[AArch64] Refine Falkor Machine Model - Part1

llvm-svn: 298768

diff --git a/llvm/lib/Target/AArch64/AArch64SchedFalkor.td b/llvm/lib/Target/AArch64/AArch64SchedFalkor.td
index ddb09ac..caf84aa 100644 (file)
--- a/llvm/lib/Target/AArch64/AArch64SchedFalkor.td
+++ b/llvm/lib/Target/AArch64/AArch64SchedFalkor.td
@@ -95,7 +95,7 @@ def : WriteRes<WriteFDiv,  [FalkorUnitVXVY, FalkorUnitVXVY]>
       { let Latency = 12; let NumMicroOps = 2; } // Fragent -1 / NoRSV +1
 def : WriteRes<WriteV,     [FalkorUnitVXVY]> { let Latency = 6; }
 def : WriteRes<WriteVLD,   [FalkorUnitLD]> { let Latency = 3; }
-def : WriteRes<WriteVST,   [FalkorUnitST]> { let Latency = 4; }
+def : WriteRes<WriteVST,   [FalkorUnitST]> { let Latency = 0; }
 
 def : WriteRes<WriteSys,     []> { let Latency = 1; }
 def : WriteRes<WriteBarrier, []> { let Latency = 1; }
@@ -116,93 +116,8 @@ def : ReadAdvance<ReadExtrHi,  0>;
 def : ReadAdvance<ReadAdrBase, 0>;
 def : ReadAdvance<ReadVLD,     0>;
 
-//===----------------------------------------------------------------------===//
-// Specialize the coarse model by associating instruction groups with the
-// subtarget-defined types. As the modeled is refined, this will override most
-// of the above SchedWriteRes and SchedAlias mappings.
-
-// -----------------------------------------------------------------------------
-// Miscellaneous
-// -----------------------------------------------------------------------------
-
-def : InstRW<[WriteI], (instrs COPY)>;
-
-// -----------------------------------------------------------------------------
-// Vector Loads
-// -----------------------------------------------------------------------------
-def : InstRW<[WriteVLD], (instregex "LD1i(8|16|32|64)$")>;
-def : InstRW<[WriteVLD], (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD1Twov(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD1Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD1i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD1Twov(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD1Threev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-
-def : InstRW<[WriteVLD], (instregex "LD2i(8|16|32|64)$")>;
-def : InstRW<[WriteVLD], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD2Twov(8b|4h|2s)$")>;
-def : InstRW<[WriteVLD], (instregex "LD2Twov(16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2i(8|16|32|64)(_POST)?$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)(_POST)?$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2Twov(8b|4h|2s)(_POST)?$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2Twov(16b|8h|4s|2d)(_POST)?$")>;
-
-def : InstRW<[WriteVLD], (instregex "LD3i(8|16|32|64)$")>;
-def : InstRW<[WriteVLD], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD3Threev(8b|4h|2s|1d|16b|8h|4s)$")>;
-def : InstRW<[WriteVLD], (instregex "LD3Threev(2d)$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3Threev(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3Threev(2d)_POST$")>;
-
-def : InstRW<[WriteVLD], (instregex "LD4i(8|16|32|64)$")>;
-def : InstRW<[WriteVLD], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVLD], (instregex "LD4Fourv(8b|4h|2s|1d|16b|8h|4s)$")>;
-def : InstRW<[WriteVLD], (instregex "LD4Fourv(2d)$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4Fourv(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
-def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4Fourv(2d)_POST$")>;
-
-// -----------------------------------------------------------------------------
-// Vector Stores
+// Detailed Refinements
 // -----------------------------------------------------------------------------
-def : InstRW<[WriteVST], (instregex "ST1i(8|16|32|64)$")>;
