// Altivec transformation functions and pattern fragments.
//
-// VSPLT_get_imm xform function: convert vector_shuffle mask to VSPLT* imm.
-def VSPLT_get_imm : SDNodeXForm<build_vector, [{
- return getI32Imm(PPC::getVSPLTImmediate(N));
+/// VPKUHUM_shuffle_mask/VPKUWUM_shuffle_mask - Return true if this is a valid
+/// shuffle mask for the VPKUHUM or VPKUWUM instructions.
+def VPKUHUM_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVPKUHUMShuffleMask(N, false);
+}]>;
+def VPKUWUM_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVPKUWUMShuffleMask(N, false);
+}]>;
+
+def VPKUHUM_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVPKUHUMShuffleMask(N, true);
+}]>;
+def VPKUWUM_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVPKUWUMShuffleMask(N, true);
+}]>;
+
+
+def VMRGLB_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGLShuffleMask(N, 1, false);
+}]>;
+def VMRGLH_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGLShuffleMask(N, 2, false);
+}]>;
+def VMRGLW_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGLShuffleMask(N, 4, false);
+}]>;
+def VMRGHB_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGHShuffleMask(N, 1, false);
+}]>;
+def VMRGHH_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGHShuffleMask(N, 2, false);
+}]>;
+def VMRGHW_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGHShuffleMask(N, 4, false);
}]>;
-def VSPLT_shuffle_mask : PatLeaf<(build_vector), [{
- return PPC::isSplatShuffleMask(N);
-}], VSPLT_get_imm>;
+def VMRGLB_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGLShuffleMask(N, 1, true);
+}]>;
+def VMRGLH_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGLShuffleMask(N, 2, true);
+}]>;
+def VMRGLW_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGLShuffleMask(N, 4, true);
+}]>;
+def VMRGHB_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGHShuffleMask(N, 1, true);
+}]>;
+def VMRGHH_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGHShuffleMask(N, 2, true);
+}]>;
+def VMRGHW_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVMRGHShuffleMask(N, 4, true);
+}]>;
+
+
+def VSLDOI_get_imm : SDNodeXForm<build_vector, [{
+ return getI32Imm(PPC::isVSLDOIShuffleMask(N, false));
+}]>;
+def VSLDOI_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVSLDOIShuffleMask(N, false) != -1;
+}], VSLDOI_get_imm>;
+
+/// VSLDOI_unary* - These are used to match vsldoi(X,X), which is turned into
+/// vector_shuffle(X,undef,mask) by the dag combiner.
+def VSLDOI_unary_get_imm : SDNodeXForm<build_vector, [{
+ return getI32Imm(PPC::isVSLDOIShuffleMask(N, true));
+}]>;
+def VSLDOI_unary_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isVSLDOIShuffleMask(N, true) != -1;
+}], VSLDOI_unary_get_imm>;
+
+
+// VSPLT*_get_imm xform function: convert vector_shuffle mask to VSPLT* imm.
+def VSPLTB_get_imm : SDNodeXForm<build_vector, [{
+ return getI32Imm(PPC::getVSPLTImmediate(N, 1));
+}]>;
+def VSPLTB_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isSplatShuffleMask(N, 1);
+}], VSPLTB_get_imm>;
+def VSPLTH_get_imm : SDNodeXForm<build_vector, [{
+ return getI32Imm(PPC::getVSPLTImmediate(N, 2));
+}]>;
+def VSPLTH_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isSplatShuffleMask(N, 2);
+}], VSPLTH_get_imm>;
+def VSPLTW_get_imm : SDNodeXForm<build_vector, [{
+ return getI32Imm(PPC::getVSPLTImmediate(N, 4));
+}]>;
+def VSPLTW_shuffle_mask : PatLeaf<(build_vector), [{
+ return PPC::isSplatShuffleMask(N, 4);
+}], VSPLTW_get_imm>;
// VSPLTISB_get_imm xform function: convert build_vector to VSPLTISB imm.
