Finish vld3 and vld4.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116140 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 78 insertions, 34 deletions

diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 24563a1a5e..4340a7d3c7 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -344,7 +344,7 @@ def  VLD3d32Pseudo : VLDQQPseudo<IIC_VLD3>;
 class VLD3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
   : NLdSt<0, 0b10, op11_8, op7_4,
           (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
-          (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3,
+          (ins addrmode6:$addr, am6offset:$offset), IIC_VLD3u,
           "vld3", Dt, "\\{$dst1, $dst2, $dst3\\}, $addr$offset",
           "$addr.addr = $wb", []>;
 
@@ -352,9 +352,9 @@ def VLD3d8_UPD  : VLD3DWB<0b0100, 0b0000, "8">;
 def VLD3d16_UPD : VLD3DWB<0b0100, 0b0100, "16">;
 def VLD3d32_UPD : VLD3DWB<0b0100, 0b1000, "32">;
 
-def VLD3d8Pseudo_UPD  : VLDQQWBPseudo<IIC_VLD3>;
-def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3>;
-def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3>;
+def VLD3d8Pseudo_UPD  : VLDQQWBPseudo<IIC_VLD3u>;
+def VLD3d16Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
+def VLD3d32Pseudo_UPD : VLDQQWBPseudo<IIC_VLD3u>;
 
 // ...with double-spaced registers (non-updating versions for disassembly only):
 def VLD3q8      : VLD3D<0b0101, 0b0000, "8">;
@@ -364,14 +364,14 @@ def VLD3q8_UPD  : VLD3DWB<0b0101, 0b0000, "8">;
 def VLD3q16_UPD : VLD3DWB<0b0101, 0b0100, "16">;
 def VLD3q32_UPD : VLD3DWB<0b0101, 0b1000, "32">;
 
-def VLD3q8Pseudo_UPD  : VLDQQQQWBPseudo<IIC_VLD3>;
-def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
-def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
+def VLD3q8Pseudo_UPD  : VLDQQQQWBPseudo<IIC_VLD3u>;
+def VLD3q16Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
+def VLD3q32Pseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
 
 // ...alternate versions to be allocated odd register numbers:
-def VLD3q8oddPseudo_UPD  : VLDQQQQWBPseudo<IIC_VLD3>;
-def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
-def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3>;
+def VLD3q8oddPseudo_UPD  : VLDQQQQWBPseudo<IIC_VLD3u>;
+def VLD3q16oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
+def VLD3q32oddPseudo_UPD : VLDQQQQWBPseudo<IIC_VLD3u>;
 
 //   VLD4     : Vector Load (multiple 4-element structures)
 class VLD4D<bits<4> op11_8, bits<4> op7_4, string Dt>
@@ -499,7 +499,7 @@ def VLD2LNq32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD2lnu>;
 class VLD3LN<bits<4> op11_8, bits<4> op7_4, string Dt>
   : NLdSt<1, 0b10, op11_8, op7_4, (outs DPR:$dst1, DPR:$dst2, DPR:$dst3),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3,
-          nohash_imm:$lane), IIC_VLD3, "vld3", Dt,
+          nohash_imm:$lane), IIC_VLD3ln, "vld3", Dt,
           "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr",
           "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3", []>;
 
@@ -507,16 +507,16 @@ def VLD3LNd8  : VLD3LN<0b0010, {?,?,?,0}, "8">;
 def VLD3LNd16 : VLD3LN<0b0110, {?,?,0,0}, "16">;
 def VLD3LNd32 : VLD3LN<0b1010, {?,0,0,0}, "32">;
 
-def VLD3LNd8Pseudo  : VLDQQLNPseudo<IIC_VLD3>;
-def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3>;
-def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3>;
+def VLD3LNd8Pseudo  : VLDQQLNPseudo<IIC_VLD3ln>;
+def VLD3LNd16Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
+def VLD3LNd32Pseudo : VLDQQLNPseudo<IIC_VLD3ln>;
 
 // ...with double-spaced registers:
 def VLD3LNq16 : VLD3LN<0b0110, {?,?,1,0}, "16">;
 def VLD3LNq32 : VLD3LN<0b1010, {?,1,0,0}, "32">;
 