-def : InstRW<[WriteVST], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVST], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVST], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVST], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST1i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
-
-def : InstRW<[WriteVST], (instregex "ST2i(8|16|32|64)$")>;
-def : InstRW<[WriteVST], (instregex "ST2Twov(8b|4h|2s)$")>;
-def : InstRW<[WriteVST], (instregex "ST2Twov(16b|8h|4s|2d)$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST2i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST2Twov(8b|4h|2s)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST2Twov(16b|8h|4s|2d)_POST$")>;
-
-def : InstRW<[WriteVST], (instregex "ST3i(8|16|32|64)$")>;
-def : InstRW<[WriteVST], (instregex "ST3Threev(8b|4h|2s|1d|16b|8h|4s)$")>;
-def : InstRW<[WriteVST], (instregex "ST3Threev(2d)$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST3i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST3Threev(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST3Threev(2d)_POST$")>;
-
-def : InstRW<[WriteVST], (instregex "ST4i(8|16|32|64)$")>;
-def : InstRW<[WriteVST], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s)$")>;
-def : InstRW<[WriteVST], (instregex "ST4Fourv(2d)$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST4i(8|16|32|64)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
-def : InstRW<[WriteVST, WriteAdr], (instregex "ST4Fourv(2d)_POST$")>;
+include "AArch64SchedFalkorDetails.td"
 
 }

diff --git a/llvm/lib/Target/AArch64/AArch64SchedFalkorDetails.td b/llvm/lib/Target/AArch64/AArch64SchedFalkorDetails.td
new file mode 100644 (file)
index 0000000..4e1ae37
--- /dev/null
+++ b/llvm/lib/Target/AArch64/AArch64SchedFalkorDetails.td
@@ -0,0 +1,151 @@
+//==- AArch64SchedFalkorDetails.td - Falkor Scheduling Defs -*- tablegen -*-==//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the uop and latency details for the machine model for the
+// Qualcomm Falkor subtarget.
+//
+//===----------------------------------------------------------------------===//
+
+include "AArch64SchedFalkorWriteRes.td"
+
+//===----------------------------------------------------------------------===//
+// Specialize the coarse model by associating instruction groups with the
+// subtarget-defined types. As the modeled is refined, this will override most
+// of the earlier mappings.
+
+// Miscellaneous
+// -----------------------------------------------------------------------------
+
+def : InstRW<[WriteI], (instrs COPY)>;
+
+// SIMD Floating-point Instructions
+// -----------------------------------------------------------------------------
+def : InstRW<[FalkorWr_1VXVY_1cyc],   (instregex "^(FABS|FNEG)(v2f32|v4f16)$")>;
+
+def : InstRW<[FalkorWr_1VXVY_2cyc],   (instregex "^(F(MAX|MIN)(NM)?P?|FAC(GE|GT))(v2f32|v4f16|v2i16p|v2i32p)$")>;
+def : InstRW<[FalkorWr_1VXVY_2cyc],   (instregex "^FAC(GE|GT)(16|32|64)$")>;
+def : InstRW<[FalkorWr_1VXVY_2cyc],   (instregex "^FCM(EQ|GE|GT)(16|32|64|v2f32|v4f16|v2i32|v4i16)$")>;
+def : InstRW<[FalkorWr_1VXVY_2cyc],   (instregex "^FCM(EQ|LE|GE|GT|LT)(v1i16|v1i32|v1i64|v2i32|v4i16)rz$")>;
+def : InstRW<[FalkorWr_1VXVY_2cyc],   (instregex "^FRINT(A|I|M|N|P|X|Z)(v2f32|v4f16)$")>;
+
+def : InstRW<[FalkorWr_1VXVY_3cyc],   (instregex "^F(MAX|MIN)(NM)?V(v4i16|v4i32|v8i16)v$")>;
+def : InstRW<[FalkorWr_1VXVY_3cyc],   (instregex "^(FABD|FADD|FSUB)(v2f32|v4f16)$")>;
+def : InstRW<[FalkorWr_1VXVY_3cyc],   (instregex "^FADDP(v2i16p|v2i32p|v2i64p|v2f32|v4f16)$")>;
+
+def : InstRW<[FalkorWr_1VXVY_4cyc],   (instregex "^FCVT(N|M|P|Z|A)(S|U)(v1i32|v1i64|v1f16|v2f32|v4f16)$")>;
+def : InstRW<[FalkorWr_1VXVY_4cyc],   (instrs FCVTXNv1i64)>;
