// Alias instructions that map zero vector to pxor / xorp* for sse.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-zeros value if folding it would be beneficial.
-// FIXME: Change encoding to pseudo!
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
isCodeGenOnly = 1 in {
-def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4f32 immAllZerosV))]>;
-def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v2f64 immAllZerosV))]>;
-let ExeDomain = SSEPackedInt in
-def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4i32 immAllZerosV))]>;
+ let Predicates = [HasSSE1] in
+ def V_SET0PS : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4f32 immAllZerosV))]>;
+ let Predicates = [HasSSE2] in {
+ def V_SET0PD : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v2f64 immAllZerosV))]>;
+ def V_SET0PI : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4i32 immAllZerosV))], SSEPackedInt>;
+ }
}
// The same as done above but for AVX. The 128-bit versions are the
// same, but re-encoded. The 256-bit does not support PI version.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
isCodeGenOnly = 1, Predicates = [HasAVX] in {
-def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
-def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
-def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
- [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
-def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
- [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
-let ExeDomain = SSEPackedInt in
-def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4i32 immAllZerosV))]>;
+def AVX_SET0PS : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4f32 immAllZerosV))]>;
+def AVX_SET0PD : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v2f64 immAllZerosV))]>;
+def AVX_SET0PI : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4i32 immAllZerosV))], SSEPackedInt>;
+def AVX_SET0PSY : I<0, Pseudo, (outs VR256:$dst), (ins), "",
+ [(set VR256:$dst, (v8f32 immAllZerosV))]>;
+def AVX_SET0PDY : I<0, Pseudo, (outs VR256:$dst), (ins), "",
+ [(set VR256:$dst, (v4f64 immAllZerosV))]>;
}
def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
// was introduced with SSE2, it's backward compatible.
def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
-// Alias instructions that map zero vector to pxor / xorp* for sse.
+// Alias instructions that map one vector to pcmpeqd for sse2 and above.
// We set canFoldAsLoad because this can be converted to a constant-pool
// load of an all-ones value if folding it would be beneficial.
let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
- isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
- // FIXME: Change encoding to pseudo.
- def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
- [(set VR128:$dst, (v4i32 immAllOnesV))]>;
+ isCodeGenOnly = 1, Predicates = [HasSSE2] in
+ def V_SETALLONES : I<0, Pseudo, (outs VR128:$dst), (ins), "",
+ [(set VR128:$dst, (v4i32 immAllOnesV))], SSEPackedInt>;
//===---------------------------------------------------------------------===//
// SSE3 - Conversion Instructions