Reland "[X86][BF16] Add libcall for F80 -> BF16 (#109116)"

[phoebewang]

This reverts commit ababfee78714313a0cad87591b819f0944b90d09.

Add X86 FP80 check.

[llvmbot]

@llvm/pr-subscribers-llvm-ir

@llvm/pr-subscribers-backend-x86

Author: Phoebe Wang (phoebewang)

Changes

This reverts commit ababfee78714313a0cad87591b819f0944b90d09.

Add X86 FP80 check.

---

Full diff: https://github.com/llvm/llvm-project/pull/109143.diff

5 Files Affected:

• (modified) compiler-rt/lib/builtins/CMakeLists.txt (+1)
• (added) compiler-rt/lib/builtins/truncxfbf2.c (+16)
• (modified) llvm/include/llvm/IR/RuntimeLibcalls.def (+1)
• (modified) llvm/lib/CodeGen/TargetLoweringBase.cpp (+2)
• (modified) llvm/test/CodeGen/X86/bfloat.ll (+22)

diff --git a/compiler-rt/lib/builtins/CMakeLists.txt b/compiler-rt/lib/builtins/CMakeLists.txt
index 2c3b0fa84a4782..9a0a50ee7003f1 100644
--- a/compiler-rt/lib/builtins/CMakeLists.txt
+++ b/compiler-rt/lib/builtins/CMakeLists.txt
@@ -192,6 +192,7 @@ set(GENERIC_SOURCES
 set(BF16_SOURCES
     extendbfsf2.c
     truncdfbf2.c
+    truncxfbf2.c
     truncsfbf2.c
 )
 
diff --git a/compiler-rt/lib/builtins/truncxfbf2.c b/compiler-rt/lib/builtins/truncxfbf2.c
new file mode 100644
index 00000000000000..aa4108c334eebc
--- /dev/null
+++ b/compiler-rt/lib/builtins/truncxfbf2.c
@@ -0,0 +1,16 @@
+//===-- lib/truncxfbf2.c - long double -> bfloat conversion -------*- C -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#if defined(CRT_HAS_TF_MODE) && __LDBL_MANT_DIG__ == 64 && defined(__x86_64__)
+#define SRC_80
+#define DST_BFLOAT
+#include "fp_trunc_impl.inc"
+
+COMPILER_RT_ABI dst_t __truncxfbf2(long double a) { return __truncXfYf2__(a); }
+
+#endif
diff --git a/llvm/include/llvm/IR/RuntimeLibcalls.def b/llvm/include/llvm/IR/RuntimeLibcalls.def
index c3d5ef9f4e4f82..69cf43140ad4bd 100644
--- a/llvm/include/llvm/IR/RuntimeLibcalls.def
+++ b/llvm/include/llvm/IR/RuntimeLibcalls.def
@@ -367,6 +367,7 @@ HANDLE_LIBCALL(FPROUND_F128_F16, "__trunctfhf2")
 HANDLE_LIBCALL(FPROUND_PPCF128_F16, "__trunctfhf2")
 HANDLE_LIBCALL(FPROUND_F32_BF16, "__truncsfbf2")
 HANDLE_LIBCALL(FPROUND_F64_BF16, "__truncdfbf2")
+HANDLE_LIBCALL(FPROUND_F80_BF16, "__truncxfbf2")
 HANDLE_LIBCALL(FPROUND_F64_F32, "__truncdfsf2")
 HANDLE_LIBCALL(FPROUND_F80_F32, "__truncxfsf2")
 HANDLE_LIBCALL(FPROUND_F128_F32, "__trunctfsf2")
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
index eb3190c7cd247a..9fdde454559171 100644
--- a/llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -169,6 +169,8 @@ RTLIB::Libcall RTLIB::getFPROUND(EVT OpVT, EVT RetVT) {
       return FPROUND_F32_BF16;
     if (OpVT == MVT::f64)
       return FPROUND_F64_BF16;
+    if (OpVT == MVT::f80)
+      return FPROUND_F80_BF16;
   } else if (RetVT == MVT::f32) {
     if (OpVT == MVT::f64)
       return FPROUND_F64_F32;
diff --git a/llvm/test/CodeGen/X86/bfloat.ll b/llvm/test/CodeGen/X86/bfloat.ll
index 3759909a2ccc8e..3144fd56d9ccf3 100644
--- a/llvm/test/CodeGen/X86/bfloat.ll
+++ b/llvm/test/CodeGen/X86/bfloat.ll
@@ -1953,3 +1953,25 @@ define void @PR92471(ptr %0, ptr %1) nounwind {
   store <7 x float> %4, ptr %1, align 4
   ret void
 }
+
+define bfloat @PR108936(x86_fp80 %0) nounwind {
+; X86-LABEL: PR108936:
+; X86:       # %bb.0:
+; X86-NEXT:    subl $12, %esp
+; X86-NEXT:    fldt {{[0-9]+}}(%esp)
+; X86-NEXT:    fstpt (%esp)
+; X86-NEXT:    calll __truncxfbf2
+; X86-NEXT:    addl $12, %esp
+; X86-NEXT:    retl
+;
+; CHECK-LABEL: PR108936:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    subq $24, %rsp
+; CHECK-NEXT:    fldt {{[0-9]+}}(%rsp)
+; CHECK-NEXT:    fstpt (%rsp)
+; CHECK-NEXT:    callq __truncxfbf2@PLT
+; CHECK-NEXT:    addq $24, %rsp
+; CHECK-NEXT:    retq
+  %2 = fptrunc x86_fp80 %0 to bfloat
+  ret bfloat %2
+}