+def : InstRW<[FalkorWr_1VXVY_4cyc],   (instregex "^FCVTZ(S|U)(v2i32|v4i16)(_shift)?$")>;
+def : InstRW<[FalkorWr_1VXVY_4cyc],   (instregex "^ML(A|S)(v8i8|v4i16|v2i32)(_indexed)?$")>;
+
+def : InstRW<[FalkorWr_1VXVY_5cyc],   (instregex "^(FML(A|S)|FMUL|FMULX)(v2f32|v4f16|(v1i16_indexed|v4i16_indexed|v1i32_indexed|v2i32_indexed))$")>;
+def : InstRW<[FalkorWr_1VXVY_5cyc],   (instrs FMULX16, FMULX32)>;
+
+def : InstRW<[FalkorWr_1VXVY_6cyc],   (instregex "^(FML(A|S)|FMUL|FMULX)v1i64_indexed$")>;
+def : InstRW<[FalkorWr_1VXVY_6cyc],   (instrs FMULX64)>;
+
+def : InstRW<[FalkorWr_2VXVY_1cyc],   (instregex "^(FABS|FNEG)(v2f64|v4f32|v8f16)$")>;
+
+def : InstRW<[FalkorWr_2VXVY_2cyc],   (instregex "^(F(MAX|MIN)(NM)?P?|FAC(GE|GT)|FCM(EQ|GE|GT))(v2f64|v4f32|v8f16|v2i64p)$")>;
+def : InstRW<[FalkorWr_2VXVY_2cyc],   (instregex "^FCM(EQ|LE|GE|GT|LT)(v2i64|v4i32|v8i16)rz$")>;
+def : InstRW<[FalkorWr_2VXVY_2cyc],   (instregex "^FRINT(A|I|M|N|P|X|Z)(v2f64|v4f32|v8f16)$")>;
+
+def : InstRW<[FalkorWr_1VX_1VY_10cyc],(instregex "^(FDIV|FSQRT)(v2f32|v4f16)$")>;
+
+def : InstRW<[FalkorWr_2VXVY_3cyc],   (instregex "^(FABD|FADD(P)?|FSUB)(v2f64|v4f32|v8f16)$")>;
+
+def : InstRW<[FalkorWr_2VXVY_4cyc],   (instregex "^FCVT(N|M|P|Z|A)(S|U)(v2f64|v4f32|v8f16)$")>;
+def : InstRW<[FalkorWr_2VXVY_4cyc],   (instregex "^(FCVTL|FCVTL2)(v2i32|v4i16|v4i32|v8i16)$")>;
+def : InstRW<[FalkorWr_2VXVY_4cyc],   (instregex "^FCVTZ(S|U)(v2i64|v4i32|v8i16)(_shift)?$")>;
+def : InstRW<[FalkorWr_2VXVY_4cyc],   (instregex "^ML(A|S)(v16i8|v8i16|v4i32|v2i64)(_indexed)?$")>;
+
+def : InstRW<[FalkorWr_2VXVY_5cyc],   (instregex "^(FML(A|S)|FMUL|FMULX)(v2f64|v4f32|v8f16|v8i16_indexed|v4i32_indexed)$")>;
+
+def : InstRW<[FalkorWr_2VXVY_6cyc],   (instregex "^(FML(A|S)|FMUL|FMULX)v2i64_indexed$")>;
+
+def : InstRW<[FalkorWr_3VXVY_4cyc],   (instregex "^(FCVTX?N|FCVTX?N2)(v1i32|v1i64|v1f16|v2f32|v4f16)$")>;
+
+def : InstRW<[FalkorWr_3VXVY_5cyc],   (instregex "^(FCVTX?N|FCVTX?N2)(v2i32|v4i16|v8i16)$")>;
+
+def : InstRW<[FalkorWr_2VX_2VY_2cyc], (instregex "^(FDIV|FSQRT)(v2f64|v4f32|v8f16)$")>;
+
+// SIMD Load Instructions
+// -----------------------------------------------------------------------------
+def : InstRW<[WriteVLD],                (instregex "LD1i(8|16|32|64)$")>;
+def : InstRW<[WriteVLD],                (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[FalkorWr_1LD_1VXVY_4cyc], (instregex "LD1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[FalkorWr_2LD_3cyc],       (instregex "LD1Twov(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[FalkorWr_3LD_3cyc],       (instregex "LD1Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[FalkorWr_4LD_3cyc],       (instregex "LD1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVLD, WriteAdr],                (instregex "LD1i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr],                (instregex "LD1Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[FalkorWr_1LD_1VXVY_4cyc, WriteAdr], (instregex "LD1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[FalkorWr_2LD_3cyc, WriteAdr],       (instregex "LD1Twov(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[FalkorWr_3LD_3cyc, WriteAdr],       (instregex "LD1Threev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[FalkorWr_4LD_3cyc, WriteAdr],       (instregex "LD1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+
+def : InstRW<[WriteVLD], (instregex "LD2i(8|16|32|64)$")>;
+def : InstRW<[WriteVLD], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVLD], (instregex "LD2Twov(8b|4h|2s)$")>;