-def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3>;
-def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3>;
+def VLD3LNq16Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
+def VLD3LNq32Pseudo : VLDQQQQLNPseudo<IIC_VLD3ln>;
 
 // ...with address register writeback:
 class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
@@ -524,7 +524,7 @@ class VLD3LNWB<bits<4> op11_8, bits<4> op7_4, string Dt>
           (outs DPR:$dst1, DPR:$dst2, DPR:$dst3, GPR:$wb),
           (ins addrmode6:$addr, am6offset:$offset,
            DPR:$src1, DPR:$src2, DPR:$src3, nohash_imm:$lane),
-          IIC_VLD3, "vld3", Dt,
+          IIC_VLD3lnu, "vld3", Dt,
           "\\{$dst1[$lane], $dst2[$lane], $dst3[$lane]\\}, $addr$offset",
           "$src1 = $dst1, $src2 = $dst2, $src3 = $dst3, $addr.addr = $wb",
           []>;
@@ -533,15 +533,15 @@ def VLD3LNd8_UPD  : VLD3LNWB<0b0010, {?,?,?,0}, "8">;
 def VLD3LNd16_UPD : VLD3LNWB<0b0110, {?,?,0,0}, "16">;
 def VLD3LNd32_UPD : VLD3LNWB<0b1010, {?,0,0,0}, "32">;
 
-def VLD3LNd8Pseudo_UPD  : VLDQQLNWBPseudo<IIC_VLD3>;
-def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
-def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3>;
+def VLD3LNd8Pseudo_UPD  : VLDQQLNWBPseudo<IIC_VLD3lnu>;
+def VLD3LNd16Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
+def VLD3LNd32Pseudo_UPD : VLDQQLNWBPseudo<IIC_VLD3lnu>;
 
 def VLD3LNq16_UPD : VLD3LNWB<0b0110, {?,?,1,0}, "16">;
 def VLD3LNq32_UPD : VLD3LNWB<0b1010, {?,1,0,0}, "32">;
 
-def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3>;
-def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3>;
+def VLD3LNq16Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
+def VLD3LNq32Pseudo_UPD : VLDQQQQLNWBPseudo<IIC_VLD3lnu>;
 
