[llvm-branch-commits] [llvm] RuntimeLibcalls: Use element alignment for vector libcall out pointers (PR #237486)
Matt Arsenault via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Sat Oct 31 21:36:21 PDT 2026
https://github.com/arsenm updated https://github.com/llvm/llvm-project/pull/237486
>From b59267394a10a6e1e6c592c9655877161c8d5088 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Fri, 30 Oct 2026 22:03:30 -0700
Subject: [PATCH 1/2] RuntimeLibcalls: Use element alignment for vector libcall
out pointers
The pointer arguments of the vector modf, sincos and sincospi functions
are linear pointers to the scalar element type (e.g. the l4 in
_ZGVnN4vl4l4_sincosf), not pointers to the vector type. Only the element
alignment is guaranteed, so stop declaring them with the ABI alignment
of the vector.
Co-Authored-By: Claude Opus 5.5 <noreply at anthropic.com>
---
llvm/lib/IR/RuntimeLibcalls.cpp | 2 +-
.../Util/DeclareRuntimeLibcalls/amdlibm.ll | 12 +++++-----
.../Util/DeclareRuntimeLibcalls/armpl.ll | 24 +++++++++----------
.../DeclareRuntimeLibcalls/sleef-riscv.ll | 12 +++++-----
.../Util/DeclareRuntimeLibcalls/sleef.ll | 24 +++++++++----------
5 files changed, 37 insertions(+), 37 deletions(-)
diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index db67c3f1ac01cb..8ed7fc27ac024c 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -645,7 +645,7 @@ RuntimeLibcallsInfo::getFunctionTy(LLVMContext &Ctx, const Triple &TT,
AttrBuilder ArgAttrBuilder(Ctx);
for (Attribute::AttrKind AK : CommonPtrArgAttrs)
ArgAttrBuilder.addAttribute(AK);
- ArgAttrBuilder.addAlignmentAttr(DL.getABITypeAlign(VecTy));
+ ArgAttrBuilder.addAlignmentAttr(DL.getABITypeAlign(VecTy->getScalarType()));
bool HasPtrArg = false;
for (auto [Idx, ParamTy] : enumerate(FuncTy->params())) {
diff --git a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll
index d40ed527442b23..11f89866e61091 100644
--- a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll
+++ b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll
@@ -47,7 +47,7 @@
; CHECK: declare <2 x double> @amd_vrd2_sin(<2 x double>) [[ATTRS]]
-; CHECK: declare void @amd_vrd2_sincos(<2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG:#[0-9]+]]
+; CHECK: declare void @amd_vrd2_sincos(<2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG:#[0-9]+]]
; CHECK: declare <2 x double> @amd_vrd2_tan(<2 x double>) [[ATTRS]]
@@ -93,7 +93,7 @@
; CHECK: declare <4 x double> @amd_vrd4_sin(<4 x double>) [[ATTRS]]
-; CHECK: declare void @amd_vrd4_sincos(<4 x double>, ptr noalias nonnull writeonly align 32, ptr noalias nonnull writeonly align 32) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrd4_sincos(<4 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
; CHECK: declare <4 x double> @amd_vrd4_tan(<4 x double>) [[ATTRS]]
@@ -139,7 +139,7 @@
; CHECK: declare <8 x double> @amd_vrd8_sin(<8 x double>) [[ATTRS]]
-; CHECK: declare void @amd_vrd8_sincos(<8 x double>, ptr noalias nonnull writeonly align 64, ptr noalias nonnull writeonly align 64) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrd8_sincos(<8 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
; CHECK: declare <8 x double> @amd_vrd8_tan(<8 x double>) [[ATTRS]]
@@ -175,7 +175,7 @@
; CHECK: declare <16 x float> @amd_vrs16_roundf(<16 x float>) [[ATTRS]]
-; CHECK: declare void @amd_vrs16_sincosf(<16 x float>, ptr noalias nonnull writeonly align 64, ptr noalias nonnull writeonly align 64) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrs16_sincosf(<16 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare <16 x float> @amd_vrs16_sinf(<16 x float>) [[ATTRS]]
@@ -219,7 +219,7 @@
; CHECK: declare <4 x float> @amd_vrs4_roundf(<4 x float>) [[ATTRS]]
-; CHECK: declare void @amd_vrs4_sincosf(<4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrs4_sincosf(<4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare <4 x float> @amd_vrs4_sinf(<4 x float>) [[ATTRS]]
@@ -259,7 +259,7 @@
; CHECK: declare <8 x float> @amd_vrs8_roundf(<8 x float>) [[ATTRS]]
-; CHECK: declare void @amd_vrs8_sincosf(<8 x float>, ptr noalias nonnull writeonly align 32, ptr noalias nonnull writeonly align 32) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrs8_sincosf(<8 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare <8 x float> @amd_vrs8_sinf(<8 x float>) [[ATTRS]]
diff --git a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/armpl.ll b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/armpl.ll
index 9d7b72fe172e0f..9e1bf15eeaf70f 100644
--- a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/armpl.ll
+++ b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/armpl.ll
@@ -135,11 +135,11 @@
; CHECK: declare void @armpl_svmodf_f32_stret_x(...)