+def : InstRW<[WriteVLD], (instregex "LD2Twov(16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2i(8|16|32|64)(_POST)?$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2Rv(8b|4h|2s|1d|16b|8h|4s|2d)(_POST)?$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2Twov(8b|4h|2s)(_POST)?$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD2Twov(16b|8h|4s|2d)(_POST)?$")>;
+
+def : InstRW<[WriteVLD], (instregex "LD3i(8|16|32|64)$")>;
+def : InstRW<[WriteVLD], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVLD], (instregex "LD3Threev(8b|4h|2s|1d|16b|8h|4s)$")>;
+def : InstRW<[WriteVLD], (instregex "LD3Threev(2d)$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3Threev(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD3Threev(2d)_POST$")>;
+
+def : InstRW<[WriteVLD], (instregex "LD4i(8|16|32|64)$")>;
+def : InstRW<[WriteVLD], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVLD], (instregex "LD4Fourv(8b|4h|2s|1d|16b|8h|4s)$")>;
+def : InstRW<[WriteVLD], (instregex "LD4Fourv(2d)$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4Rv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4Fourv(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
+def : InstRW<[WriteVLD, WriteAdr], (instregex "LD4Fourv(2d)_POST$")>;
+
+// SIMD Store Instructions
+// -----------------------------------------------------------------------------
+def : InstRW<[WriteVST], (instregex "ST1i(8|16|32|64)$")>;
+def : InstRW<[WriteVST], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVST], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVST], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVST], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST1i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Onev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Twov(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Threev(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST1Fourv(8b|4h|2s|1d|16b|8h|4s|2d)_POST$")>;
+
+def : InstRW<[WriteVST], (instregex "ST2i(8|16|32|64)$")>;
+def : InstRW<[WriteVST], (instregex "ST2Twov(8b|4h|2s)$")>;
+def : InstRW<[WriteVST], (instregex "ST2Twov(16b|8h|4s|2d)$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST2i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST2Twov(8b|4h|2s)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST2Twov(16b|8h|4s|2d)_POST$")>;
+
+def : InstRW<[WriteVST], (instregex "ST3i(8|16|32|64)$")>;
+def : InstRW<[WriteVST], (instregex "ST3Threev(8b|4h|2s|1d|16b|8h|4s)$")>;
+def : InstRW<[WriteVST], (instregex "ST3Threev(2d)$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST3i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST3Threev(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST3Threev(2d)_POST$")>;
+
+def : InstRW<[WriteVST], (instregex "ST4i(8|16|32|64)$")>;
+def : InstRW<[WriteVST], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s)$")>;
+def : InstRW<[WriteVST], (instregex "ST4Fourv(2d)$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST4i(8|16|32|64)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST4Fourv(8b|4h|2s|1d|16b|8h|4s)_POST$")>;
+def : InstRW<[WriteVST, WriteAdr], (instregex "ST4Fourv(2d)_POST$")>;

diff --git a/llvm/lib/Target/AArch64/AArch64SchedFalkorWriteRes.td b/llvm/lib/Target/AArch64/AArch64SchedFalkorWriteRes.td
new file mode 100644 (file)
index 0000000..7ef3c67
--- /dev/null
+++ b/llvm/lib/Target/AArch64/AArch64SchedFalkorWriteRes.td
@@ -0,0 +1,268 @@
+//=- AArch64SchedFalkorWrRes.td - Falkor Write Res ---*- tablegen -*-=//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Contains all of the Falkor specific SchedWriteRes types. The approach
+// below is to define a generic SchedWriteRes for every combination of
+// latency and microOps. The naming conventions is to use a prefix, one field
+// for latency, and one or more microOp count/type designators.
