Skip to content

Commit 6b9cd72

Browse files
committed
[SelectionDAG][X86] Support inline assembly returning an mmx register into a type with fewer than 64 bits.
It's possible to use the 'y' mmx constraint with a type narrower than 64-bits. This patch supports this by bitcasting the mmx type to 64-bits and then truncating to the desired type. There are probably other missing type combinations we need to support, but this is the case we have a bug report for. Fixes PR41748. Differential Revision: https://reviews.llvm.org/D61582 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@360069 91177308-0d34-0410-b5e6-96231b3b80d8
1 parent 4ad9ccc commit 6b9cd72

File tree

2 files changed

+23
-0
lines changed

2 files changed

+23
-0
lines changed

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -322,6 +322,14 @@ static SDValue getCopyFromParts(SelectionDAG &DAG, const SDLoc &DL,
322322
return DAG.getNode(ISD::FP_EXTEND, DL, ValueVT, Val);
323323
}
324324

325+
// Handle MMX to a narrower integer type by bitcasting MMX to integer and
326+
// then truncating.
327+
if (PartEVT == MVT::x86mmx && ValueVT.isInteger() &&
328+
ValueVT.bitsLT(PartEVT)) {
329+
Val = DAG.getNode(ISD::BITCAST, DL, MVT::i64, Val);
330+
return DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
331+
}
332+
325333
report_fatal_error("Unknown mismatch in getCopyFromParts!");
326334
}
327335

test/CodeGen/X86/pr41748.ll

Lines changed: 15 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,15 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2+
; RUN: llc < %s -mtriple=x86_64-apple-macosx10.14.0 -mattr=mmx | FileCheck %s
3+
4+
define i32 @foo(i32 %a) {
5+
; CHECK-LABEL: foo:
6+
; CHECK: ## %bb.0: ## %entry
7+
; CHECK-NEXT: ## InlineAsm Start
8+
; CHECK-NEXT: movd %edi, %mm0
9+
; CHECK-NEXT: ## InlineAsm End
10+
; CHECK-NEXT: movd %mm0, %eax
11+
; CHECK-NEXT: retq
12+
entry:
13+
%0 = tail call i32 asm sideeffect "movd $1, $0", "=y,r,~{dirflag},~{fpsr},~{flags}"(i32 %a)
14+
ret i32 %0
15+
}

0 commit comments

Comments
 (0)