From a3b104176cc416bc2083ba46a8edf6efd489230a Mon Sep 17 00:00:00 2001
From: Trevor Gross <tgross@intrepidcs.com>
Date: Thu, 1 May 2025 00:38:21 -0400
Subject: [PATCH] Align f128 to 16 bytes when passing on x86

LLVM does not seem to enforce the required 16-byte alignment, which is
causing an ABI mismatch.

This is not yet working locally.
---
 llvm/lib/Target/X86/X86CallingConv.td | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/llvm/lib/Target/X86/X86CallingConv.td b/llvm/lib/Target/X86/X86CallingConv.td
index 0d087e057a2bd..52665050244e9 100644
--- a/llvm/lib/Target/X86/X86CallingConv.td
+++ b/llvm/lib/Target/X86/X86CallingConv.td
@@ -829,6 +829,9 @@ def CC_X86_32_Common : CallingConv<[
   // Long doubles get slots whose size and alignment depends on the subtarget.
   CCIfType<[f80], CCAssignToStack<0, 0>>,
 
+  // f128 get 16-byte slots that are 16-byte aligned.
+  CCIfType<[f128], CCAssignToStack<16, 16>>,
+
   // Boolean vectors of AVX-512 are passed in SIMD registers.
   // The call from AVX to AVX-512 function should work,
   // since the boolean types in AVX/AVX2 are promoted by default.