def VSPLTISB_get_imm : SDNodeXForm<build_vector, [{
- char Val;
- PPC::isVecSplatImm(N, 1, &Val);
- return getI32Imm(Val);
+ return PPC::get_VSPLTI_elt(N, 1, *CurDAG);
}]>;
def vecspltisb : PatLeaf<(build_vector), [{
- return PPC::isVecSplatImm(N, 1);
+ return PPC::get_VSPLTI_elt(N, 1, *CurDAG).Val != 0;
}], VSPLTISB_get_imm>;
// VSPLTISH_get_imm xform function: convert build_vector to VSPLTISH imm.
def VSPLTISH_get_imm : SDNodeXForm<build_vector, [{
- char Val;
- PPC::isVecSplatImm(N, 2, &Val);
- return getI32Imm(Val);
+ return PPC::get_VSPLTI_elt(N, 2, *CurDAG);
}]>;
def vecspltish : PatLeaf<(build_vector), [{
- return PPC::isVecSplatImm(N, 2);
+ return PPC::get_VSPLTI_elt(N, 2, *CurDAG).Val != 0;
}], VSPLTISH_get_imm>;
// VSPLTISW_get_imm xform function: convert build_vector to VSPLTISW imm.
def VSPLTISW_get_imm : SDNodeXForm<build_vector, [{
- char Val;
- PPC::isVecSplatImm(N, 4, &Val);
- return getI32Imm(Val);
+ return PPC::get_VSPLTI_elt(N, 4, *CurDAG);
}]>;
def vecspltisw : PatLeaf<(build_vector), [{
- return PPC::isVecSplatImm(N, 4);
+ return PPC::get_VSPLTI_elt(N, 4, *CurDAG).Val != 0;
}], VSPLTISW_get_imm>;
-class isVDOT { // vector dot instruction.
- list<Register> Defs = [CR6];
- bit RC = 1;
-}
-
//===----------------------------------------------------------------------===//
// Helpers for defining instructions that directly correspond to intrinsics.
//===----------------------------------------------------------------------===//
// Instruction Definitions.
-def IMPLICIT_DEF_VRRC : Pseudo<(ops VRRC:$rD), "; $rD = IMPLICIT_DEF_VRRC",
- [(set VRRC:$rD, (v4f32 (undef)))]>;
+def IMPLICIT_DEF_VRRC : Pseudo<(ops VRRC:$rD), "; IMPLICIT_DEF_VRRC $rD",
+ [(set VRRC:$rD, (v4i32 (undef)))]>;
+
+let noResults = 1 in {
+def DSS : DSS_Form<822, (ops u5imm:$A, u5imm:$STRM,u5imm:$ZERO1,u5imm:$ZERO2),
+ "dss $STRM, $A", LdStGeneral /*FIXME*/, []>;
+def DST : DSS_Form<342, (ops u5imm:$T, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
+ "dst $rA, $rB, $STRM, $T", LdStGeneral /*FIXME*/, []>;
+def DSTST : DSS_Form<374, (ops u5imm:$T, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
+ "dstst $rA, $rB, $STRM, $T", LdStGeneral /*FIXME*/, []>;
+}
+
+def MFVSCR : VXForm_4<1540, (ops VRRC:$vD),
+ "mfvcr $vD", LdStGeneral,
+ [(set VRRC:$vD, (int_ppc_altivec_mfvscr))]>;
+def MTVSCR : VXForm_5<1604, (ops VRRC:$vB),
+ "mtvcr $vB", LdStGeneral,
+ [(int_ppc_altivec_mtvscr VRRC:$vB)]>;
let isLoad = 1, PPC970_Unit = 2 in { // Loads.
def LVEBX: XForm_1<31, 7, (ops VRRC:$vD, memrr:$src),
[(set VRRC:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
PPC970_Unit_LSU;
def LVSR : XForm_1<31, 38, (ops VRRC:$vD, memrr:$src),
- "lvsl $vD, $src", LdStGeneral,
+ "lvsr $vD, $src", LdStGeneral,
[(set VRRC:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
PPC970_Unit_LSU;
[(set VRRC:$vD, (fneg (fsub (fmul VRRC:$vA, VRRC:$vC),
VRRC:$vB)))]>,
Requires<[FPContractions]>;
+
def VMHADDSHS : VA1a_Int<32, "vmhaddshs", int_ppc_altivec_vmhaddshs>;
def VMHRADDSHS : VA1a_Int<33, "vmhraddshs", int_ppc_altivec_vmhraddshs>;
+def VMLADDUHM : VA1a_Int<34, "vmladduhm", int_ppc_altivec_vmladduhm>;
def VPERM : VA1a_Int<43, "vperm", int_ppc_altivec_vperm>;
def VSEL : VA1a_Int<42, "vsel", int_ppc_altivec_vsel>;
+// Shuffles.
