[llvm] r220533 - [SelectionDAG] Teach the vector scalarizer about FP conversions.

Ahmed Bougacha ahmed.bougacha at gmail.com
Thu Oct 23 15:49:25 PDT 2014


Author: ab
Date: Thu Oct 23 17:49:25 2014
New Revision: 220533

URL: http://llvm.org/viewvc/llvm-project?rev=220533&view=rev
Log:
[SelectionDAG] Teach the vector scalarizer about FP conversions.

This adds support for legalization of instructions of the form:

  [fp_conv] <1 x i1> %op to <1 x double>

where fp_conv is one of fpto[us]i, [us]itofp.  This used to assert
because they were simply missing from the vector operand scalarizer.

A similar problem arose in r190830, with trunc instead.

Fixes PR20778.

Differential Revision: http://reviews.llvm.org/D5810

Added:
    llvm/trunk/test/CodeGen/AArch64/fpconv-vector-op-scalarize.ll
Modified:
    llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp

Modified: llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp?rev=220533&r1=220532&r2=220533&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp (original)
+++ llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp Thu Oct 23 17:49:25 2014
@@ -411,6 +411,10 @@ bool DAGTypeLegalizer::ScalarizeVectorOp
     case ISD::ZERO_EXTEND:
     case ISD::SIGN_EXTEND:
     case ISD::TRUNCATE:
+    case ISD::FP_TO_SINT:
+    case ISD::FP_TO_UINT:
+    case ISD::SINT_TO_FP:
+    case ISD::UINT_TO_FP:
       Res = ScalarizeVecOp_UnaryOp(N);
       break;
     case ISD::CONCAT_VECTORS:

Added: llvm/trunk/test/CodeGen/AArch64/fpconv-vector-op-scalarize.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/AArch64/fpconv-vector-op-scalarize.ll?rev=220533&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/AArch64/fpconv-vector-op-scalarize.ll (added)
+++ llvm/trunk/test/CodeGen/AArch64/fpconv-vector-op-scalarize.ll Thu Oct 23 17:49:25 2014
@@ -0,0 +1,44 @@
+; RUN: llc < %s -mtriple=arm64-apple-darwin | FileCheck %s
+
+; PR20778
+; Check that the legalizer doesn't crash when scalarizing FP conversion
+; instructions' operands.  The operands are all illegal on AArch64,
+; ensuring they are legalized.  The results are all legal.
+
+define <1 x double> @test_sitofp(<1 x i1> %in) {
+; CHECK-LABEL: test_sitofp:
+; CHECK:       sbfx  w8, w0, #0, #1
+; CHECK-NEXT:  scvtf d0, w8
+; CHECK-NEXT:  ret
+entry:
+  %0 = sitofp <1 x i1> %in to <1 x double>
+  ret <1 x double> %0
+}
+
+define <1 x double> @test_uitofp(<1 x i1> %in) {
+; CHECK-LABEL: test_uitofp:
+; CHECK:       and   w8, w0, #0x1
+; CHECK-NEXT:  ucvtf d0, w8
+; CHECK-NEXT:  ret
+entry:
+  %0 = uitofp <1 x i1> %in to <1 x double>
+  ret <1 x double> %0
+}
+
+define <1 x i64> @test_fptosi(<1 x fp128> %in) {
+; CHECK-LABEL: test_fptosi:
+; CHECK:       bl    ___fixtfdi
+; CHECK-NEXT:  fmov  d0, x0
+entry:
+  %0 = fptosi <1 x fp128> %in to <1 x i64>
+  ret <1 x i64> %0
+}
+
+define <1 x i64> @test_fptoui(<1 x fp128> %in) {
+; CHECK-LABEL: test_fptoui:
+; CHECK:       bl    ___fixunstfdi
+; CHECK-NEXT:  fmov  d0, x0
+entry:
+  %0 = fptoui <1 x fp128> %in to <1 x i64>
+  ret <1 x i64> %0
+}





More information about the llvm-commits mailing list