-//===- PPCInstrAltivec.td - The PowerPC Altivec Extension --*- tablegen -*-===//
-//
+//===-- PPCInstrAltivec.td - The PowerPC Altivec Extension -*- tablegen -*-===//
+//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
-//
+//
//===----------------------------------------------------------------------===//
//
// This file describes the Altivec extension to the PowerPC instruction set.
// Altivec transformation functions and pattern fragments.
//
+// Since we canonicalize buildvectors to v16i8, all vnots "-1" operands will be
+// of that type.
+def vnot_ppc : PatFrag<(ops node:$in),
+ (xor node:$in, (bitconvert (v16i8 immAllOnesV)))>;
def vpkuhum_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
(vector_shuffle node:$lhs, node:$rhs), [{
def vmrglb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 1, false);
}]>;
def vmrglh_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 2, false);
}]>;
def vmrglw_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 4, false);
}]>;
def vmrghb_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 1, false);
}]>;
def vmrghh_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 2, false);
}]>;
def vmrghw_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGHShuffleMask(cast<ShuffleVectorSDNode>(N), 4, false);
}]>;
def vmrglb_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
- (vector_shuffle node:$lhs, node:$rhs), [{
+ (vector_shuffle (v16i8 node:$lhs), node:$rhs), [{
return PPC::isVMRGLShuffleMask(cast<ShuffleVectorSDNode>(N), 1, true);
}]>;
def vmrglh_unary_shuffle : PatFrag<(ops node:$lhs, node:$rhs),
return PPC::get_VSPLTI_elt(N, 4, *CurDAG).getNode() != 0;
}], VSPLTISW_get_imm>;
-def V_immneg0 : PatLeaf<(build_vector), [{
- return PPC::isAllNegativeZeroVector(N);
-}]>;
-
//===----------------------------------------------------------------------===//
// Helpers for defining instructions that directly correspond to intrinsics.
def DSS : DSS_Form<822, (outs),
(ins u5imm:$ZERO0, u5imm:$STRM,u5imm:$ZERO1,u5imm:$ZERO2),
- "dss $STRM", LdStGeneral /*FIXME*/, []>;
+ "dss $STRM", LdStLoad /*FIXME*/, []>;
def DSSALL : DSS_Form<822, (outs),
(ins u5imm:$ONE, u5imm:$ZERO0,u5imm:$ZERO1,u5imm:$ZERO2),
- "dssall", LdStGeneral /*FIXME*/, []>;
+ "dssall", LdStLoad /*FIXME*/, []>;
def DST : DSS_Form<342, (outs),
(ins u5imm:$ZERO, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
- "dst $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DSTT : DSS_Form<342, (outs),
(ins u5imm:$ONE, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
- "dstt $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dstt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DSTST : DSS_Form<374, (outs),
(ins u5imm:$ZERO, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
- "dstst $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dstst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DSTSTT : DSS_Form<374, (outs),
(ins u5imm:$ONE, u5imm:$STRM, GPRC:$rA, GPRC:$rB),
- "dststt $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dststt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DST64 : DSS_Form<342, (outs),
(ins u5imm:$ZERO, u5imm:$STRM, G8RC:$rA, GPRC:$rB),
- "dst $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DSTT64 : DSS_Form<342, (outs),
(ins u5imm:$ONE, u5imm:$STRM, G8RC:$rA, GPRC:$rB),
- "dstt $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dstt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DSTST64 : DSS_Form<374, (outs),
(ins u5imm:$ZERO, u5imm:$STRM, G8RC:$rA, GPRC:$rB),
- "dstst $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dstst $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def DSTSTT64 : DSS_Form<374, (outs),
(ins u5imm:$ONE, u5imm:$STRM, G8RC:$rA, GPRC:$rB),
- "dststt $rA, $rB, $STRM", LdStGeneral /*FIXME*/, []>;
+ "dststt $rA, $rB, $STRM", LdStLoad /*FIXME*/, []>;
def MFVSCR : VXForm_4<1540, (outs VRRC:$vD), (ins),
- "mfvscr $vD", LdStGeneral,
+ "mfvscr $vD", LdStStore,
[(set VRRC:$vD, (int_ppc_altivec_mfvscr))]>;
def MTVSCR : VXForm_5<1604, (outs), (ins VRRC:$vB),
- "mtvscr $vB", LdStGeneral,
+ "mtvscr $vB", LdStLoad,
[(int_ppc_altivec_mtvscr VRRC:$vB)]>;
let canFoldAsLoad = 1, PPC970_Unit = 2 in { // Loads.