def VSLDOI : VAForm_2<44, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB, u5imm:$SH),
"vsldoi $vD, $vA, $vB, $SH", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vsldoi VRRC:$vA, VRRC:$vB,
- imm:$SH))]>;
+ [(set VRRC:$vD,
+ (vector_shuffle (v16i8 VRRC:$vA), VRRC:$vB,
+ VSLDOI_shuffle_mask:$SH))]>;
// VX-Form instructions. AltiVec arithmetic ops.
-def VADDCUW : VXForm_1<384, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vaddcuw $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vaddcuw VRRC:$vA, VRRC:$vB))]>;
def VADDFP : VXForm_1<10, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
"vaddfp $vD, $vA, $vB", VecFP,
[(set VRRC:$vD, (fadd VRRC:$vA, VRRC:$vB))]>;
"vadduwm $vD, $vA, $vB", VecGeneral,
[(set VRRC:$vD, (add (v4i32 VRRC:$vA), VRRC:$vB))]>;
-def VADDSBS : VXForm_1<768, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vaddsbs $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vaddsbs VRRC:$vA, VRRC:$vB))]>;
-def VADDSHS : VXForm_1<832, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vaddshs $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vaddshs VRRC:$vA, VRRC:$vB))]>;
-def VADDSWS : VXForm_1<896, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vaddsws $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vaddsws VRRC:$vA, VRRC:$vB))]>;
+def VADDCUW : VX1_Int<384, "vaddcuw", int_ppc_altivec_vaddcuw>;
+def VADDSBS : VX1_Int<768, "vaddsbs", int_ppc_altivec_vaddsbs>;
+def VADDSHS : VX1_Int<832, "vaddshs", int_ppc_altivec_vaddshs>;
+def VADDSWS : VX1_Int<896, "vaddsws", int_ppc_altivec_vaddsws>;
+def VADDUBS : VX1_Int<512, "vaddubs", int_ppc_altivec_vaddubs>;
+def VADDUHS : VX1_Int<576, "vadduhs", int_ppc_altivec_vadduhs>;
+def VADDUWS : VX1_Int<640, "vadduws", int_ppc_altivec_vadduws>;
+
-def VADDUBS : VXForm_1<512, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vaddubs $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vaddubs VRRC:$vA, VRRC:$vB))]>;
-def VADDUHS : VXForm_1<576, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vadduhs $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vadduhs VRRC:$vA, VRRC:$vB))]>;
-def VADDUWS : VXForm_1<640, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vadduws $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD,
- (int_ppc_altivec_vadduws VRRC:$vA, VRRC:$vB))]>;
def VAND : VXForm_1<1028, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
"vand $vD, $vA, $vB", VecFP,
[(set VRRC:$vD, (and (v4i32 VRRC:$vA), VRRC:$vB))]>;
(int_ppc_altivec_vcfux VRRC:$vB, imm:$UIMM))]>;
def VCTSXS : VXForm_1<970, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
"vctsxs $vD, $vB, $UIMM", VecFP,
- []>;
+ [(set VRRC:$vD,
+ (int_ppc_altivec_vctsxs VRRC:$vB, imm:$UIMM))]>;
def VCTUXS : VXForm_1<906, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
"vctuxs $vD, $vB, $UIMM", VecFP,
- []>;
-def VEXPTEFP : VXForm_2<394, (ops VRRC:$vD, VRRC:$vB),
- "vexptefp $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vexptefp VRRC:$vB))]>;
