Fix a long standing deficiency in the X86 backend: we would
sometimes emit "zero" and "all one" vectors multiple times,
for example:
_test2:
pcmpeqd %mm0, %mm0
movq %mm0, _M1
pcmpeqd %mm0, %mm0
movq %mm0, _M2
ret
instead of:
_test2:
pcmpeqd %mm0, %mm0
movq %mm0, _M1
movq %mm0, _M2
ret
This patch fixes this by always arranging for zero/one vectors
to be defined as v4i32 or v2i32 (SSE/MMX) instead of letting them be
any random type. This ensures they get trivially CSE'd on the dag.
This fix is also important for LegalizeDAGTypes, as it gets unhappy
when the x86 backend wants BUILD_VECTOR(i64 0) to be legal even when
'i64' isn't legal.
This patch makes the following changes:
1) X86TargetLowering::LowerBUILD_VECTOR now lowers 0/1 vectors into
their canonical types.
2) The now-dead patterns are removed from the SSE/MMX .td files.
3) All the patterns in the .td file that referred to immAllOnesV or
immAllZerosV in the wrong form now use *_bc to match them with a
bitcast wrapped around them.
4) X86DAGToDAGISel::SelectScalarSSELoad is generalized to handle
bitcast'd zero vectors, which simplifies the code actually.
5) getShuffleVectorZeroOrUndef is updated to generate a shuffle that
is legal, instead of generating one that is illegal and expecting
a later legalize pass to clean it up.
6) isZeroShuffle is generalized to handle bitcast of zeros.
7) several other minor tweaks.
This patch is definite goodness, but has the potential to cause random
code quality regressions. Please be on the lookout for these and let
me know if they happen.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@44310 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/X86/X86InstrMMX.td b/lib/Target/X86/X86InstrMMX.td
index b7024bc..c892c34 100644
--- a/lib/Target/X86/X86InstrMMX.td
+++ b/lib/Target/X86/X86InstrMMX.td
@@ -486,14 +486,13 @@
//===----------------------------------------------------------------------===//
// Alias instructions that map zero vector to pxor.
-// FIXME: remove when we can teach regalloc that xor reg, reg is ok.
let isReMaterializable = 1 in {
def MMX_V_SET0 : MMXI<0xEF, MRMInitReg, (outs VR64:$dst), (ins),
"pxor\t$dst, $dst",
- [(set VR64:$dst, (v1i64 immAllZerosV))]>;
+ [(set VR64:$dst, (v2i32 immAllZerosV))]>;
def MMX_V_SETALLONES : MMXI<0x76, MRMInitReg, (outs VR64:$dst), (ins),
"pcmpeqd\t$dst, $dst",
- [(set VR64:$dst, (v1i64 immAllOnesV))]>;
+ [(set VR64:$dst, (v2i32 immAllOnesV))]>;
}
//===----------------------------------------------------------------------===//
@@ -510,18 +509,6 @@
def : Pat<(store (v1i64 VR64:$src), addr:$dst),
(MMX_MOVQ64mr addr:$dst, VR64:$src)>;
-// 64-bit vector all zero's.
-def : Pat<(v8i8 immAllZerosV), (MMX_V_SET0)>;
-def : Pat<(v4i16 immAllZerosV), (MMX_V_SET0)>;
-def : Pat<(v2i32 immAllZerosV), (MMX_V_SET0)>;
-def : Pat<(v1i64 immAllZerosV), (MMX_V_SET0)>;
-
-// 64-bit vector all one's.
-def : Pat<(v8i8 immAllOnesV), (MMX_V_SETALLONES)>;
-def : Pat<(v4i16 immAllOnesV), (MMX_V_SETALLONES)>;
-def : Pat<(v2i32 immAllOnesV), (MMX_V_SETALLONES)>;
-def : Pat<(v1i64 immAllOnesV), (MMX_V_SETALLONES)>;
-
// Bit convert.
def : Pat<(v8i8 (bitconvert (v1i64 VR64:$src))), (v8i8 VR64:$src)>;
def : Pat<(v8i8 (bitconvert (v2i32 VR64:$src))), (v8i8 VR64:$src)>;
@@ -551,10 +538,10 @@
// Move scalar to XMM zero-extended
// movd to XMM register zero-extends
let AddedComplexity = 15 in {
- def : Pat<(v8i8 (vector_shuffle immAllZerosV,
+ def : Pat<(v8i8 (vector_shuffle immAllZerosV_bc,
(v8i8 (MMX_X86s2vec GR32:$src)), MMX_MOVL_shuffle_mask)),
(MMX_MOVZDI2PDIrr GR32:$src)>;
- def : Pat<(v4i16 (vector_shuffle immAllZerosV,
+ def : Pat<(v4i16 (vector_shuffle immAllZerosV_bc,
(v4i16 (MMX_X86s2vec GR32:$src)), MMX_MOVL_shuffle_mask)),
(MMX_MOVZDI2PDIrr GR32:$src)>;
def : Pat<(v2i32 (vector_shuffle immAllZerosV,
@@ -606,19 +593,19 @@
def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v2i32 immAllOnesV))),
VR64:$src2)),
(MMX_PANDNrr VR64:$src1, VR64:$src2)>;
-def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v4i16 immAllOnesV))),
+def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v4i16 immAllOnesV_bc))),
VR64:$src2)),
(MMX_PANDNrr VR64:$src1, VR64:$src2)>;
-def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v8i8 immAllOnesV))),
+def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v8i8 immAllOnesV_bc))),
VR64:$src2)),
(MMX_PANDNrr VR64:$src1, VR64:$src2)>;
def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v2i32 immAllOnesV))),
(load addr:$src2))),
(MMX_PANDNrm VR64:$src1, addr:$src2)>;
-def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v4i16 immAllOnesV))),
+def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v4i16 immAllOnesV_bc))),
(load addr:$src2))),
(MMX_PANDNrm VR64:$src1, addr:$src2)>;
-def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v8i8 immAllOnesV))),
+def : Pat<(v1i64 (and (xor VR64:$src1, (bc_v1i64 (v8i8 immAllOnesV_bc))),
(load addr:$src2))),
(MMX_PANDNrm VR64:$src1, addr:$src2)>;