Add ARMv6 itineraries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89218 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 201 insertions, 13 deletions

diff --git a/lib/Target/ARM/ARM.td b/lib/Target/ARM/ARM.td
index cb9bd6a..7033861 100644
--- a/lib/Target/ARM/ARM.td
+++ b/lib/Target/ARM/ARM.td
@@ -89,16 +89,18 @@ def : ProcNoItin<"xscale",          [ArchV5TE]>;
 def : ProcNoItin<"iwmmxt",          [ArchV5TE]>;
 
 // V6 Processors.
-def : ProcNoItin<"arm1136j-s",      [ArchV6]>;
-def : ProcNoItin<"arm1136jf-s",     [ArchV6, FeatureVFP2]>;
-def : ProcNoItin<"arm1176jz-s",     [ArchV6]>;
-def : ProcNoItin<"arm1176jzf-s",    [ArchV6, FeatureVFP2]>;
-def : ProcNoItin<"mpcorenovfp",     [ArchV6]>;
-def : ProcNoItin<"mpcore",          [ArchV6, FeatureVFP2]>;
+def : Processor<"arm1136j-s",       ARMV6Itineraries, [ArchV6]>;
+def : Processor<"arm1136jf-s",      ARMV6Itineraries, [ArchV6, FeatureVFP2]>;
+def : Processor<"arm1176jz-s",      ARMV6Itineraries, [ArchV6]>;
+def : Processor<"arm1176jzf-s",     ARMV6Itineraries, [ArchV6, FeatureVFP2]>;
+def : Processor<"mpcorenovfp",      ARMV6Itineraries, [ArchV6]>;
+def : Processor<"mpcore",           ARMV6Itineraries, [ArchV6, FeatureVFP2]>;
 
 // V6T2 Processors.
-def : ProcNoItin<"arm1156t2-s",     [ArchV6T2, FeatureThumb2]>;
-def : ProcNoItin<"arm1156t2f-s",    [ArchV6T2, FeatureThumb2, FeatureVFP2]>;
+def : Processor<"arm1156t2-s",     ARMV6Itineraries,
+                 [ArchV6T2, FeatureThumb2]>;
+def : Processor<"arm1156t2f-s",    ARMV6Itineraries,
+                 [ArchV6T2, FeatureThumb2, FeatureVFP2]>;
 
 // V7 Processors.
 def : Processor<"cortex-a8",        CortexA8Itineraries,
diff --git a/lib/Target/ARM/ARMScheduleV6.td b/lib/Target/ARM/ARMScheduleV6.td
index 1ace718..0fef466 100644
--- a/lib/Target/ARM/ARMScheduleV6.td
+++ b/lib/Target/ARM/ARMScheduleV6.td
@@ -11,4 +11,190 @@
 //
 //===----------------------------------------------------------------------===//
 
-// TODO: Add model for an ARM11
+// Model based on ARM1176
+//
+// Scheduling information derived from "ARM1176JZF-S Technical Reference Manual".