-def VLOGEFP : VXForm_2<458, (ops VRRC:$vD, VRRC:$vB),
- "vlogefp $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vlogefp VRRC:$vB))]>;
-def VMAXFP : VXForm_1<1034, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vmaxfp $vD, $vA, $vB", VecFP,
- []>;
-def VMINFP : VXForm_1<1098, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
- "vminfp $vD, $vA, $vB", VecFP,
- []>;
-
-def VMRGHH : VX1_Int<76 , "vmrghh", int_ppc_altivec_vmrghh>;
-def VMRGHW : VX1_Int<140, "vmrghw", int_ppc_altivec_vmrghw>;
-def VMRGLH : VX1_Int<332, "vmrglh", int_ppc_altivec_vmrglh>;
-def VMRGLW : VX1_Int<396, "vmrglw", int_ppc_altivec_vmrglw>;
+ [(set VRRC:$vD,
+ (int_ppc_altivec_vctuxs VRRC:$vB, imm:$UIMM))]>;
+def VEXPTEFP : VX2_Int<394, "vexptefp", int_ppc_altivec_vexptefp>;
+def VLOGEFP : VX2_Int<458, "vlogefp", int_ppc_altivec_vlogefp>;
+
+def VAVGSB : VX1_Int<1282, "vavgsb", int_ppc_altivec_vavgsb>;
+def VAVGSH : VX1_Int<1346, "vavgsh", int_ppc_altivec_vavgsh>;
+def VAVGSW : VX1_Int<1410, "vavgsw", int_ppc_altivec_vavgsw>;
+def VAVGUB : VX1_Int<1026, "vavgub", int_ppc_altivec_vavgub>;
+def VAVGUH : VX1_Int<1090, "vavguh", int_ppc_altivec_vavguh>;
+def VAVGUW : VX1_Int<1154, "vavguw", int_ppc_altivec_vavguw>;
+
+def VMAXFP : VX1_Int<1034, "vmaxfp", int_ppc_altivec_vmaxfp>;
+def VMAXSB : VX1_Int< 258, "vmaxsb", int_ppc_altivec_vmaxsb>;
+def VMAXSH : VX1_Int< 322, "vmaxsh", int_ppc_altivec_vmaxsh>;
+def VMAXSW : VX1_Int< 386, "vmaxsw", int_ppc_altivec_vmaxsw>;
+def VMAXUB : VX1_Int< 2, "vmaxub", int_ppc_altivec_vmaxub>;
+def VMAXUH : VX1_Int< 66, "vmaxuh", int_ppc_altivec_vmaxuh>;
+def VMAXUW : VX1_Int< 130, "vmaxuw", int_ppc_altivec_vmaxuw>;
+def VMINFP : VX1_Int<1098, "vminfp", int_ppc_altivec_vminfp>;
+def VMINSB : VX1_Int< 770, "vminsb", int_ppc_altivec_vminsb>;
+def VMINSH : VX1_Int< 834, "vminsh", int_ppc_altivec_vminsh>;
+def VMINSW : VX1_Int< 896, "vminsw", int_ppc_altivec_vminsw>;
+def VMINUB : VX1_Int< 514, "vminub", int_ppc_altivec_vminub>;
+def VMINUH : VX1_Int< 578, "vminuh", int_ppc_altivec_vminuh>;
+def VMINUW : VX1_Int< 642, "vminuw", int_ppc_altivec_vminuw>;
+
+def VMRGHB : VXForm_1< 12, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
+ "vmrghb $vD, $vA, $vB", VecFP,
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VMRGHB_shuffle_mask))]>;
+def VMRGHH : VXForm_1< 76, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
+ "vmrghh $vD, $vA, $vB", VecFP,
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VMRGHH_shuffle_mask))]>;
+def VMRGHW : VXForm_1<140, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
+ "vmrghw $vD, $vA, $vB", VecFP,
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VMRGHW_shuffle_mask))]>;
+def VMRGLB : VXForm_1<268, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
+ "vmrglb $vD, $vA, $vB", VecFP,
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VMRGLB_shuffle_mask))]>;
+def VMRGLH : VXForm_1<332, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