+//   Prefix: FalkorWr
+//   MicroOp Count/Types: #(B|X|Y|Z|LD|ST|SD|VX|VY|VSD)
+//   Latency: #cyc
+//
+// e.g. FalkorWr_1Z_6SD_4VX_6cyc means there are 11 micro-ops to be issued
+//      down one Z pipe, six SD pipes, four VX pipes and the total latency is
+//      six cycles.
+//
+//===----------------------------------------------------------------------===//
+
+//===----------------------------------------------------------------------===//
+// Define 1 micro-op types
+
+
+def FalkorWr_1X_2cyc    : SchedWriteRes<[FalkorUnitX]>   { let Latency = 2; }
+def FalkorWr_1X_4cyc    : SchedWriteRes<[FalkorUnitX]>   { let Latency = 4; }
+def FalkorWr_1X_5cyc    : SchedWriteRes<[FalkorUnitX]>   { let Latency = 5; }
+def FalkorWr_1Z_1cyc    : SchedWriteRes<[FalkorUnitZ]>   { let Latency = 1; }
+def FalkorWr_1Z_2cyc    : SchedWriteRes<[FalkorUnitZ]>   { let Latency = 2; }
+def FalkorWr_1ZB_1cyc   : SchedWriteRes<[FalkorUnitZB]>  { let Latency = 1; }
+def FalkorWr_1LD_3cyc   : SchedWriteRes<[FalkorUnitLD]>  { let Latency = 3; }
+def FalkorWr_1LD_4cyc   : SchedWriteRes<[FalkorUnitLD]>  { let Latency = 4; }
+def FalkorWr_1XYZ_1cyc  : SchedWriteRes<[FalkorUnitXYZ]> { let Latency = 1; }
+def FalkorWr_1XYZ_2cyc  : SchedWriteRes<[FalkorUnitXYZ]> { let Latency = 2; }
+def FalkorWr_1XYZB_1cyc : SchedWriteRes<[FalkorUnitXYZB]>{ let Latency = 1; }
+
+def FalkorWr_1VXVY_1cyc : SchedWriteRes<[FalkorUnitVXVY]>{ let Latency = 1; }
+def FalkorWr_1VXVY_2cyc : SchedWriteRes<[FalkorUnitVXVY]>{ let Latency = 2; }
+def FalkorWr_1VXVY_3cyc : SchedWriteRes<[FalkorUnitVXVY]>{ let Latency = 3; }
+def FalkorWr_1VXVY_4cyc : SchedWriteRes<[FalkorUnitVXVY]>{ let Latency = 4; }
+def FalkorWr_1VXVY_5cyc : SchedWriteRes<[FalkorUnitVXVY]>{ let Latency = 5; }
+def FalkorWr_1VXVY_6cyc : SchedWriteRes<[FalkorUnitVXVY]>{ let Latency = 6; }
+
+def FalkorWr_1LD_1cyc   : SchedWriteRes<[FalkorUnitLD]>  { let Latency = 1; }
+def FalkorWr_1ST_1cyc   : SchedWriteRes<[FalkorUnitST]>  { let Latency = 1; }
+
+//===----------------------------------------------------------------------===//
+// Define 2 micro-op types
+
+def FalkorWr_2VXVY_1cyc   : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 1;
+  let NumMicroOps = 2;
+}
+def FalkorWr_2VXVY_2cyc   : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+def FalkorWr_2VXVY_3cyc   : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 3;
+  let NumMicroOps = 2;
+}
+def FalkorWr_2VXVY_4cyc   : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 2;
+}
+def FalkorWr_2VXVY_5cyc   : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 5;
+  let NumMicroOps = 2;
+}
+def FalkorWr_2VXVY_6cyc   : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 6;
+  let NumMicroOps = 2;
+}
+
+def FalkorWr_1LD_1VXVY_4cyc : SchedWriteRes<[FalkorUnitLD, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 2;
+}
