Skip to content

[RISCV] Mark sincos libcalls as available for RISC-V #168708

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.

Sign up for GitHub

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

Jump to bottom
Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[RISCV] Mark sincos libcalls as available for RISC-V #168708

wants to merge 1 commit into from
+17 −24

Conversation

@asb
Copy link
Contributor

@asb asb commented Nov 19, 2025

A standard glibc build for RISC-V does contain these libcalls, and GCC will generate calls to them. Found by comparing benchmark data for SPEC, spotting that the GCC build of 526.blender_r spends 3.02% of dynamically executed instructions in libm functions, while the Clang build spends 3.41%. Calling sincos rather than separate sin and cos functions is the reason for the difference.

@asb
[RISCV] Mark sincos libcalls as available for RISC-V
f8d8c00 
A standard glibc build for RISC-V does contain these libcalls, and GCC
will generate them. Found by comparing benchmark data for SPEC, spotting
that the GCC build of 526.blender_r spends 3.02% of dynamically executed
instructions in libm functions, while the Clang build spends 3.41%. And
calling sincos vs separate sin and cos functions is the reason for the
difference.
@llvmbot
Copy link
Member

llvmbot commented Nov 19, 2025

@llvm/pr-subscribers-llvm-ir

Author: Alex Bradbury (asb)

Changes

A standard glibc build for RISC-V does contain these libcalls, and GCC will generate calls to them. Found by comparing benchmark data for SPEC, spotting that the GCC build of 526.blender_r spends 3.02% of dynamically executed instructions in libm functions, while the Clang build spends 3.41%. Calling sincos rather than separate sin and cos functions is the reason for the difference.

Full diff: https://github.com/llvm/llvm-project/pull/168708.diff

2 Files Affected:

  • (modified) llvm/include/llvm/IR/RuntimeLibcalls.td (+1)
  • (modified) llvm/test/CodeGen/RISCV/sincos-expansion.ll (+16-24)
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.td b/llvm/include/llvm/IR/RuntimeLibcalls.td
index ce7e836f66446..9230b39e6f45e 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.td
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.td
@@ -2452,6 +2452,7 @@ def RISCVSystemLibrary
       (add DefaultRuntimeLibcallImpls,
            exp10f, exp10, exp10l_f128,
            __riscv_flush_icache,
+           LibmHasSinCosF32, LibmHasSinCosF64, LibmHasSinCosF128,
            LibcallImpls<(add Int128RTLibcalls), isRISCV64>,
            DefaultStackProtector)>;
 
diff --git a/llvm/test/CodeGen/RISCV/sincos-expansion.ll b/llvm/test/CodeGen/RISCV/sincos-expansion.ll
index 6f37d67134cb3..606768a8e2f53 100644
--- a/llvm/test/CodeGen/RISCV/sincos-expansion.ll
+++ b/llvm/test/CodeGen/RISCV/sincos-expansion.ll
@@ -7,20 +7,16 @@
 define float @test_sincos_f32(float %f) nounwind {
 ; CHECK-LABEL: test_sincos_f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    addi sp, sp, -32
-; CHECK-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs0, 16(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs1, 8(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fmv.s fs0, fa0
-; CHECK-NEXT:    call sinf
-; CHECK-NEXT:    fmv.s fs1, fa0
-; CHECK-NEXT:    fmv.s fa0, fs0
-; CHECK-NEXT:    call cosf
-; CHECK-NEXT:    fadd.s fa0, fs1, fa0
-; CHECK-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs0, 16(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs1, 8(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    addi sp, sp, 32
+; CHECK-NEXT:    addi sp, sp, -16
+; CHECK-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    addi a0, sp, 4
+; CHECK-NEXT:    mv a1, sp
+; CHECK-NEXT:    call sincosf
+; CHECK-NEXT:    flw fa5, 0(sp)
+; CHECK-NEXT:    flw fa4, 4(sp)
+; CHECK-NEXT:    fadd.s fa0, fa4, fa5
+; CHECK-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    addi sp, sp, 16
 ; CHECK-NEXT:    ret
   %sin = call float @sinf(float %f) readnone
   %cos = call float @cosf(float %f) readnone
@@ -57,17 +53,13 @@ define double @test_sincos_f64(double %f) nounwind {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    addi sp, sp, -32
 ; CHECK-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs0, 16(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs1, 8(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fmv.d fs0, fa0
-; CHECK-NEXT:    call sin
-; CHECK-NEXT:    fmv.d fs1, fa0
-; CHECK-NEXT:    fmv.d fa0, fs0
-; CHECK-NEXT:    call cos
-; CHECK-NEXT:    fadd.d fa0, fs1, fa0
+; CHECK-NEXT:    addi a0, sp, 16
+; CHECK-NEXT:    addi a1, sp, 8
+; CHECK-NEXT:    call sincos
+; CHECK-NEXT:    fld fa5, 8(sp)
+; CHECK-NEXT:    fld fa4, 16(sp)
+; CHECK-NEXT:    fadd.d fa0, fa4, fa5
 ; CHECK-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs0, 16(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs1, 8(sp) # 8-byte Folded Reload
 ; CHECK-NEXT:    addi sp, sp, 32
 ; CHECK-NEXT:    ret
   %sin = call double @sin(double %f) readnone

@llvmbot
Copy link
Member

llvmbot commented Nov 19, 2025

@llvm/pr-subscribers-backend-risc-v

Author: Alex Bradbury (asb)

Changes

A standard glibc build for RISC-V does contain these libcalls, and GCC will generate calls to them. Found by comparing benchmark data for SPEC, spotting that the GCC build of 526.blender_r spends 3.02% of dynamically executed instructions in libm functions, while the Clang build spends 3.41%. Calling sincos rather than separate sin and cos functions is the reason for the difference.