+ "vmrglh $vD, $vA, $vB", VecFP,
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VMRGLH_shuffle_mask))]>;
+def VMRGLW : VXForm_1<396, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
+ "vmrglw $vD, $vA, $vB", VecFP,
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VMRGLW_shuffle_mask))]>;
def VMSUMMBM : VA1a_Int<37, "vmsummbm", int_ppc_altivec_vmsummbm>;
def VMSUMSHM : VA1a_Int<40, "vmsumshm", int_ppc_altivec_vmsumshm>;
def VRLB : VX1_Int< 4, "vrlb", int_ppc_altivec_vrlb>;
def VRLH : VX1_Int< 68, "vrlh", int_ppc_altivec_vrlh>;
def VRLW : VX1_Int< 132, "vrlw", int_ppc_altivec_vrlw>;
+
+def VSL : VX1_Int< 452, "vsl" , int_ppc_altivec_vsl >;
def VSLO : VX1_Int<1036, "vslo", int_ppc_altivec_vslo>;
def VSLB : VX1_Int< 260, "vslb", int_ppc_altivec_vslb>;
def VSLH : VX1_Int< 324, "vslh", int_ppc_altivec_vslh>;
def VSPLTB : VXForm_1<524, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
"vspltb $vD, $vB, $UIMM", VecPerm,
- []>;
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vB), (undef),
+ VSPLTB_shuffle_mask:$UIMM))]>;
def VSPLTH : VXForm_1<588, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
"vsplth $vD, $vB, $UIMM", VecPerm,
- []>;
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vB), (undef),
+ VSPLTH_shuffle_mask:$UIMM))]>;
def VSPLTW : VXForm_1<652, (ops VRRC:$vD, u5imm:$UIMM, VRRC:$vB),
"vspltw $vD, $vB, $UIMM", VecPerm,
- [(set VRRC:$vD, (vector_shuffle (v4f32 VRRC:$vB), (undef),
- VSPLT_shuffle_mask:$UIMM))]>;
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vB), (undef),
+ VSPLTW_shuffle_mask:$UIMM))]>;
def VSR : VX1_Int< 708, "vsr" , int_ppc_altivec_vsr>;
def VSRO : VX1_Int<1100, "vsro" , int_ppc_altivec_vsro>;
def VSPLTISB : VXForm_3<780, (ops VRRC:$vD, s5imm:$SIMM),
"vspltisb $vD, $SIMM", VecPerm,
- [(set VRRC:$vD, (v4f32 vecspltisb:$SIMM))]>;
+ [(set VRRC:$vD, (v16i8 vecspltisb:$SIMM))]>;
def VSPLTISH : VXForm_3<844, (ops VRRC:$vD, s5imm:$SIMM),
"vspltish $vD, $SIMM", VecPerm,
- [(set VRRC:$vD, (v4f32 vecspltish:$SIMM))]>;
+ [(set VRRC:$vD, (v8i16 vecspltish:$SIMM))]>;
def VSPLTISW : VXForm_3<908, (ops VRRC:$vD, s5imm:$SIMM),
"vspltisw $vD, $SIMM", VecPerm,
- [(set VRRC:$vD, (v4f32 vecspltisw:$SIMM))]>;
+ [(set VRRC:$vD, (v4i32 vecspltisw:$SIMM))]>;
// Vector Pack.
def VPKPX : VX1_Int<782, "vpkpx", int_ppc_altivec_vpkpx>;
def VPKSWUS : VX1_Int<334, "vpkswus", int_ppc_altivec_vpkswus>;
def VPKUHUM : VXForm_1<14, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
"vpkuhum $vD, $vA, $vB", VecFP,
- [/*TODO*/]>;
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VPKUHUM_shuffle_mask))]>;
def VPKUHUS : VX1_Int<142, "vpkuhus", int_ppc_altivec_vpkuhus>;
def VPKUWUM : VXForm_1<78, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB),
"vpkuwum $vD, $vA, $vB", VecFP,
- [/*TODO*/]>;
+ [(set VRRC:$vD, (vector_shuffle (v16i8 VRRC:$vA),
+ VRRC:$vB, VPKUWUM_shuffle_mask))]>;
def VPKUWUS : VX1_Int<206, "vpkuwus", int_ppc_altivec_vpkuwus>;
// Vector Unpack.