+//
+def ARMV6Itineraries : ProcessorItineraries<[
+  //
+  // No operand cycles
+  InstrItinData<IIC_iALUx    , [InstrStage<1, [FU_Pipe0]>]>,
+  //
+  // Binary Instructions that produce a result
+  InstrItinData<IIC_iALUi    , [InstrStage<1, [FU_Pipe0]>], [2, 2]>,
+  InstrItinData<IIC_iALUr    , [InstrStage<1, [FU_Pipe0]>], [2, 2, 2]>,
+  InstrItinData<IIC_iALUsi   , [InstrStage<1, [FU_Pipe0]>], [2, 2, 1]>,
+  InstrItinData<IIC_iALUsr   , [InstrStage<2, [FU_Pipe0]>], [3, 3, 2, 1]>,
+  //
+  // Unary Instructions that produce a result
+  InstrItinData<IIC_iUNAr    , [InstrStage<1, [FU_Pipe0]>], [2, 2]>,
+  InstrItinData<IIC_iUNAsi   , [InstrStage<1, [FU_Pipe0]>], [2, 1]>,
+  InstrItinData<IIC_iUNAsr   , [InstrStage<2, [FU_Pipe0]>], [3, 2, 1]>,
+  //
+  // Compare instructions
+  InstrItinData<IIC_iCMPi    , [InstrStage<1, [FU_Pipe0]>], [2]>,
+  InstrItinData<IIC_iCMPr    , [InstrStage<1, [FU_Pipe0]>], [2, 2]>,
+  InstrItinData<IIC_iCMPsi   , [InstrStage<1, [FU_Pipe0]>], [2, 1]>,
+  InstrItinData<IIC_iCMPsr   , [InstrStage<2, [FU_Pipe0]>], [3, 2, 1]>,
+  //
+  // Move instructions, unconditional
+  InstrItinData<IIC_iMOVi    , [InstrStage<1, [FU_Pipe0]>], [2]>,
+  InstrItinData<IIC_iMOVr    , [InstrStage<1, [FU_Pipe0]>], [2, 2]>,
+  InstrItinData<IIC_iMOVsi   , [InstrStage<1, [FU_Pipe0]>], [2, 1]>,
+  InstrItinData<IIC_iMOVsr   , [InstrStage<2, [FU_Pipe0]>], [3, 2, 1]>,
+  //
+  // Move instructions, conditional
+  InstrItinData<IIC_iCMOVi   , [InstrStage<1, [FU_Pipe0]>], [3]>,
+  InstrItinData<IIC_iCMOVr   , [InstrStage<1, [FU_Pipe0]>], [3, 2]>,
+  InstrItinData<IIC_iCMOVsi  , [InstrStage<1, [FU_Pipe0]>], [3, 1]>,
+  InstrItinData<IIC_iCMOVsr  , [InstrStage<1, [FU_Pipe0]>], [4, 2, 1]>,
+
+  // Integer multiply pipeline
+  //
+  InstrItinData<IIC_iMUL16   , [InstrStage<1, [FU_Pipe0]>], [4, 1, 1]>,
+  InstrItinData<IIC_iMAC16   , [InstrStage<1, [FU_Pipe0]>], [4, 1, 1, 2]>,
+  InstrItinData<IIC_iMUL32   , [InstrStage<2, [FU_Pipe0]>], [5, 1, 1]>,
+  InstrItinData<IIC_iMAC32   , [InstrStage<2, [FU_Pipe0]>], [5, 1, 1, 2]>,
+  InstrItinData<IIC_iMUL64   , [InstrStage<3, [FU_Pipe0]>], [6, 1, 1]>,
+  InstrItinData<IIC_iMAC64   , [InstrStage<3, [FU_Pipe0]>], [6, 1, 1, 2]>,
+  
+  // Integer load pipeline
+  //
+  // Immediate offset
+  InstrItinData<IIC_iLoadi   , [InstrStage<1, [FU_Pipe0]>], [4, 1]>,
+  //
+  // Register offset
+  InstrItinData<IIC_iLoadr   , [InstrStage<1, [FU_Pipe0]>], [4, 1, 1]>,
+  //
+  // Scaled register offset, issues over 2 cycles
+  InstrItinData<IIC_iLoadsi  , [InstrStage<2, [FU_Pipe0]>], [5, 2, 1]>,
+  //
+  // Immediate offset with update
+  InstrItinData<IIC_iLoadiu  , [InstrStage<1, [FU_Pipe0]>], [4, 2, 1]>,
+  //
+  // Register offset with update
+  InstrItinData<IIC_iLoadru  , [InstrStage<1, [FU_Pipe0]>], [4, 2, 1, 1]>,
+  //
+  // Scaled register offset with update, issues over 2 cycles
+  InstrItinData<IIC_iLoadsiu , [InstrStage<2, [FU_Pipe0]>], [5, 2, 2, 1]>,
+
+  //
+  // Load multiple
+  InstrItinData<IIC_iLoadm   , [InstrStage<3, [FU_Pipe0]>]>,
+
+  // Integer store pipeline
+  //
+  // Immediate offset