-; CHECK: declare <vscale x 4 x float> @armpl_svmodf_f32_x(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, <vscale x 4 x i1>) [[ATTRS_PTR_ARG:#[0-9]+]]
+; CHECK: declare <vscale x 4 x float> @armpl_svmodf_f32_x(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, <vscale x 4 x i1>) [[ATTRS_PTR_ARG:#[0-9]+]]
; CHECK: declare void @armpl_svmodf_f64_stret_x(...)
-; CHECK: declare <vscale x 2 x double> @armpl_svmodf_f64_x(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, <vscale x 2 x i1>) [[ATTRS_PTR_ARG]]
+; CHECK: declare <vscale x 2 x double> @armpl_svmodf_f64_x(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, <vscale x 2 x i1>) [[ATTRS_PTR_ARG]]
; CHECK: declare <vscale x 4 x float> @armpl_svnextafter_f32_x(<vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x i1>) [[ATTRS]]
@@ -153,13 +153,13 @@
; CHECK: declare <vscale x 2 x double> @armpl_svsin_f64_x(<vscale x 2 x double>, <vscale x 2 x i1>) [[ATTRS]]
-; CHECK: declare void @armpl_svsincos_f32_x(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16, <vscale x 4 x i1>) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @armpl_svsincos_f32_x(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4, <vscale x 4 x i1>) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @armpl_svsincos_f64_x(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16, <vscale x 2 x i1>) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @armpl_svsincos_f64_x(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8, <vscale x 2 x i1>) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @armpl_svsincospi_f32_x(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16, <vscale x 4 x i1>) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @armpl_svsincospi_f32_x(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4, <vscale x 4 x i1>) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @armpl_svsincospi_f64_x(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16, <vscale x 2 x i1>) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @armpl_svsincospi_f64_x(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8, <vscale x 2 x i1>) [[ATTRS_PTR_ARG]]
; CHECK: declare <vscale x 4 x float> @armpl_svsinh_f32_x(<vscale x 4 x float>, <vscale x 4 x i1>) [[ATTRS]]
@@ -317,11 +317,11 @@
; CHECK: declare aarch64_vector_pcs <2 x double> @armpl_vlogq_f64(<2 x double>) [[ATTRS]]
-; CHECK: declare aarch64_vector_pcs <4 x float> @armpl_vmodfq_f32(<4 x float>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare aarch64_vector_pcs <4 x float> @armpl_vmodfq_f32(<4 x float>, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare aarch64_vector_pcs void @armpl_vmodfq_f32_stret(...)
-; CHECK: declare aarch64_vector_pcs <2 x double> @armpl_vmodfq_f64(<2 x double>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare aarch64_vector_pcs <2 x double> @armpl_vmodfq_f64(<2 x double>, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
; CHECK: declare aarch64_vector_pcs void @armpl_vmodfq_f64_stret(...)
@@ -333,13 +333,13 @@
; CHECK: declare aarch64_vector_pcs <2 x double> @armpl_vpowq_f64(<2 x double>, <2 x double>) [[ATTRS]]
-; CHECK: declare aarch64_vector_pcs void @armpl_vsincospiq_f32(<4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare aarch64_vector_pcs void @armpl_vsincospiq_f32(<4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare aarch64_vector_pcs void @armpl_vsincospiq_f64(<2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare aarch64_vector_pcs void @armpl_vsincospiq_f64(<2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
-; CHECK: declare aarch64_vector_pcs void @armpl_vsincosq_f32(<4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare aarch64_vector_pcs void @armpl_vsincosq_f32(<4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare aarch64_vector_pcs void @armpl_vsincosq_f64(<2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare aarch64_vector_pcs void @armpl_vsincosq_f64(<2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
; CHECK: declare aarch64_vector_pcs <4 x float> @armpl_vsinhq_f32(<4 x float>) [[ATTRS]]
diff --git a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef-riscv.ll b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef-riscv.ll
index 1a634ffdf1d1ab..1607726efabc1e 100644
--- a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef-riscv.ll
+++ b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef-riscv.ll
@@ -125,9 +125,9 @@
