6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2025-01-18 13:34:04 +00:00
Code Issues Projects Releases Wiki Activity

Fix PR10845. SUBREG_TO_REG shouldn't be used when the input and

Browse Source 
destination types are equal!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139553 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Bruno Cardoso Lopes 2011-09-12 22:59:23 +00:00
parent 27cea8e34b
commit 5fc48100ee
2 changed files with 22 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
lib/Target/X86/X86InstrSSE.td
View File

@ -4250,11 +4250,11 @@ let Predicates = [HasAVX] in {
// AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
let AddedComplexity = 20 in {
def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
(SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrm addr:$src), sub_xmm)>;
(VMOVZDI2PDIrm addr:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
(SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrm addr:$src), sub_xmm)>;
(VMOVZDI2PDIrm addr:$src)>;
def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
(SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrm addr:$src), sub_xmm)>;
(VMOVZDI2PDIrm addr:$src)>;
}
// Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
@ -4347,11 +4347,11 @@ let Predicates = [HasSSE2], AddedComplexity = 20 in {
let Predicates = [HasAVX], AddedComplexity = 20 in {
def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
(SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrm addr:$src), sub_xmm)>;
(VMOVZQI2PQIrm addr:$src)>;
def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
(SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrm addr:$src), sub_xmm)>;
(VMOVZQI2PQIrm addr:$src)>;
def : Pat<(v2i64 (X86vzload addr:$src)),
(SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrm addr:$src), sub_xmm)>;
(VMOVZQI2PQIrm addr:$src)>;
}
//===---------------------------------------------------------------------===//
@ -4392,9 +4392,9 @@ let AddedComplexity = 20 in {
}
let Predicates = [HasAVX] in {
def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
(SUBREG_TO_REG (i64 0), (VMOVZPQILo2PQIrm addr:$src), sub_xmm)>;
(VMOVZPQILo2PQIrm addr:$src)>;
def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
(SUBREG_TO_REG (i64 0), (MOVZPQILo2PQIrr VR128:$src), sub_xmm)>;
(VMOVZPQILo2PQIrr VR128:$src)>;
}
}

14
test/CodeGen/X86/avx-basic.ll
View File

@ -91,3 +91,17 @@ entry:
ret <8 x i32> %shuffle
}
;;; Don't crash on movd
; CHECK: _VMOVZQI2PQI
; CHECK: vmovd (%
define <8 x i32> @VMOVZQI2PQI([0 x float]* nocapture %aFOO) nounwind {
allocas:
%ptrcast.i33.i = bitcast [0 x float]* %aFOO to i32*
%val.i34.i = load i32* %ptrcast.i33.i, align 4
%ptroffset.i22.i992 = getelementptr [0 x float]* %aFOO, i64 0, i64 1
%ptrcast.i23.i = bitcast float* %ptroffset.i22.i992 to i32*
%val.i24.i = load i32* %ptrcast.i23.i, align 4
%updatedret.i30.i = insertelement <8 x i32> undef, i32 %val.i34.i, i32 1
ret <8 x i32> %updatedret.i30.i
}