[llvm-branch-commits] [llvm] [AArch64] Use 64-bit registers for wide "r" inline asm operands (PR #239127)
Eli Friedman via llvm-branch-commits
llvm-branch-commits at lists.llvm.org
Tue Nov 17 15:47:22 PST 2026
================
@@ -14487,8 +14487,14 @@ AArch64TargetLowering::getRegForInlineAsmConstraint(
return std::make_pair(0U, nullptr);
if (Subtarget->hasLS64() && VT.getSizeInBits() == 512)
return std::make_pair(0U, &AArch64::GPR64x8ClassRegClass);
- if (VT.getFixedSizeInBits() == 64)
+
+ // An integer wider than 64 bits, such as an i128, takes a 64-bit
+ // register for each 64 bits of it. 32-bit registers would hold only half
+ // of it.
----------------
efriedma-quic wrote:
Is there any way to actually refer to the high half? "Supporting" this seems kind of useless without that.
https://github.com/llvm/llvm-project/pull/239127
More information about the llvm-branch-commits
mailing list