Full diff: https://github.com/llvm/llvm-project/pull/168708.diff

2 Files Affected:

  • (modified) llvm/include/llvm/IR/RuntimeLibcalls.td (+1)
  • (modified) llvm/test/CodeGen/RISCV/sincos-expansion.ll (+16-24)
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.td b/llvm/include/llvm/IR/RuntimeLibcalls.td
index ce7e836f66446..9230b39e6f45e 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.td
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.td
@@ -2452,6 +2452,7 @@ def RISCVSystemLibrary
       (add DefaultRuntimeLibcallImpls,
            exp10f, exp10, exp10l_f128,
            __riscv_flush_icache,
+           LibmHasSinCosF32, LibmHasSinCosF64, LibmHasSinCosF128,
            LibcallImpls<(add Int128RTLibcalls), isRISCV64>,
            DefaultStackProtector)>;
 
diff --git a/llvm/test/CodeGen/RISCV/sincos-expansion.ll b/llvm/test/CodeGen/RISCV/sincos-expansion.ll
index 6f37d67134cb3..606768a8e2f53 100644
--- a/llvm/test/CodeGen/RISCV/sincos-expansion.ll
+++ b/llvm/test/CodeGen/RISCV/sincos-expansion.ll
@@ -7,20 +7,16 @@
 define float @test_sincos_f32(float %f) nounwind {
 ; CHECK-LABEL: test_sincos_f32:
 ; CHECK:       # %bb.0:
-; CHECK-NEXT:    addi sp, sp, -32
-; CHECK-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs0, 16(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs1, 8(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fmv.s fs0, fa0
-; CHECK-NEXT:    call sinf
-; CHECK-NEXT:    fmv.s fs1, fa0
-; CHECK-NEXT:    fmv.s fa0, fs0
-; CHECK-NEXT:    call cosf
-; CHECK-NEXT:    fadd.s fa0, fs1, fa0
-; CHECK-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs0, 16(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs1, 8(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    addi sp, sp, 32
+; CHECK-NEXT:    addi sp, sp, -16
+; CHECK-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill
+; CHECK-NEXT:    addi a0, sp, 4
+; CHECK-NEXT:    mv a1, sp
+; CHECK-NEXT:    call sincosf
+; CHECK-NEXT:    flw fa5, 0(sp)
+; CHECK-NEXT:    flw fa4, 4(sp)
+; CHECK-NEXT:    fadd.s fa0, fa4, fa5
+; CHECK-NEXT:    ld ra, 8(sp) # 8-byte Folded Reload
+; CHECK-NEXT:    addi sp, sp, 16
 ; CHECK-NEXT:    ret
   %sin = call float @sinf(float %f) readnone
   %cos = call float @cosf(float %f) readnone
@@ -57,17 +53,13 @@ define double @test_sincos_f64(double %f) nounwind {
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    addi sp, sp, -32
 ; CHECK-NEXT:    sd ra, 24(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs0, 16(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fsd fs1, 8(sp) # 8-byte Folded Spill
-; CHECK-NEXT:    fmv.d fs0, fa0
-; CHECK-NEXT:    call sin
-; CHECK-NEXT:    fmv.d fs1, fa0
-; CHECK-NEXT:    fmv.d fa0, fs0
-; CHECK-NEXT:    call cos
-; CHECK-NEXT:    fadd.d fa0, fs1, fa0
+; CHECK-NEXT:    addi a0, sp, 16
+; CHECK-NEXT:    addi a1, sp, 8
+; CHECK-NEXT:    call sincos
+; CHECK-NEXT:    fld fa5, 8(sp)
+; CHECK-NEXT:    fld fa4, 16(sp)
+; CHECK-NEXT:    fadd.d fa0, fa4, fa5
 ; CHECK-NEXT:    ld ra, 24(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs0, 16(sp) # 8-byte Folded Reload
-; CHECK-NEXT:    fld fs1, 8(sp) # 8-byte Folded Reload
 ; CHECK-NEXT:    addi sp, sp, 32
 ; CHECK-NEXT:    ret
   %sin = call double @sin(double %f) readnone

lukel97
lukel97 reviewed Nov 19, 2025
View reviewed changes
llvm/include/llvm/IR/RuntimeLibcalls.td
(add DefaultRuntimeLibcallImpls,
exp10f, exp10, exp10l_f128,
__riscv_flush_icache,
LibmHasSinCosF32, LibmHasSinCosF64, LibmHasSinCosF128,
Copy link
Contributor

@lukel97 lukel97 Nov 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Is it possible to test the sincos f128 libcall?

Copy link
Contributor Author

@asb asb Nov 19, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looking at AArch64 it seems it should kick in for f128 as well and it's not for some reason. I'll look into that.

@github-actions
Copy link

github-actions bot commented Nov 19, 2025

🐧 Linux x64 Test Results

  • 186387 tests passed
  • 4855 tests skipped

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@lukel97 lukel97 lukel97 left review comments

@lenary lenary Awaiting requested review from lenary

@arsenm arsenm Awaiting requested review from arsenm

@preames preames Awaiting requested review from preames

@topperc topperc Awaiting requested review from topperc

Assignees

No one assigned

Labels

backend:RISC-V llvm:ir

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@asb @llvmbot @lukel97