def LVEBX: XForm_1<31, 7, (outs VRRC:$vD), (ins memrr:$src),
- "lvebx $vD, $src", LdStGeneral,
+ "lvebx $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvebx xoaddr:$src))]>;
def LVEHX: XForm_1<31, 39, (outs VRRC:$vD), (ins memrr:$src),
- "lvehx $vD, $src", LdStGeneral,
+ "lvehx $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvehx xoaddr:$src))]>;
def LVEWX: XForm_1<31, 71, (outs VRRC:$vD), (ins memrr:$src),
- "lvewx $vD, $src", LdStGeneral,
+ "lvewx $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvewx xoaddr:$src))]>;
def LVX : XForm_1<31, 103, (outs VRRC:$vD), (ins memrr:$src),
- "lvx $vD, $src", LdStGeneral,
+ "lvx $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvx xoaddr:$src))]>;
def LVXL : XForm_1<31, 359, (outs VRRC:$vD), (ins memrr:$src),
- "lvxl $vD, $src", LdStGeneral,
+ "lvxl $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvxl xoaddr:$src))]>;
}
def LVSL : XForm_1<31, 6, (outs VRRC:$vD), (ins memrr:$src),
- "lvsl $vD, $src", LdStGeneral,
+ "lvsl $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvsl xoaddr:$src))]>,
PPC970_Unit_LSU;
def LVSR : XForm_1<31, 38, (outs VRRC:$vD), (ins memrr:$src),
- "lvsr $vD, $src", LdStGeneral,
+ "lvsr $vD, $src", LdStLoad,
[(set VRRC:$vD, (int_ppc_altivec_lvsr xoaddr:$src))]>,
PPC970_Unit_LSU;
let PPC970_Unit = 2 in { // Stores.
def STVEBX: XForm_8<31, 135, (outs), (ins VRRC:$rS, memrr:$dst),
- "stvebx $rS, $dst", LdStGeneral,
+ "stvebx $rS, $dst", LdStStore,
[(int_ppc_altivec_stvebx VRRC:$rS, xoaddr:$dst)]>;
def STVEHX: XForm_8<31, 167, (outs), (ins VRRC:$rS, memrr:$dst),
- "stvehx $rS, $dst", LdStGeneral,
+ "stvehx $rS, $dst", LdStStore,
[(int_ppc_altivec_stvehx VRRC:$rS, xoaddr:$dst)]>;
def STVEWX: XForm_8<31, 199, (outs), (ins VRRC:$rS, memrr:$dst),
- "stvewx $rS, $dst", LdStGeneral,
+ "stvewx $rS, $dst", LdStStore,
[(int_ppc_altivec_stvewx VRRC:$rS, xoaddr:$dst)]>;
def STVX : XForm_8<31, 231, (outs), (ins VRRC:$rS, memrr:$dst),
- "stvx $rS, $dst", LdStGeneral,
+ "stvx $rS, $dst", LdStStore,
[(int_ppc_altivec_stvx VRRC:$rS, xoaddr:$dst)]>;
def STVXL : XForm_8<31, 487, (outs), (ins VRRC:$rS, memrr:$dst),
- "stvxl $rS, $dst", LdStGeneral,
+ "stvxl $rS, $dst", LdStStore,
[(int_ppc_altivec_stvxl VRRC:$rS, xoaddr:$dst)]>;
}
// VA-Form instructions. 3-input AltiVec ops.