; CHECK: declare <vscale x 4 x float> @Sleef_logfx_u10rvvm2(<vscale x 4 x float>) [[ATTRS]]
-; CHECK: declare <vscale x 2 x double> @Sleef_modfdx_rvvm2(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG:#[0-9]+]]
+; CHECK: declare <vscale x 2 x double> @Sleef_modfdx_rvvm2(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG:#[0-9]+]]
-; CHECK: declare <vscale x 4 x float> @Sleef_modffx_rvvm2(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare <vscale x 4 x float> @Sleef_modffx_rvvm2(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare <vscale x 2 x double> @Sleef_nextafterdx_rvvm2(<vscale x 2 x double>, <vscale x 2 x double>) [[ATTRS]]
@@ -137,13 +137,13 @@
; CHECK: declare <vscale x 4 x float> @Sleef_powfx_u10rvvm2(<vscale x 4 x float>, <vscale x 4 x float>) [[ATTRS]]
-; CHECK: declare void @Sleef_sincosdx_u10rvvm2(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @Sleef_sincosdx_u10rvvm2(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @Sleef_sincosfx_u10rvvm2(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @Sleef_sincosfx_u10rvvm2(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @Sleef_sincospidx_u10rvvm2(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @Sleef_sincospidx_u10rvvm2(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @Sleef_sincospifx_u10rvvm2(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @Sleef_sincospifx_u10rvvm2(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare <vscale x 2 x double> @Sleef_sindx_u10rvvm2(<vscale x 2 x double>) [[ATTRS]]
diff --git a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef.ll b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef.ll
index 0a76beb039b5bf..916fa049b9f55b 100644
--- a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef.ll
+++ b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/sleef.ll
@@ -59,11 +59,11 @@
; CHECK: declare <2 x double> @_ZGVnN2v_tgamma(<2 x double>) [[ATTRS]]
-; CHECK: declare <2 x double> @_ZGVnN2vl8_modf(<2 x double>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG:#[0-9]+]]
+; CHECK: declare <2 x double> @_ZGVnN2vl8_modf(<2 x double>, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG:#[0-9]+]]
-; CHECK: declare void @_ZGVnN2vl8l8_sincos(<2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVnN2vl8l8_sincos(<2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVnN2vl8l8_sincospi(<2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVnN2vl8l8_sincospi(<2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
; CHECK: declare <2 x double> @_ZGVnN2vv_atan2(<2 x double>, <2 x double>) [[ATTRS]]
@@ -145,11 +145,11 @@
; CHECK: declare <4 x float> @_ZGVnN4v_tgammaf(<4 x float>) [[ATTRS]]
-; CHECK: declare <4 x float> @_ZGVnN4vl4_modff(<4 x float>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare <4 x float> @_ZGVnN4vl4_modff(<4 x float>, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVnN4vl4l4_sincosf(<4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVnN4vl4l4_sincosf(<4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVnN4vl4l4_sincospif(<4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVnN4vl4l4_sincospif(<4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
; CHECK: declare <4 x float> @_ZGVnN4vv_atan2f(<4 x float>, <4 x float>) [[ATTRS]]
@@ -333,17 +333,17 @@
; CHECK: declare <vscale x 4 x float> @_ZGVsMxvvv_fmaf(<vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x float>, <vscale x 4 x i1>) [[ATTRS]]
-; CHECK: declare <vscale x 4 x float> @_ZGVsNxvl4_modff(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare <vscale x 4 x float> @_ZGVsNxvl4_modff(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVsNxvl4l4_sincosf(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVsNxvl4l4_sincosf(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVsNxvl4l4_sincospif(<vscale x 4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVsNxvl4l4_sincospif(<vscale x 4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
-; CHECK: declare <vscale x 2 x double> @_ZGVsNxvl8_modf(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare <vscale x 2 x double> @_ZGVsNxvl8_modf(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVsNxvl8l8_sincos(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVsNxvl8l8_sincos(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
