return X86::isPSHUFDMask(N);
}], MMX_SHUFFLE_get_shuf_imm>;
-// Patterns for: vector_shuffle v1, v2, <4, 5, 2, 3>; etc.
-def MMX_MOVL_shuffle_mask : PatLeaf<(build_vector), [{
- return X86::isMOVLMask(N);
-}]>;
-
//===----------------------------------------------------------------------===//
// MMX Multiclasses
//===----------------------------------------------------------------------===//
def MMX_MOVD64rr : MMXI<0x6E, MRMSrcReg, (outs VR64:$dst), (ins GR32:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (v2i32 (scalar_to_vector GR32:$src)))]>;
-let isSimpleLoad = 1, isReMaterializable = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1 in
def MMX_MOVD64rm : MMXI<0x6E, MRMSrcMem, (outs VR64:$dst), (ins i32mem:$src),
"movd\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (v2i32 (scalar_to_vector (loadi32 addr:$src))))]>;
let neverHasSideEffects = 1 in
def MMX_MOVQ64rr : MMXI<0x6F, MRMSrcReg, (outs VR64:$dst), (ins VR64:$src),
"movq\t{$src, $dst|$dst, $src}", []>;
-let isSimpleLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
+let canFoldAsLoad = 1, isReMaterializable = 1, mayHaveSideEffects = 1 in
def MMX_MOVQ64rm : MMXI<0x6F, MRMSrcMem, (outs VR64:$dst), (ins i64mem:$src),
"movq\t{$src, $dst|$dst, $src}",
[(set VR64:$dst, (load_mmx addr:$src))]>;
"movq\t{$src, $dst|$dst, $src}",
[(store (v1i64 VR64:$src), addr:$dst)]>;
-def MMX_MOVDQ2Qrr : MMXID<0xD6, MRMDestMem, (outs VR64:$dst), (ins VR128:$src),
+def MMX_MOVDQ2Qrr : SDIi8<0xD6, MRMDestMem, (outs VR64:$dst), (ins VR128:$src),
"movdq2q\t{$src, $dst|$dst, $src}",
[(set VR64:$dst,
(v1i64 (bitconvert
(i64 (vector_extract (v2i64 VR128:$src),
(iPTR 0))))))]>;
-def MMX_MOVQ2DQrr : MMXIS<0xD6, MRMDestMem, (outs VR128:$dst), (ins VR64:$src),
- "movq2dq\t{$src, $dst|$dst, $src}",
+def MMX_MOVQ2DQrr : SSDIi8<0xD6, MRMDestMem, (outs VR128:$dst), (ins VR64:$src),
+ "movq2dq\t{$src, $dst|$dst, $src}",
[(set VR128:$dst,
(v2i64 (vector_shuffle immAllZerosV,
(v2i64 (scalar_to_vector (i64 (bitconvert VR64:$src)))),
def MMX_MASKMOVQ : MMXI<0xF7, MRMDestMem, (outs), (ins VR64:$src, VR64:$mask),
"maskmovq\t{$mask, $src|$src, $mask}",
[(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, EDI)]>;
+let Uses = [RDI] in
+def MMX_MASKMOVQ64: MMXI64<0xF7, MRMDestMem, (outs), (ins VR64:$src, VR64:$mask),
+ "maskmovq\t{$mask, $src|$src, $mask}",
+ [(int_x86_mmx_maskmovq VR64:$src, VR64:$mask, RDI)]>;
//===----------------------------------------------------------------------===//
// Alias Instructions
(MMX_MOVQ64mr addr:$dst, VR64:$src)>;
def : Pat<(store (v2i32 VR64:$src), addr:$dst),
(MMX_MOVQ64mr addr:$dst, VR64:$src)>;
+def : Pat<(store (v2f32 VR64:$src), addr:$dst),
+ (MMX_MOVQ64mr addr:$dst, VR64:$src)>;
def : Pat<(store (v1i64 VR64:$src), addr:$dst),
(MMX_MOVQ64mr addr:$dst, VR64:$src)>;
// Bit convert.