def VMADDFP : VAForm_1<46, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vC, VRRC:$vB),
"vmaddfp $vD, $vA, $vC, $vB", VecFP,
- [(set VRRC:$vD, (fadd (fmul VRRC:$vA, VRRC:$vC),
- VRRC:$vB))]>,
- Requires<[FPContractions]>;
+ [(set VRRC:$vD, (fma VRRC:$vA, VRRC:$vC, VRRC:$vB))]>;
def VNMSUBFP: VAForm_1<47, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vC, VRRC:$vB),
"vnmsubfp $vD, $vA, $vC, $vB", VecFP,
- [(set VRRC:$vD, (fsub V_immneg0,
- (fsub (fmul VRRC:$vA, VRRC:$vC),
- VRRC:$vB)))]>,
- Requires<[FPContractions]>;
+ [(set VRRC:$vD, (fneg (fma VRRC:$vA, VRRC:$vC,
+ (fneg VRRC:$vB))))]>;
def VMHADDSHS : VA1a_Int<32, "vmhaddshs", int_ppc_altivec_vmhaddshs>;
def VMHRADDSHS : VA1a_Int<33, "vmhraddshs", int_ppc_altivec_vmhraddshs>;
[(set VRRC:$vD, (and (v4i32 VRRC:$vA), VRRC:$vB))]>;
def VANDC : VXForm_1<1092, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
"vandc $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD, (and (v4i32 VRRC:$vA), (vnot VRRC:$vB)))]>;
+ [(set VRRC:$vD, (and (v4i32 VRRC:$vA),
+ (vnot_ppc VRRC:$vB)))]>;
def VCFSX : VXForm_1<842, (outs VRRC:$vD), (ins u5imm:$UIMM, VRRC:$vB),
"vcfsx $vD, $vB, $UIMM", VecFP,
"vctuxs $vD, $vB, $UIMM", VecFP,
[(set VRRC:$vD,
(int_ppc_altivec_vctuxs VRRC:$vB, imm:$UIMM))]>;
+
+// Defines with the UIM field set to 0 for floating-point
+// to integer (fp_to_sint/fp_to_uint) conversions and integer
+// to floating-point (sint_to_fp/uint_to_fp) conversions.
+let VA = 0 in {
+def VCFSX_0 : VXForm_1<842, (outs VRRC:$vD), (ins VRRC:$vB),
+ "vcfsx $vD, $vB, 0", VecFP,
+ [(set VRRC:$vD,
+ (int_ppc_altivec_vcfsx VRRC:$vB, 0))]>;
+def VCTUXS_0 : VXForm_1<906, (outs VRRC:$vD), (ins VRRC:$vB),
+ "vctuxs $vD, $vB, 0", VecFP,
+ [(set VRRC:$vD,
+ (int_ppc_altivec_vctuxs VRRC:$vB, 0))]>;
+def VCFUX_0 : VXForm_1<778, (outs VRRC:$vD), (ins VRRC:$vB),
+ "vcfux $vD, $vB, 0", VecFP,
+ [(set VRRC:$vD,
+ (int_ppc_altivec_vcfux VRRC:$vB, 0))]>;
+def VCTSXS_0 : VXForm_1<970, (outs VRRC:$vD), (ins VRRC:$vB),
+ "vctsxs $vD, $vB, 0", VecFP,
+ [(set VRRC:$vD,
+ (int_ppc_altivec_vctsxs VRRC:$vB, 0))]>;
+}
def VEXPTEFP : VX2_Int<394, "vexptefp", int_ppc_altivec_vexptefp>;
def VLOGEFP : VX2_Int<458, "vlogefp", int_ppc_altivec_vlogefp>;
def VNOR : VXForm_1<1284, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
"vnor $vD, $vA, $vB", VecFP,
- [(set VRRC:$vD, (vnot (or (v4i32 VRRC:$vA), VRRC:$vB)))]>;
+ [(set VRRC:$vD, (vnot_ppc (or (v4i32 VRRC:$vA),
+ VRRC:$vB)))]>;
def VOR : VXForm_1<1156, (outs VRRC:$vD), (ins VRRC:$vA, VRRC:$vB),
"vor $vD, $vA, $vB", VecFP,
[(set VRRC:$vD, (or (v4i32 VRRC:$vA), VRRC:$vB))]>;
def V_SET0 : VXForm_setzero<1220, (outs VRRC:$vD), (ins),
"vxor $vD, $vD, $vD", VecFP,
[(set VRRC:$vD, (v4i32 immAllZerosV))]>;
+let IMM=-1 in {
+def V_SETALLONES : VXForm_3<908, (outs VRRC:$vD), (ins),
+ "vspltisw $vD, -1", VecFP,
+ [(set VRRC:$vD, (v4i32 immAllOnesV))]>;
}
+} // VALU Operations.
