RuntimeLibcalls: Remove darwin override of half convert libcalls

[arsenm]

These are already the default calls set for these conversions, so
they should not require explicit setting. The non-default cases are
currently overridden in ARMISelLowering. Just delete this until
the list of calls and lowering decisions are separated.

This was added back in 6402ad2. It
appears to not be relevant for AArch64, where calls appear to never
be used for these. It also appears to not be relevant for x86, where
the default calls seem to always end up used anyway.

[arsenm]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• RuntimeLibcalls: Remove target check for sjlj config #148792
• RuntimeLibcalls: Move __stack_chk_fail config to tablegen #148789
• RuntimeLibcalls: Remove darwin override of half convert libcalls #148782 👈 (View in Graphite)
• RuntimeLibcalls: Really move default libcall handling to tablegen #148780
• RuntimeLibcalls: Stop opting out of exp10 #148604
• RuntimeLibcalls: Mostly move x86 configuration into tablegen #148575
• RuntimeLibcalls: Invert handling of 64-bit only libcalls #148571
• RuntimeLibcalls: Remove __muloti4 from default libcall set #148562
• main

This stack of pull requests is managed by Graphite. Learn more about stacking.

[llvmbot]

@llvm/pr-subscribers-backend-arm

@llvm/pr-subscribers-llvm-ir

Author: Matt Arsenault (arsenm)

Changes

These are already the default calls set for these conversions, so
they should not require explicit setting. The non-default cases are
currently overridden in ARMISelLowering. Just delete this until
the list of calls and lowering decisions are separated.

This was added back in 6402ad2. It
appears to not be relevant for AArch64, where calls appear to never
be used for these. It also appears to not be relevant for x86, where
the default calls seem to always end up used anyway.

---

Full diff: https://github.com/llvm/llvm-project/pull/148782.diff

1 Files Affected:

• (modified) llvm/lib/IR/RuntimeLibcalls.cpp (-9)

diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index df529bc013cac..8ca31dff900cb 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -77,15 +77,6 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT,
       setLibcallImpl(RTLIB::UNWIND_RESUME, RTLIB::_Unwind_SjLj_Resume);
   }
 
-  // A few names are different on particular architectures or environments.
-  if (TT.isOSDarwin()) {
-    // For f16/f32 conversions, Darwin uses the standard naming scheme,
-    // instead of the gnueabi-style __gnu_*_ieee.
-    // FIXME: What about other targets?
-    setLibcallImpl(RTLIB::FPEXT_F16_F32, RTLIB::__extendhfsf2);
-    setLibcallImpl(RTLIB::FPROUND_F32_F16, RTLIB::__truncsfhf2);
-  }
-
   if (TT.isOSOpenBSD()) {
     setLibcallImpl(RTLIB::STACKPROTECTOR_CHECK_FAIL, RTLIB::Unsupported);
   }

[llvmbot]

@llvm/pr-subscribers-tablegen

Author: Matt Arsenault (arsenm)

Changes

These are already the default calls set for these conversions, so
they should not require explicit setting. The non-default cases are
currently overridden in ARMISelLowering. Just delete this until
the list of calls and lowering decisions are separated.

This was added back in 6402ad2. It
appears to not be relevant for AArch64, where calls appear to never
be used for these. It also appears to not be relevant for x86, where
the default calls seem to always end up used anyway.

---

Full diff: https://github.com/llvm/llvm-project/pull/148782.diff

1 Files Affected:

• (modified) llvm/lib/IR/RuntimeLibcalls.cpp (-9)

diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index df529bc013cac..8ca31dff900cb 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -77,15 +77,6 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT,
       setLibcallImpl(RTLIB::UNWIND_RESUME, RTLIB::_Unwind_SjLj_Resume);
   }
 
-  // A few names are different on particular architectures or environments.
-  if (TT.isOSDarwin()) {
-    // For f16/f32 conversions, Darwin uses the standard naming scheme,
-    // instead of the gnueabi-style __gnu_*_ieee.
-    // FIXME: What about other targets?
-    setLibcallImpl(RTLIB::FPEXT_F16_F32, RTLIB::__extendhfsf2);
-    setLibcallImpl(RTLIB::FPROUND_F32_F16, RTLIB::__truncsfhf2);
-  }
-
   if (TT.isOSOpenBSD()) {
     setLibcallImpl(RTLIB::STACKPROTECTOR_CHECK_FAIL, RTLIB::Unsupported);
   }

[llvmbot]

@llvm/pr-subscribers-backend-x86

Author: Matt Arsenault (arsenm)

Changes

These are already the default calls set for these conversions, so
they should not require explicit setting. The non-default cases are
currently overridden in ARMISelLowering. Just delete this until
the list of calls and lowering decisions are separated.

This was added back in 6402ad2. It
appears to not be relevant for AArch64, where calls appear to never
be used for these. It also appears to not be relevant for x86, where
the default calls seem to always end up used anyway.

---

Full diff: https://github.com/llvm/llvm-project/pull/148782.diff

1 Files Affected:

• (modified) llvm/lib/IR/RuntimeLibcalls.cpp (-9)

diff --git a/llvm/lib/IR/RuntimeLibcalls.cpp b/llvm/lib/IR/RuntimeLibcalls.cpp
index df529bc013cac..8ca31dff900cb 100644
--- a/llvm/lib/IR/RuntimeLibcalls.cpp
+++ b/llvm/lib/IR/RuntimeLibcalls.cpp
@@ -77,15 +77,6 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT,
       setLibcallImpl(RTLIB::UNWIND_RESUME, RTLIB::_Unwind_SjLj_Resume);
   }
 
-  // A few names are different on particular architectures or environments.
-  if (TT.isOSDarwin()) {
-    // For f16/f32 conversions, Darwin uses the standard naming scheme,
-    // instead of the gnueabi-style __gnu_*_ieee.
-    // FIXME: What about other targets?
-    setLibcallImpl(RTLIB::FPEXT_F16_F32, RTLIB::__extendhfsf2);
-    setLibcallImpl(RTLIB::FPROUND_F32_F16, RTLIB::__truncsfhf2);
-  }
-
   if (TT.isOSOpenBSD()) {
     setLibcallImpl(RTLIB::STACKPROTECTOR_CHECK_FAIL, RTLIB::Unsupported);
   }

[tgross35]

LGTM, this matches what I've seen while inspecting the tablegen output