.0:
+; RV32IZHINXMIN-NEXT: addi sp, sp, -16
+; RV32IZHINXMIN-NEXT: sw ra, 12(sp) # 4-byte Folded Spill
+; RV32IZHINXMIN-NEXT: fcvt.s.h a0, a0
+; RV32IZHINXMIN-NEXT: call ldexpf
+; RV32IZHINXMIN-NEXT: fcvt.h.s a0, a0
+; RV32IZHINXMIN-NEXT: lw ra, 12(sp) # 4-byte Folded Reload
+; RV32IZHINXMIN-NEXT: addi sp, sp, 16
+; RV32IZHINXMIN-NEXT: ret
+;
+; RV64IZHINXMIN-LABEL: ldexp_half:
+; RV64IZHINXMIN: # %bb.0:
+; RV64IZHINXMIN-NEXT: addi sp, sp, -16
+; RV64IZHINXMIN-NEXT: sd ra, 8(sp) # 8-byte Folded Spill
+; RV64IZHINXMIN-NEXT: fcvt.s.h a0, a0
+; RV64IZHINXMIN-NEXT: call ldexpf
+; RV64IZHINXMIN-NEXT: fcvt.h.s a0, a0
+; RV64IZHINXMIN-NEXT: ld ra, 8(sp) # 8-byte Folded Reload
+; RV64IZHINXMIN-NEXT: addi sp, sp, 16
+; RV64IZHINXMIN-NEXT: ret
+ %z = call half @llvm.ldexp.f16.i32(half %x, i32 %y)
+ ret half %z
+}
|
arsenm
approved these changes
Nov 22, 2024
You can test this locally with the following command:git-clang-format --diff b4e000e6005bd0f11240133aa335efcbb8424a23 c1f67428e89ecb04f836b6c88abd43d55ca22fa5 --extensions cpp -- llvm/lib/Target/RISCV/RISCVISelLowering.cppView the diff from clang-format here.diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 8ca3e714c3..a09a83a864 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -495,10 +495,9 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
setOperationAction(
ISD::FNEARBYINT, MVT::f16,
Subtarget.hasStdExtZfh() && Subtarget.hasStdExtZfa() ? Legal : Promote);
- setOperationAction({ISD::FREM, ISD::FPOW, ISD::FPOWI,
- ISD::FCOS, ISD::FSIN, ISD::FSINCOS, ISD::FEXP,
- ISD::FEXP2, ISD::FEXP10, ISD::FLOG, ISD::FLOG2,
- ISD::FLOG10, ISD::FLDEXP},
+ setOperationAction({ISD::FREM, ISD::FPOW, ISD::FPOWI, ISD::FCOS, ISD::FSIN,
+ ISD::FSINCOS, ISD::FEXP, ISD::FEXP2, ISD::FEXP10,
+ ISD::FLOG, ISD::FLOG2, ISD::FLOG10, ISD::FLDEXP},
MVT::f16, Promote);
// FIXME: Need to promote f16 STRICT_* to f32 libcalls, but we don't have
|
lukel97
approved these changes
Nov 25, 2024
Contributor
lukel97
left a comment
lukel97
left a comment
There was a problem hiding this comment.
Do we also need to promote frexp?
Collaborator
Author
Probably. I'll check later today |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
The default expansion tries to create i16 operations after type legalization.
Fixes #117349