//===----------------------------------------------------------------------===//
// Additional Altivec Patterns
(VMRGHW VRRC:$vA, VRRC:$vA)>;
// Logical Operations
-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_ppc VRRC:$vA)), (VNOR VRRC:$vA, VRRC:$vA)>;
-def : Pat<(v4i32 (vnot_conv (or VRRC:$A, VRRC:$B))),
+def : Pat<(v4i32 (vnot_ppc (or VRRC:$A, VRRC:$B))),
(VNOR VRRC:$A, VRRC:$B)>;
-def : Pat<(v4i32 (and VRRC:$A, (vnot_conv VRRC:$B))),
+def : Pat<(v4i32 (and VRRC:$A, (vnot_ppc VRRC:$B))),
(VANDC VRRC:$A, VRRC:$B)>;
def : Pat<(fmul VRRC:$vA, VRRC:$vB),
- (VMADDFP VRRC:$vA, VRRC:$vB, (v4i32 (V_SET0)))>;
+ (VMADDFP VRRC:$vA, VRRC:$vB,
+ (v4i32 (VSLW (V_SETALLONES), (V_SETALLONES))))>;
// Fused multiply add and multiply sub for packed float. These are represented
// separately from the real instructions above, for operations that must have
def : Pat<(PPCvperm (v16i8 VRRC:$vA), VRRC:$vB, VRRC:$vC),
(VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC)>;
+
+// Vector shifts
+def : Pat<(v16i8 (shl (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
+ (v16i8 (VSLB VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v8i16 (shl (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
+ (v8i16 (VSLH VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v4i32 (shl (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
+ (v4i32 (VSLW VRRC:$vA, VRRC:$vB))>;
+
+def : Pat<(v16i8 (srl (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
+ (v16i8 (VSRB VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v8i16 (srl (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
+ (v8i16 (VSRH VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v4i32 (srl (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
+ (v4i32 (VSRW VRRC:$vA, VRRC:$vB))>;
+
+def : Pat<(v16i8 (sra (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
+ (v16i8 (VSRAB VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v8i16 (sra (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
+ (v8i16 (VSRAH VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v4i32 (sra (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
+ (v4i32 (VSRAW VRRC:$vA, VRRC:$vB))>;
+
+// Float to integer and integer to float conversions
+def : Pat<(v4i32 (fp_to_sint (v4f32 VRRC:$vA))),
+ (VCTSXS_0 VRRC:$vA)>;
+def : Pat<(v4i32 (fp_to_uint (v4f32 VRRC:$vA))),
+ (VCTUXS_0 VRRC:$vA)>;
+def : Pat<(v4f32 (sint_to_fp (v4i32 VRRC:$vA))),
+ (VCFSX_0 VRRC:$vA)>;
+def : Pat<(v4f32 (uint_to_fp (v4i32 VRRC:$vA))),
+ (VCFUX_0 VRRC:$vA)>;
+
+// Floating-point rounding
+def : Pat<(v4f32 (ffloor (v4f32 VRRC:$vA))),
+ (VRFIM VRRC:$vA)>;
+def : Pat<(v4f32 (fceil (v4f32 VRRC:$vA))),
+ (VRFIP VRRC:$vA)>;
+def : Pat<(v4f32 (ftrunc (v4f32 VRRC:$vA))),
+ (VRFIZ VRRC:$vA)>;
+def : Pat<(v4f32 (fnearbyint (v4f32 VRRC:$vA))),
+ (VRFIN VRRC:$vA)>;
projects / oota-llvm.git / blobdiff