+  InstrItinData<IIC_iStorei  , [InstrStage<1, [FU_Pipe0]>], [2, 1]>,
+  //
+  // Register offset
+  InstrItinData<IIC_iStorer  , [InstrStage<1, [FU_Pipe0]>], [2, 1, 1]>,
+
+  //
+  // Scaled register offset, issues over 2 cycles
+  InstrItinData<IIC_iStoresi , [InstrStage<2, [FU_Pipe0]>], [2, 2, 1]>,
+  //
+  // Immediate offset with update
+  InstrItinData<IIC_iStoreiu , [InstrStage<1, [FU_Pipe0]>], [2, 2, 1]>,
+  //
+  // Register offset with update
+  InstrItinData<IIC_iStoreru , [InstrStage<1, [FU_Pipe0]>], [2, 2, 1, 1]>,
+  //
+  // Scaled register offset with update, issues over 2 cycles
+  InstrItinData<IIC_iStoresiu, [InstrStage<2, [FU_Pipe0]>], [2, 2, 2, 1]>,
+  //
+  // Store multiple
+  InstrItinData<IIC_iStorem   , [InstrStage<3, [FU_Pipe0]>]>,
+  
+  // Branch
+  //
+  // no delay slots, so the latency of a branch is unimportant
+  InstrItinData<IIC_Br      , [InstrStage<1, [FU_Pipe0]>]>,
+
+  // VFP
+  // Issue through integer pipeline, and execute in NEON unit. We assume
+  // RunFast mode so that NFP pipeline is used for single-precision when
+  // possible.
+  //
+  // FP Special Register to Integer Register File Move
+  InstrItinData<IIC_fpSTAT , [InstrStage<1, [FU_Pipe0]>], [3]>,
+  //
+  // Single-precision FP Unary
+  InstrItinData<IIC_fpUNA32 , [InstrStage<1, [FU_Pipe0]>], [5, 2]>,
+  //
+  // Double-precision FP Unary
+  InstrItinData<IIC_fpUNA64 , [InstrStage<1, [FU_Pipe0]>], [5, 2]>,
+  //
+  // Single-precision FP Compare
+  InstrItinData<IIC_fpCMP32 , [InstrStage<1, [FU_Pipe0]>], [2, 2]>,
+  //
+  // Double-precision FP Compare
+  InstrItinData<IIC_fpCMP64 , [InstrStage<1, [FU_Pipe0]>], [2, 2]>,
+  //
+  // Single to Double FP Convert
+  InstrItinData<IIC_fpCVTSD , [InstrStage<1, [FU_Pipe0]>], [5, 2]>,
+  //
+  // Double to Single FP Convert
+  InstrItinData<IIC_fpCVTDS , [InstrStage<1, [FU_Pipe0]>], [5, 2]>,
+  //
+  // Single-Precision FP to Integer Convert
+  InstrItinData<IIC_fpCVTSI , [InstrStage<1, [FU_Pipe0]>], [9, 2]>,
+  //
+  // Double-Precision FP to Integer Convert
+  InstrItinData<IIC_fpCVTDI , [InstrStage<1, [FU_Pipe0]>], [9, 2]>,
+  //
+  // Integer to Single-Precision FP Convert
+  InstrItinData<IIC_fpCVTIS , [InstrStage<1, [FU_Pipe0]>], [9, 2]>,
+  //
+  // Integer to Double-Precision FP Convert
+  InstrItinData<IIC_fpCVTID , [InstrStage<1, [FU_Pipe0]>], [9, 2]>,
+  //
+  // Single-precision FP ALU
+  InstrItinData<IIC_fpALU32 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2]>,
+  //
+  // Double-precision FP ALU
+  InstrItinData<IIC_fpALU64 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2]>,
+  //
+  // Single-precision FP Multiply
+  InstrItinData<IIC_fpMUL32 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2]>,
+  //
+  // Double-precision FP Multiply
+  InstrItinData<IIC_fpMUL64 , [InstrStage<2, [FU_Pipe0]>], [9, 2, 2]>,
+  //
+  // Single-precision FP MAC
+  InstrItinData<IIC_fpMAC32 , [InstrStage<1, [FU_Pipe0]>], [9, 2, 2, 2]>,
+  //
+  // Double-precision FP MAC
+  InstrItinData<IIC_fpMAC64 , [InstrStage<2, [FU_Pipe0]>], [9, 2, 2, 2]>,
+  //
+  // Single-precision FP DIV