-def VUPKHPX : VXForm_2<846, (ops VRRC:$vD, VRRC:$vB),
- "vupkhpx $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vupkhpx VRRC:$vB))]>;
-def VUPKHSB : VXForm_2<526, (ops VRRC:$vD, VRRC:$vB),
- "vupkhsb $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vupkhsb VRRC:$vB))]>;
-def VUPKHSH : VXForm_2<590, (ops VRRC:$vD, VRRC:$vB),
- "vupkhsh $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vupkhsh VRRC:$vB))]>;
-def VUPKLPX : VXForm_2<974, (ops VRRC:$vD, VRRC:$vB),
- "vupklpx $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vupklpx VRRC:$vB))]>;
-def VUPKLSB : VXForm_2<654, (ops VRRC:$vD, VRRC:$vB),
- "vupklsb $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vupklsb VRRC:$vB))]>;
-def VUPKLSH : VXForm_2<718, (ops VRRC:$vD, VRRC:$vB),
- "vupklsh $vD, $vB", VecFP,
- [(set VRRC:$vD, (int_ppc_altivec_vupklsh VRRC:$vB))]>;
+def VUPKHPX : VX2_Int<846, "vupkhpx", int_ppc_altivec_vupkhpx>;
+def VUPKHSB : VX2_Int<526, "vupkhsb", int_ppc_altivec_vupkhsb>;
+def VUPKHSH : VX2_Int<590, "vupkhsh", int_ppc_altivec_vupkhsh>;
+def VUPKLPX : VX2_Int<974, "vupklpx", int_ppc_altivec_vupklpx>;
+def VUPKLSB : VX2_Int<654, "vupklsb", int_ppc_altivec_vupklsb>;
+def VUPKLSH : VX2_Int<718, "vupklsh", int_ppc_altivec_vupklsh>;
// Altivec Comparisons.
[(set VRRC:$vD, (Ty (PPCvcmp VRRC:$vA, VRRC:$vB, xo)))]>;
class VCMPo<bits<10> xo, string asmstr, ValueType Ty>
: VXRForm_1<xo, (ops VRRC:$vD, VRRC:$vA, VRRC:$vB), asmstr, VecFPCompare,
- [(set VRRC:$vD, (Ty (PPCvcmp_o VRRC:$vA, VRRC:$vB, xo)))]>,isVDOT;
+ [(set VRRC:$vD, (Ty (PPCvcmp_o VRRC:$vA, VRRC:$vB, xo)))]> {
+ let Defs = [CR6];
+ let RC = 1;
+}
// f32 element comparisons.0
def VCMPBFP : VCMP <966, "vcmpbfp $vD, $vA, $vB" , v4f32>;
def V_SET0 : VXForm_setzero<1220, (ops VRRC:$vD),
"vxor $vD, $vD, $vD", VecFP,
- [(set VRRC:$vD, (v4f32 immAllZerosV))]>;
+ [(set VRRC:$vD, (v4i32 immAllZerosV))]>;
}
//===----------------------------------------------------------------------===//
// Additional Altivec Patterns
//
-// Undef/Zero.
-def : Pat<(v16i8 (undef)), (v16i8 (IMPLICIT_DEF_VRRC))>;
-def : Pat<(v8i16 (undef)), (v8i16 (IMPLICIT_DEF_VRRC))>;
-def : Pat<(v4i32 (undef)), (v4i32 (IMPLICIT_DEF_VRRC))>;
-def : Pat<(v16i8 immAllZerosV), (v16i8 (V_SET0))>;
-def : Pat<(v8i16 immAllZerosV), (v8i16 (V_SET0))>;
-def : Pat<(v4i32 immAllZerosV), (v4i32 (V_SET0))>;
+// DS* intrinsics.
+def : Pat<(int_ppc_altivec_dss imm:$STRM), (DSS 0, imm:$STRM, 0, 0)>;
+def : Pat<(int_ppc_altivec_dssall), (DSS 1, 0, 0, 0)>;
+def : Pat<(int_ppc_altivec_dst GPRC:$rA, GPRC:$rB, imm:$STRM),
+ (DST 0, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
+def : Pat<(int_ppc_altivec_dstt GPRC:$rA, GPRC:$rB, imm:$STRM),
+ (DST 1, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
+def : Pat<(int_ppc_altivec_dstst GPRC:$rA, GPRC:$rB, imm:$STRM),
+ (DSTST 0, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
+def : Pat<(int_ppc_altivec_dststt GPRC:$rA, GPRC:$rB, imm:$STRM),
+ (DSTST 1, imm:$STRM, GPRC:$rA, GPRC:$rB)>;
+
+// Undef.