-; CHECK: declare void @_ZGVsNxvl8l8_sincospi(<vscale x 2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @_ZGVsNxvl8l8_sincospi(<vscale x 2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
; CHECK: attributes [[ATTRS]] = { nocallback nofree nosync nounwind willreturn }
; CHECK: attributes [[ATTRS_PTR_ARG]] = { nocallback nofree nosync nounwind willreturn memory(argmem: write) }
>From f40a3840020e353087634662fb44b921e6348213 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Sat, 31 Oct 2026 10:10:54 -0700
Subject: [PATCH 2/2] Use vector alignment for AMDLIBM sincos out pointers
---
llvm/lib/IR/RuntimeLibcalls.cpp | 16 +++++++++++++++-
.../Util/DeclareRuntimeLibcalls/amdlibm.ll | 12 ++++++------
2 files changed, 21 insertions(+), 7 deletions(-)
diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index 8ed7fc27ac024c..62fdab520ff53e 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -642,10 +642,24 @@ RuntimeLibcallsInfo::getFunctionTy(LLVMContext &Ctx, const Triple &TT,
return {FuncTy, Attrs};
Type *VecTy = FuncTy->getParamType(0);
+ Align PtrAlign = DL.getABITypeAlign(VecTy->getScalarType());
+ switch (LibcallImpl) {
+ case RTLIB::impl_amd_vrd2_sincos:
+ case RTLIB::impl_amd_vrd4_sincos:
+ case RTLIB::impl_amd_vrd8_sincos:
+ case RTLIB::impl_amd_vrs4_sincosf:
+ case RTLIB::impl_amd_vrs8_sincosf:
+ case RTLIB::impl_amd_vrs16_sincosf:
+ PtrAlign = DL.getABITypeAlign(VecTy);
+ break;
+ default:
+ break;
+ }
+
AttrBuilder ArgAttrBuilder(Ctx);
for (Attribute::AttrKind AK : CommonPtrArgAttrs)
ArgAttrBuilder.addAttribute(AK);
- ArgAttrBuilder.addAlignmentAttr(DL.getABITypeAlign(VecTy->getScalarType()));
+ ArgAttrBuilder.addAlignmentAttr(PtrAlign);
bool HasPtrArg = false;
for (auto [Idx, ParamTy] : enumerate(FuncTy->params())) {
diff --git a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll
index 11f89866e61091..d40ed527442b23 100644
--- a/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll
+++ b/llvm/test/Transforms/Util/DeclareRuntimeLibcalls/amdlibm.ll
@@ -47,7 +47,7 @@
; CHECK: declare <2 x double> @amd_vrd2_sin(<2 x double>) [[ATTRS]]
-; CHECK: declare void @amd_vrd2_sincos(<2 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG:#[0-9]+]]
+; CHECK: declare void @amd_vrd2_sincos(<2 x double>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG:#[0-9]+]]
; CHECK: declare <2 x double> @amd_vrd2_tan(<2 x double>) [[ATTRS]]
@@ -93,7 +93,7 @@
; CHECK: declare <4 x double> @amd_vrd4_sin(<4 x double>) [[ATTRS]]
-; CHECK: declare void @amd_vrd4_sincos(<4 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrd4_sincos(<4 x double>, ptr noalias nonnull writeonly align 32, ptr noalias nonnull writeonly align 32) [[ATTRS_PTR_ARG]]
; CHECK: declare <4 x double> @amd_vrd4_tan(<4 x double>) [[ATTRS]]
@@ -139,7 +139,7 @@
; CHECK: declare <8 x double> @amd_vrd8_sin(<8 x double>) [[ATTRS]]
-; CHECK: declare void @amd_vrd8_sincos(<8 x double>, ptr noalias nonnull writeonly align 8, ptr noalias nonnull writeonly align 8) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrd8_sincos(<8 x double>, ptr noalias nonnull writeonly align 64, ptr noalias nonnull writeonly align 64) [[ATTRS_PTR_ARG]]
; CHECK: declare <8 x double> @amd_vrd8_tan(<8 x double>) [[ATTRS]]
@@ -175,7 +175,7 @@
; CHECK: declare <16 x float> @amd_vrs16_roundf(<16 x float>) [[ATTRS]]
-; CHECK: declare void @amd_vrs16_sincosf(<16 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrs16_sincosf(<16 x float>, ptr noalias nonnull writeonly align 64, ptr noalias nonnull writeonly align 64) [[ATTRS_PTR_ARG]]
; CHECK: declare <16 x float> @amd_vrs16_sinf(<16 x float>) [[ATTRS]]
@@ -219,7 +219,7 @@
; CHECK: declare <4 x float> @amd_vrs4_roundf(<4 x float>) [[ATTRS]]
-; CHECK: declare void @amd_vrs4_sincosf(<4 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrs4_sincosf(<4 x float>, ptr noalias nonnull writeonly align 16, ptr noalias nonnull writeonly align 16) [[ATTRS_PTR_ARG]]
; CHECK: declare <4 x float> @amd_vrs4_sinf(<4 x float>) [[ATTRS]]
@@ -259,7 +259,7 @@
; CHECK: declare <8 x float> @amd_vrs8_roundf(<8 x float>) [[ATTRS]]
-; CHECK: declare void @amd_vrs8_sincosf(<8 x float>, ptr noalias nonnull writeonly align 4, ptr noalias nonnull writeonly align 4) [[ATTRS_PTR_ARG]]
+; CHECK: declare void @amd_vrs8_sincosf(<8 x float>, ptr noalias nonnull writeonly align 32, ptr noalias nonnull writeonly align 32) [[ATTRS_PTR_ARG]]
; CHECK: declare <8 x float> @amd_vrs8_sinf(<8 x float>) [[ATTRS]]
More information about the llvm-branch-commits
mailing list