+def FalkorWr_1XYZ_1LD_4cyc  : SchedWriteRes<[FalkorUnitXYZ, FalkorUnitLD]> {
+  let Latency = 4;
+  let NumMicroOps = 2;
+}
+def FalkorWr_2LD_3cyc   : SchedWriteRes<[FalkorUnitLD, FalkorUnitLD]> {
+  let Latency = 3;
+  let NumMicroOps = 2;
+}
+
+
+def FalkorWr_1VY_1VX_3cyc : SchedWriteRes<[FalkorUnitVY, FalkorUnitVX]> {
+  let Latency = 3;
+  let NumMicroOps = 2;
+}
+def FalkorWr_1VY_1VX_5cyc : SchedWriteRes<[FalkorUnitVY, FalkorUnitVX]> {
+  let Latency = 5;
+  let NumMicroOps = 2;
+}
+
+def FalkorWr_1VX_1VY_2cyc : SchedWriteRes<[FalkorUnitVX, FalkorUnitVY]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+
+def FalkorWr_1VX_1VY_10cyc : SchedWriteRes<[FalkorUnitVX, FalkorUnitVY]> {
+  let Latency = 10;
+  let NumMicroOps = 2;
+}
+
+def FalkorWr_1XYZ_1ST_4cyc: SchedWriteRes<[FalkorUnitXYZ, FalkorUnitST]> {
+  let Latency = 4;
+  let NumMicroOps = 2;
+}
+def FalkorWr_1XYZ_1LD_5cyc: SchedWriteRes<[FalkorUnitXYZ, FalkorUnitLD]> {
+  let Latency = 5;
+  let NumMicroOps = 2;
+}
+
+def FalkorWr_2XYZ_2cyc   : SchedWriteRes<[FalkorUnitXYZ, FalkorUnitXYZ]> {
+  let Latency = 2;
+  let NumMicroOps = 2;
+}
+
+def FalkorWr_1LD_1Z_3cyc : SchedWriteRes<[FalkorUnitLD, FalkorUnitZ]> {
+  let Latency = 3;
+  let NumMicroOps = 2;
+}
+
+//===----------------------------------------------------------------------===//
+// Define 3 micro-op types
+
+def FalkorWr_3VXVY_3cyc : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 3;
+  let NumMicroOps = 3;
+}
+def FalkorWr_3VXVY_4cyc : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 3;
+}
+def FalkorWr_3VXVY_5cyc : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 5;
+  let NumMicroOps = 3;
+}
+def FalkorWr_3VXVY_6cyc : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 6;
+  let NumMicroOps = 3;
+}
+
+def FalkorWr_1LD_2VXVY_4cyc  : SchedWriteRes<[FalkorUnitLD, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 3;
+}
+def FalkorWr_2LD_1none_3cyc  : SchedWriteRes<[FalkorUnitLD, FalkorUnitLD]> {
+  let Latency = 3;
+  let NumMicroOps = 3;
+}
+def FalkorWr_3LD_3cyc        : SchedWriteRes<[FalkorUnitLD, FalkorUnitLD,
+                                              FalkorUnitLD]> {
+  let Latency = 3;
+  let NumMicroOps = 3;
+}
+
+def FalkorWr_2LD_1Z_3cyc     : SchedWriteRes<[FalkorUnitLD, FalkorUnitLD,
+                                             FalkorUnitZ]> {
+  let Latency = 3;
+  let NumMicroOps = 3;
+}
+def FalkorWr_1LD_1ST_1SD_3cyc: SchedWriteRes<[FalkorUnitLD, FalkorUnitST,
+                                              FalkorUnitSD]> {
+  let Latency = 3;
+  let NumMicroOps = 3;
+}
+
+//===----------------------------------------------------------------------===//
+// Define 4 micro-op types
+
+def FalkorWr_2VX_2VY_2cyc  : SchedWriteRes<[FalkorUnitVX, FalkorUnitVY,
+                                            FalkorUnitVX, FalkorUnitVY]> {
+  let Latency = 2;