+def : Pat<(v16i8 (undef)), (IMPLICIT_DEF_VRRC)>;
+def : Pat<(v8i16 (undef)), (IMPLICIT_DEF_VRRC)>;
+def : Pat<(v4f32 (undef)), (IMPLICIT_DEF_VRRC)>;
// Loads.
-def : Pat<(v16i8 (load xoaddr:$src)), (v16i8 (LVX xoaddr:$src))>;
-def : Pat<(v8i16 (load xoaddr:$src)), (v8i16 (LVX xoaddr:$src))>;
-def : Pat<(v4i32 (load xoaddr:$src)), (v4i32 (LVX xoaddr:$src))>;
-def : Pat<(v4f32 (load xoaddr:$src)), (v4f32 (LVX xoaddr:$src))>;
+def : Pat<(v4i32 (load xoaddr:$src)), (LVX xoaddr:$src)>;
// Stores.
-def : Pat<(store (v16i8 VRRC:$rS), xoaddr:$dst),
- (STVX (v16i8 VRRC:$rS), xoaddr:$dst)>;
-def : Pat<(store (v8i16 VRRC:$rS), xoaddr:$dst),
- (STVX (v8i16 VRRC:$rS), xoaddr:$dst)>;
def : Pat<(store (v4i32 VRRC:$rS), xoaddr:$dst),
(STVX (v4i32 VRRC:$rS), xoaddr:$dst)>;
-def : Pat<(store (v4f32 VRRC:$rS), xoaddr:$dst),
- (STVX (v4f32 VRRC:$rS), xoaddr:$dst)>;
// Bit conversions.
def : Pat<(v16i8 (bitconvert (v8i16 VRRC:$src))), (v16i8 VRRC:$src)>;
def : Pat<(v4f32 (bitconvert (v8i16 VRRC:$src))), (v4f32 VRRC:$src)>;
def : Pat<(v4f32 (bitconvert (v4i32 VRRC:$src))), (v4f32 VRRC:$src)>;
-// Immediate vector formation with vsplti*.
-def : Pat<(v16i8 vecspltisb:$invec), (v16i8 (VSPLTISB vecspltisb:$invec))>;
-def : Pat<(v16i8 vecspltish:$invec), (v16i8 (VSPLTISH vecspltish:$invec))>;
-def : Pat<(v16i8 vecspltisw:$invec), (v16i8 (VSPLTISW vecspltisw:$invec))>;
-
-def : Pat<(v8i16 vecspltisb:$invec), (v8i16 (VSPLTISB vecspltisb:$invec))>;
-def : Pat<(v8i16 vecspltish:$invec), (v8i16 (VSPLTISH vecspltish:$invec))>;
-def : Pat<(v8i16 vecspltisw:$invec), (v8i16 (VSPLTISW vecspltisw:$invec))>;
-
-def : Pat<(v4i32 vecspltisb:$invec), (v4i32 (VSPLTISB vecspltisb:$invec))>;
-def : Pat<(v4i32 vecspltish:$invec), (v4i32 (VSPLTISH vecspltish:$invec))>;
-def : Pat<(v4i32 vecspltisw:$invec), (v4i32 (VSPLTISW vecspltisw:$invec))>;
+// Shuffles.