[phoebewang]

I assume this would solve the buildbot failures on other targets, but I'm surprised it was also failed on llvm-clang-x86_64-gcc-ubuntu. I'll check it locally tomorrow.

[kazutakahirata]

I've fixed a warning from this PR with b55d4a3.

[llvm-ci]

LLVM Buildbot has detected a new failure on builder lld-x86_64-win running on as-worker-93 while building compiler-rt,llvm at step 7 "test-build-unified-tree-check-all".

Full details are available at: https://lab.llvm.org/buildbot/#/builders/146/builds/1194

Here is the relevant piece of the build log for the reference

Step 7 (test-build-unified-tree-check-all) failure: test (failure)
******************** TEST 'LLVM-Unit :: Support/./SupportTests.exe/36/87' FAILED ********************
Script(shard):
--
GTEST_OUTPUT=json:C:\a\lld-x86_64-win\build\unittests\Support\.\SupportTests.exe-LLVM-Unit-6336-36-87.json GTEST_SHUFFLE=0 GTEST_TOTAL_SHARDS=87 GTEST_SHARD_INDEX=36 C:\a\lld-x86_64-win\build\unittests\Support\.\SupportTests.exe
--

Script:
--
C:\a\lld-x86_64-win\build\unittests\Support\.\SupportTests.exe --gtest_filter=ProgramEnvTest.CreateProcessLongPath
--
C:\a\lld-x86_64-win\llvm-project\llvm\unittests\Support\ProgramTest.cpp(160): error: Expected equality of these values:
  0
  RC
    Which is: -2

C:\a\lld-x86_64-win\llvm-project\llvm\unittests\Support\ProgramTest.cpp(163): error: fs::remove(Twine(LongPath)): did not return errc::success.
error number: 13
error message: permission denied



C:\a\lld-x86_64-win\llvm-project\llvm\unittests\Support\ProgramTest.cpp:160
Expected equality of these values:
  0
  RC
    Which is: -2

C:\a\lld-x86_64-win\llvm-project\llvm\unittests\Support\ProgramTest.cpp:163
fs::remove(Twine(LongPath)): did not return errc::success.
error number: 13
error message: permission denied




********************