From 191c1b88cde9ff8c34ffd7027aad1d7ae1c54bcd Mon Sep 17 00:00:00 2001
From: Trevor Gross <tmgross@umich.edu>
Date: Fri, 10 May 2024 19:46:50 -0500
Subject: [PATCH] Implement `f128` multiplication

---
 library/compiler-builtins/README.md             |  2 +-
 library/compiler-builtins/build.rs              |  5 -----
 library/compiler-builtins/src/float/mul.rs      | 11 +++++++++++
 .../compiler-builtins/testcrate/tests/mul.rs    | 17 +++++++++++++++++
 4 files changed, 29 insertions(+), 6 deletions(-)

diff --git a/library/compiler-builtins/README.md b/library/compiler-builtins/README.md
index c7241ec7c0fd..1986c21c6692 100644
--- a/library/compiler-builtins/README.md
+++ b/library/compiler-builtins/README.md
@@ -249,7 +249,7 @@ These builtins are needed to support `f16` and `f128`, which are in the process
 - [ ] floatsitf.c
 - [ ] floatunditf.c
 - [ ] floatunsitf.c
-- [ ] multf3.c
+- [x] multf3.c
 - [ ] powitf2.c
 - [ ] ppc/fixtfdi.c
 - [ ] ppc/fixunstfdi.c
diff --git a/library/compiler-builtins/build.rs b/library/compiler-builtins/build.rs
index a57052429460..ec830ecb34b2 100644
--- a/library/compiler-builtins/build.rs
+++ b/library/compiler-builtins/build.rs
@@ -543,7 +543,6 @@ mod c {
                 ("__floatsitf", "floatsitf.c"),
                 ("__floatunditf", "floatunditf.c"),
                 ("__floatunsitf", "floatunsitf.c"),
-                ("__multf3", "multf3.c"),
                 ("__divtf3", "divtf3.c"),
                 ("__powitf2", "powitf2.c"),
                 ("__fe_getround", "fp_mode.c"),
@@ -562,26 +561,22 @@ mod c {
         if target_arch == "mips64" {
             sources.extend(&[
                 ("__netf2", "comparetf2.c"),
-                ("__multf3", "multf3.c"),
                 ("__fixtfsi", "fixtfsi.c"),
                 ("__floatsitf", "floatsitf.c"),
                 ("__fixunstfsi", "fixunstfsi.c"),
                 ("__floatunsitf", "floatunsitf.c"),
                 ("__fe_getround", "fp_mode.c"),
-                ("__divtf3", "divtf3.c"),
             ]);
         }
 
         if target_arch == "loongarch64" {
             sources.extend(&[
                 ("__netf2", "comparetf2.c"),
-                ("__multf3", "multf3.c"),
                 ("__fixtfsi", "fixtfsi.c"),
                 ("__floatsitf", "floatsitf.c"),
                 ("__fixunstfsi", "fixunstfsi.c"),
                 ("__floatunsitf", "floatunsitf.c"),
                 ("__fe_getround", "fp_mode.c"),
-                ("__divtf3", "divtf3.c"),
             ]);
         }
 
diff --git a/library/compiler-builtins/src/float/mul.rs b/library/compiler-builtins/src/float/mul.rs
index 46c41d09f191..9866b62804da 100644
--- a/library/compiler-builtins/src/float/mul.rs
+++ b/library/compiler-builtins/src/float/mul.rs
@@ -199,6 +199,17 @@ intrinsics! {
         mul(a, b)
     }
 
+    #[cfg(not(any(feature = "no-f16-f128", target_arch = "powerpc", target_arch = "powerpc64")))]
+    pub extern "C" fn __multf3(a: f128, b: f128) -> f128 {
+        mul(a, b)
+    }
+
+
+    #[cfg(all(not(feature = "no-f16-f128"), any(target_arch = "powerpc", target_arch = "powerpc64")))]
+    pub extern "C" fn __mulkf3(a: f128, b: f128) -> f128 {
+        mul(a, b)
+    }
+
     #[cfg(target_arch = "arm")]
     pub extern "C" fn __mulsf3vfp(a: f32, b: f32) -> f32 {
         a * b
diff --git a/library/compiler-builtins/testcrate/tests/mul.rs b/library/compiler-builtins/testcrate/tests/mul.rs
index 4fca2c13fa66..ffbe638644d1 100644
--- a/library/compiler-builtins/testcrate/tests/mul.rs
+++ b/library/compiler-builtins/testcrate/tests/mul.rs
@@ -1,4 +1,6 @@
 #![allow(unused_macros)]
+#![feature(f128)]
+#![feature(f16)]
 
 use testcrate::*;
 
@@ -114,6 +116,21 @@ fn float_mul() {
         f32, __mulsf3, Single, all();
         f64, __muldf3, Double, all();
     );
+
+    #[cfg(not(feature = "no-f16-f128"))]
+    {
+        #[cfg(any(target_arch = "powerpc", target_arch = "powerpc64"))]
+        use compiler_builtins::float::mul::__mulkf3 as __multf3;
+        #[cfg(not(any(target_arch = "powerpc", target_arch = "powerpc64")))]
+        use compiler_builtins::float::mul::__multf3;
+
+        float_mul!(
+            f128, __multf3, Quad,
+            // FIXME(llvm): there is a bug in LLVM rt.
+            // See <https://github.com/llvm/llvm-project/issues/91840>.
+            not(any(feature = "no-sys-f128", all(target_arch = "aarch64", target_os = "linux")));
+        );
+    }
 }
 
 #[cfg(target_arch = "arm")]