+
+// Match vsldoi(x,x), vpkuwum(x,x), vpkuhum(x,x)
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VSLDOI_unary_shuffle_mask:$in),
+ (VSLDOI VRRC:$vA, VRRC:$vA, VSLDOI_unary_shuffle_mask:$in)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef,VPKUWUM_unary_shuffle_mask:$in),
+ (VPKUWUM VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef,VPKUHUM_unary_shuffle_mask:$in),
+ (VPKUHUM VRRC:$vA, VRRC:$vA)>;
+
+// Match vmrg*(x,x)
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VMRGLB_unary_shuffle_mask:$in),
+ (VMRGLB VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VMRGLH_unary_shuffle_mask:$in),
+ (VMRGLH VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VMRGLW_unary_shuffle_mask:$in),
+ (VMRGLW VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VMRGHB_unary_shuffle_mask:$in),
+ (VMRGHB VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VMRGHH_unary_shuffle_mask:$in),
+ (VMRGHH VRRC:$vA, VRRC:$vA)>;
+def:Pat<(vector_shuffle (v16i8 VRRC:$vA), undef, VMRGHW_unary_shuffle_mask:$in),
+ (VMRGHW VRRC:$vA, VRRC:$vA)>;
// Logical Operations
-def : Pat<(v16i8 (and VRRC:$A, VRRC:$B)), (v16i8 (VAND VRRC:$A, VRRC:$B))>;
-def : Pat<(v8i16 (and VRRC:$A, VRRC:$B)), (v8i16 (VAND VRRC:$A, VRRC:$B))>;
-def : Pat<(v16i8 (or VRRC:$A, VRRC:$B)), (v16i8 (VOR VRRC:$A, VRRC:$B))>;
-def : Pat<(v8i16 (or VRRC:$A, VRRC:$B)), (v8i16 (VOR VRRC:$A, VRRC:$B))>;
-def : Pat<(v16i8 (xor VRRC:$A, VRRC:$B)), (v16i8 (VXOR VRRC:$A, VRRC:$B))>;
-def : Pat<(v8i16 (xor VRRC:$A, VRRC:$B)), (v8i16 (VXOR VRRC:$A, VRRC:$B))>;
-def : Pat<(v16i8 (vnot (or VRRC:$A, VRRC:$B))),(v16i8 (VNOR VRRC:$A, VRRC:$B))>;
-def : Pat<(v8i16 (vnot (or VRRC:$A, VRRC:$B))),(v8i16 (VNOR VRRC:$A, VRRC:$B))>;
-def : Pat<(v16i8 (and VRRC:$A, (vnot VRRC:$B))),
- (v16i8 (VANDC VRRC:$A, VRRC:$B))>;
-def : Pat<(v8i16 (and VRRC:$A, (vnot VRRC:$B))),
- (v8i16 (VANDC VRRC:$A, VRRC:$B))>;
+def : Pat<(v4i32 (vnot VRRC:$vA)), (VNOR VRRC:$vA, VRRC:$vA)>;
+def : Pat<(v4i32 (vnot_conv VRRC:$vA)), (VNOR VRRC:$vA, VRRC:$vA)>;
+
+def : Pat<(v4i32 (vnot_conv (or VRRC:$A, VRRC:$B))),
+ (VNOR VRRC:$A, VRRC:$B)>;
+def : Pat<(v4i32 (and VRRC:$A, (vnot_conv VRRC:$B))),
+ (VANDC VRRC:$A, VRRC:$B)>;
def : Pat<(fmul VRRC:$vA, VRRC:$vB),
- (VMADDFP VRRC:$vA, VRRC:$vB, (V_SET0))>;
+ (VMADDFP VRRC:$vA, VRRC:$vB, (v4i32 (V_SET0)))>;
// Fused multiply add and multiply sub for packed float. These are represented
// separately from the real instructions above, for operations that must have
(VMADDFP VRRC:$A, VRRC:$B, VRRC:$C)>;
def : Pat<(int_ppc_altivec_vnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
(VNMSUBFP VRRC:$A, VRRC:$B, VRRC:$C)>;
-def : Pat<(vector_shuffle (v4i32 VRRC:$vB), (undef), VSPLT_shuffle_mask:$UIMM),
- (v4i32 (VSPLTW VSPLT_shuffle_mask:$UIMM, VRRC:$vB))>;
-
-def : Pat<(PPCvperm (v4i32 VRRC:$vA), VRRC:$vB, VRRC:$vC),
- (v4i32 (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC))>;
-def : Pat<(PPCvperm (v4f32 VRRC:$vA), VRRC:$vB, VRRC:$vC),
- (v4f32 (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC))>;
-def : Pat<(PPCvperm (v8i16 VRRC:$vA), VRRC:$vB, VRRC:$vC),
- (v8i16 (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC))>;
+
def : Pat<(PPCvperm (v16i8 VRRC:$vA), VRRC:$vB, VRRC:$vC),
- (v16i8 (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC))>;
+ (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC)>;