+  let NumMicroOps = 4;
+}
+
+def FalkorWr_4VXVY_2cyc    : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 2;
+  let NumMicroOps = 4;
+}
+def FalkorWr_4VXVY_3cyc    : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 3;
+  let NumMicroOps = 4;
+}
+def FalkorWr_4VXVY_4cyc    : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 4;
+}
+def FalkorWr_4VXVY_6cyc    : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 6;
+  let NumMicroOps = 4;
+}
+
+def FalkorWr_4LD_3cyc      : SchedWriteRes<[FalkorUnitLD, FalkorUnitLD,
+                                            FalkorUnitLD, FalkorUnitLD]> {
+  let Latency = 3;
+  let NumMicroOps = 4;
+}
+
+def FalkorWr_1LD_1ST_1SD_1LD_3cyc: SchedWriteRes<[FalkorUnitLD, FalkorUnitST,
+                                                  FalkorUnitSD, FalkorUnitLD]> {
+  let Latency = 3;
+  let NumMicroOps = 4;
+}
+
+//===----------------------------------------------------------------------===//
+// Define 5 micro-op types
+
+def FalkorWr_1LD_4VXVY_4cyc: SchedWriteRes<[FalkorUnitLD, FalkorUnitVXVY,
+                                            FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 5;
+}
+def FalkorWr_2LD_2VXVY_4cyc: SchedWriteRes<[FalkorUnitLD, FalkorUnitLD,
+                                            FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 5;
+}
+def FalkorWr_5VXVY_7cyc    : SchedWriteRes<[FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY, FalkorUnitVXVY,
+                                            FalkorUnitVXVY]> {
+  let Latency = 7;
+  let NumMicroOps = 5;
+}
+
+//===----------------------------------------------------------------------===//
+// Define 6 micro-op types
+
+def FalkorWr_2LD_2VXVY_2none_4cyc: SchedWriteRes<[FalkorUnitLD, FalkorUnitLD,
+                                            FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 6;
+}
+
+//===----------------------------------------------------------------------===//
+// Define 8 micro-op types
+
+def FalkorWr_2LD_2VXVY_2LD_2VXVY_4cyc:SchedWriteRes<[FalkorUnitLD, FalkorUnitLD,
+                                             FalkorUnitVXVY, FalkorUnitVXVY,
+                                             FalkorUnitLD, FalkorUnitLD,
+                                             FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 8;
+}
+
+//===----------------------------------------------------------------------===//
+// Define 9 micro-op types
+
+def FalkorWr_2LD_2VXVY_1XYZ_2LD_2VXVY_4cyc:SchedWriteRes<[FalkorUnitLD,
+                                             FalkorUnitLD, FalkorUnitVXVY,
+                                             FalkorUnitVXVY, FalkorUnitXYZ,
+                                             FalkorUnitLD, FalkorUnitLD,
+                                             FalkorUnitVXVY, FalkorUnitVXVY]> {
+  let Latency = 4;
+  let NumMicroOps = 9;
+}