[PATCH] D43179: [Thumb] Handle addressing mode AddrMode5FP16
Oliver Stannard via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Feb 12 09:53:17 PST 2018
olista01 added inline comments.
================
Comment at: lib/Target/ARM/Thumb2InstrInfo.cpp:611
+ Scale = 2;
+ Offset += InstrOffs * 4;
+ assert((Offset & (Scale-1)) == 0 && "Can't encode this offset!");
----------------
Why is this multiplied by 4, and not 2?
================
Comment at: test/CodeGen/ARM/fp16-instructions.ll:708
+ store volatile half 0xH3C00, half* %S, align 2
+ %S.0.S.0. = load volatile half, half* %S, align 2
+ %add = fadd half %S.0.S.0., %0
----------------
There is a half-precision load in the IR, does this generate a vldr.16? if so, we should be checking that.
================
Comment at: test/CodeGen/ARM/fp16-instructions.ll:715
+; CHECK-LABEL: ThumbAddrMode5FP16
+; CHECK-SOFTFP-FULLFP16: vstr.16 s0, [sp, #2]
+}
----------------
This test will be fragile WRT changes in stack layout, so we should use a regex for the offset.
https://reviews.llvm.org/D43179
More information about the llvm-commits
mailing list