This patch adds support for 16 bit floating point registers to the inline asm registe...

Without this patch, register allocation for the example below fails.

define half @test(half %a1, half %a2) #0 {
entry:
  %0 = tail call half asm "sqrshl ${0:h}, ${1:h}, ${2:h}", "=w,w,w" (half %a1, half %a2) #1
  ret half %0
}

Patch by Florian Hahn.

Differential Revision: https://reviews.llvm.org/D25080

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@286111 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AArch64/AArch64ISelLowering.cpp b/lib/Target/AArch64/AArch64ISelLowering.cpp
index 076145c36b2d877cea64baf4ff5b145424ba3e54..04219baf041f66a0b2b277d82316d71a9abfa67a 100644 (file)
--- a/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ b/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -4780,6 +4780,8 @@ AArch64TargetLowering::getRegForInlineAsmConstraint(
         return std::make_pair(0U, &AArch64::GPR64commonRegClass);
       return std::make_pair(0U, &AArch64::GPR32commonRegClass);
     case 'w':
+      if (VT.getSizeInBits() == 16)
+        return std::make_pair(0U, &AArch64::FPR16RegClass);
       if (VT.getSizeInBits() == 32)
         return std::make_pair(0U, &AArch64::FPR32RegClass);
       if (VT.getSizeInBits() == 64)

diff --git a/test/CodeGen/AArch64/neon-inline-asm-16-bit-fp.ll b/test/CodeGen/AArch64/neon-inline-asm-16-bit-fp.ll
new file mode 100644 (file)
index 0000000..3656a78
--- /dev/null
+++ b/test/CodeGen/AArch64/neon-inline-asm-16-bit-fp.ll
@@ -0,0 +1,20 @@
+; RUN: llc -mtriple=aarch64-none-linux-gnu -mattr=+neon < %s | FileCheck %s
+
+; generated from
+; __fp16 test(__fp16 a1, __fp16 a2) {
+;    __fp16 res0;
+;    __asm__("sqrshl %h[__res], %h[__A], %h[__B]"
+;             : [__res] "=w" (res0)
+;             : [__A] "w" (a1), [__B] "w" (a2)
+;             :
+;             );
+;    return res0;
+;}
+
+; Function Attrs: nounwind readnone
+define half @test(half %a1, half %a2) #0 {
+entry:
+  ;CHECK: sqrshl {{h[0-9]+}}, {{h[0-9]+}}, {{h[0-9]+}}
+  %0 = tail call half asm "sqrshl ${0:h}, ${1:h}, ${2:h}", "=w,w,w" (half %a1, half %a2) #1
+  ret half %0
+}