def : Pat<(v8i8 (bitconvert (v1i64 VR64:$src))), (v8i8 VR64:$src)>;
def : Pat<(v8i8 (bitconvert (v2i32 VR64:$src))), (v8i8 VR64:$src)>;
+def : Pat<(v8i8 (bitconvert (v2f32 VR64:$src))), (v8i8 VR64:$src)>;
def : Pat<(v8i8 (bitconvert (v4i16 VR64:$src))), (v8i8 VR64:$src)>;
def : Pat<(v4i16 (bitconvert (v1i64 VR64:$src))), (v4i16 VR64:$src)>;
def : Pat<(v4i16 (bitconvert (v2i32 VR64:$src))), (v4i16 VR64:$src)>;
+def : Pat<(v4i16 (bitconvert (v2f32 VR64:$src))), (v4i16 VR64:$src)>;
def : Pat<(v4i16 (bitconvert (v8i8 VR64:$src))), (v4i16 VR64:$src)>;
def : Pat<(v2i32 (bitconvert (v1i64 VR64:$src))), (v2i32 VR64:$src)>;
+def : Pat<(v2i32 (bitconvert (v2f32 VR64:$src))), (v2i32 VR64:$src)>;
def : Pat<(v2i32 (bitconvert (v4i16 VR64:$src))), (v2i32 VR64:$src)>;
def : Pat<(v2i32 (bitconvert (v8i8 VR64:$src))), (v2i32 VR64:$src)>;
+def : Pat<(v2f32 (bitconvert (v1i64 VR64:$src))), (v2f32 VR64:$src)>;
+def : Pat<(v2f32 (bitconvert (v2i32 VR64:$src))), (v2f32 VR64:$src)>;
+def : Pat<(v2f32 (bitconvert (v4i16 VR64:$src))), (v2f32 VR64:$src)>;
+def : Pat<(v2f32 (bitconvert (v8i8 VR64:$src))), (v2f32 VR64:$src)>;
def : Pat<(v1i64 (bitconvert (v2i32 VR64:$src))), (v1i64 VR64:$src)>;
+def : Pat<(v1i64 (bitconvert (v2f32 VR64:$src))), (v1i64 VR64:$src)>;
def : Pat<(v1i64 (bitconvert (v4i16 VR64:$src))), (v1i64 VR64:$src)>;
def : Pat<(v1i64 (bitconvert (v8i8 VR64:$src))), (v1i64 VR64:$src)>;
(MMX_MOVD64to64rr GR64:$src)>;
def : Pat<(v2i32 (bitconvert (i64 GR64:$src))),
(MMX_MOVD64to64rr GR64:$src)>;
+def : Pat<(v2f32 (bitconvert (i64 GR64:$src))),
+ (MMX_MOVD64to64rr GR64:$src)>;
def : Pat<(v4i16 (bitconvert (i64 GR64:$src))),
(MMX_MOVD64to64rr GR64:$src)>;
def : Pat<(v8i8 (bitconvert (i64 GR64:$src))),
(MMX_MOVD64from64rr VR64:$src)>;
def : Pat<(i64 (bitconvert (v2i32 VR64:$src))),
(MMX_MOVD64from64rr VR64:$src)>;
+def : Pat<(i64 (bitconvert (v2f32 VR64:$src))),
+ (MMX_MOVD64from64rr VR64:$src)>;
def : Pat<(i64 (bitconvert (v4i16 VR64:$src))),
(MMX_MOVD64from64rr VR64:$src)>;
def : Pat<(i64 (bitconvert (v8i8 VR64:$src))),
(MMX_MOVZDI2PDIrr GR32:$src)>;
}
+let AddedComplexity = 20 in {
+ def : Pat<(v8i8 (X86vzmovl (bc_v8i8 (load_mmx addr:$src)))),
+ (MMX_MOVZDI2PDIrm addr:$src)>;
+ def : Pat<(v4i16 (X86vzmovl (bc_v4i16 (load_mmx addr:$src)))),
+ (MMX_MOVZDI2PDIrm addr:$src)>;
+ def : Pat<(v2i32 (X86vzmovl (bc_v2i32 (load_mmx addr:$src)))),
+ (MMX_MOVZDI2PDIrm addr:$src)>;
+}
+
+// Clear top half.
+let AddedComplexity = 15 in {
+ def : Pat<(v8i8 (X86vzmovl VR64:$src)),
+ (MMX_PUNPCKLDQrr VR64:$src, (MMX_V_SET0))>;
+ def : Pat<(v4i16 (X86vzmovl VR64:$src)),
+ (MMX_PUNPCKLDQrr VR64:$src, (MMX_V_SET0))>;
+ def : Pat<(v2i32 (X86vzmovl VR64:$src)),
+ (MMX_PUNPCKLDQrr VR64:$src, (MMX_V_SET0))>;
+}
+
// Scalar to v4i16 / v8i8. The source may be a GR32, but only the lower
// 8 or 16-bits matter.
def : Pat<(bc_v8i8 (v2i32 (scalar_to_vector GR32:$src))),