+  InstrItinData<IIC_fpDIV32 , [InstrStage<15, [FU_Pipe0]>], [20, 2, 2]>,
+  //
+  // Double-precision FP DIV
+  InstrItinData<IIC_fpDIV64 , [InstrStage<29, [FU_Pipe0]>], [34, 2, 2]>,
+  //
+  // Single-precision FP SQRT
+  InstrItinData<IIC_fpSQRT32 , [InstrStage<15, [FU_Pipe0]>], [20, 2, 2]>,
+  //
+  // Double-precision FP SQRT
+  InstrItinData<IIC_fpSQRT64 , [InstrStage<29, [FU_Pipe0]>], [34, 2, 2]>,
+  //
+  // Single-precision FP Load
+  InstrItinData<IIC_fpLoad32 , [InstrStage<1, [FU_Pipe0]>], [5, 2, 2]>,
+  //
+  // Double-precision FP Load
+  InstrItinData<IIC_fpLoad64 , [InstrStage<1, [FU_Pipe0]>], [5, 2, 2]>,
+  //
+  // FP Load Multiple
+  InstrItinData<IIC_fpLoadm , [InstrStage<3, [FU_Pipe0]>]>,
+  //
+  // Single-precision FP Store
+  InstrItinData<IIC_fpStore32 , [InstrStage<1, [FU_Pipe0]>], [2, 2, 2]>,
+  //
+  // Double-precision FP Store
+  // use FU_Issue to enforce the 1 load/store per cycle limit
+  InstrItinData<IIC_fpStore64 , [InstrStage<1, [FU_Pipe0]>], [2, 2, 2]>,
+  //
+  // FP Store Multiple
+  InstrItinData<IIC_fpStorem , [InstrStage<3, [FU_Pipe0]>]>
+]>;
diff --git a/lib/Target/ARM/ARMScheduleV7.td b/lib/Target/ARM/ARMScheduleV7.td
index e565813..427645c 100644
--- a/lib/Target/ARM/ARMScheduleV7.td
+++ b/lib/Target/ARM/ARMScheduleV7.td
@@ -184,7 +184,7 @@ def CortexA8Itineraries : ProcessorItineraries<[
   //
   // Single-precision FP Compare
   InstrItinData<IIC_fpCMP32 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
-                               InstrStage<1, [FU_NPipe]>], [7, 1]>,
+                               InstrStage<1, [FU_NPipe]>], [1, 1]>,
   //
   // Double-precision FP Compare
   InstrItinData<IIC_fpCMP64 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
@@ -221,7 +221,7 @@ def CortexA8Itineraries : ProcessorItineraries<[
   //
   // Single-precision FP ALU
   InstrItinData<IIC_fpALU32 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
-                               InstrStage<1, [FU_NPipe]>], [7, 1]>,
+                               InstrStage<1, [FU_NPipe]>], [7, 1, 1]>,
   //
   // Double-precision FP ALU
   InstrItinData<IIC_fpALU64 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
@@ -230,7 +230,7 @@ def CortexA8Itineraries : ProcessorItineraries<[
   //
   // Single-precision FP Multiply
   InstrItinData<IIC_fpMUL32 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
-                               InstrStage<1, [FU_NPipe]>], [7, 1]>,
+                               InstrStage<1, [FU_NPipe]>], [7, 1, 1]>,
   //
   // Double-precision FP Multiply
   InstrItinData<IIC_fpMUL64 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
@@ -239,7 +239,7 @@ def CortexA8Itineraries : ProcessorItineraries<[
   //
   // Single-precision FP MAC
   InstrItinData<IIC_fpMAC32 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,
-                               InstrStage<1, [FU_NPipe]>], [7, 1]>,
+                               InstrStage<1, [FU_NPipe]>], [7, 2, 1, 1]>,
   //
   // Double-precision FP MAC
   InstrItinData<IIC_fpMAC64 , [InstrStage<1, [FU_Pipe0, FU_Pipe1]>,