 //   VLD4LN   : Vector Load (single 4-element structure to one lane)
 class VLD4LN<bits<4> op11_8, bits<4> op7_4, string Dt>
diff --git a/lib/Target/ARM/ARMSchedule.td b/lib/Target/ARM/ARMSchedule.td
index 521faa196d..73c677e0d4 100644
--- a/lib/Target/ARM/ARMSchedule.td
+++ b/lib/Target/ARM/ARMSchedule.td
@@ -141,6 +141,9 @@ def IIC_VLD2x2u    : InstrItinClass;
 def IIC_VLD2ln     : InstrItinClass;
 def IIC_VLD2lnu    : InstrItinClass;
 def IIC_VLD3       : InstrItinClass;
+def IIC_VLD3ln     : InstrItinClass;
+def IIC_VLD3u      : InstrItinClass;
+def IIC_VLD3lnu    : InstrItinClass;
 def IIC_VLD4       : InstrItinClass;
 def IIC_VST        : InstrItinClass;
 def IIC_VUNAD      : InstrItinClass;
diff --git a/lib/Target/ARM/ARMScheduleA8.td b/lib/Target/ARM/ARMScheduleA8.td
index e711839845..6c4cf8f122 100644
--- a/lib/Target/ARM/ARMScheduleA8.td
+++ b/lib/Target/ARM/ARMScheduleA8.td
@@ -441,39 +441,58 @@ def CortexA8Itineraries : ProcessorItineraries<
                               [2, 2, 1]>,
   //
   // VLD2x2
-  InstrItinData<IIC_VLD2,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+  InstrItinData<IIC_VLD2x2,   [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
                                InstrStage<3, [A8_NLSPipe], 1>,
                                InstrStage<3, [A8_LSPipe]>],
                               [2, 2, 3, 3, 1]>,
   //
   // VLD2ln
-  InstrItinData<IIC_VLD2,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+  InstrItinData<IIC_VLD2ln,   [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
                                InstrStage<3, [A8_NLSPipe], 1>,
                                InstrStage<3, [A8_LSPipe]>],
                               [3, 3, 1, 1, 1, 1]>,
   //
   // VLD2u
-  InstrItinData<IIC_VLD2,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+  InstrItinData<IIC_VLD2u,    [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
                                InstrStage<1, [A8_NLSPipe], 1>,
                                InstrStage<1, [A8_LSPipe]>],
                               [2, 2, 2, 1, 1, 1]>,
   //
   // VLD2x2u
-  InstrItinData<IIC_VLD2,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+  InstrItinData<IIC_VLD2x2u,  [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
                                InstrStage<3, [A8_NLSPipe], 1>,
                                InstrStage<3, [A8_LSPipe]>],
                               [2, 2, 3, 3, 2, 1]>,
   //
   // VLD2lnu
-  InstrItinData<IIC_VLD2,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+  InstrItinData<IIC_VLD2lnu,  [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
                                InstrStage<3, [A8_NLSPipe], 1>,
                                InstrStage<3, [A8_LSPipe]>],
                               [3, 3, 2, 1, 1, 1, 1, 1]>,
   //
   // VLD3
   InstrItinData<IIC_VLD3,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
-                               InstrStage<1, [A8_NLSPipe]>,
-                               InstrStage<1, [A8_LSPipe]>], [2, 2, 2, 1]>,
+                               InstrStage<4, [A8_NLSPipe], 1>,
+                               InstrStage<4, [A8_LSPipe]>],
+                              [3, 3, 4, 1]>,
+  //
+  // VLD3ln
+  InstrItinData<IIC_VLD3ln,   [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+                               InstrStage<5, [A8_NLSPipe], 1>,
+                               InstrStage<5, [A8_LSPipe]>],
+                              [4, 4, 5, 1, 1, 1, 1, 2]>,
+  //
+  // VLD3u
+  InstrItinData<IIC_VLD3u,    [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+                               InstrStage<4, [A8_NLSPipe], 1>,
+                               InstrStage<4, [A8_LSPipe]>],
+                              [3, 3, 4, 2, 1]>,
+  //
+  // VLD3lnu
+  InstrItinData<IIC_VLD3lnu,  [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
+                               InstrStage<5, [A8_NLSPipe], 1>,
+                               InstrStage<5, [A8_LSPipe]>],
+                              [4, 4, 5, 2, 1, 1, 1, 1, 1, 2]>,
   //
   // VLD4
   InstrItinData<IIC_VLD4,     [InstrStage<1, [A8_Pipe0, A8_Pipe1]>,
diff --git a/lib/Target/ARM/ARMScheduleA9.td b/lib/Target/ARM/ARMScheduleA9.td
index c199ef7f2b..2d2bc370f5 100644
--- a/lib/Target/ARM/ARMScheduleA9.td
+++ b/lib/Target/ARM/ARMScheduleA9.td
@@ -822,14 +822,36 @@ def CortexA9Itineraries : ProcessorItineraries<
                               [4, 4, 2, 1, 1, 1, 1, 1]>,
   //
   // VLD3
-  // FIXME: We don't model this instruction properly
   InstrItinData<IIC_VLD3,     [InstrStage<1, [A9_DRegsN],   0, Required>,
-                               // Extra latency cycles since wbck is 6 cycles
-                               InstrStage<7, [A9_DRegsVFP], 0, Reserved>,
+                               InstrStage<10, [A9_DRegsVFP], 0, Reserved>,
                                InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
                                InstrStage<1, [A9_MUX0], 0>,
-                               InstrStage<1, [A9_NPipe]>],
-                              [2, 2, 2, 1]>,
+                               InstrStage<4, [A9_NPipe]>],
+                              [4, 4, 5, 1]>,
+  //
+  // VLD3ln
+  InstrItinData<IIC_VLD3ln,   [InstrStage<1, [A9_DRegsN],   0, Required>,
+                               InstrStage<11, [A9_DRegsVFP], 0, Reserved>,
+                               InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
+                               InstrStage<1, [A9_MUX0], 0>,
+                               InstrStage<5, [A9_NPipe]>],
+                              [5, 5, 6, 1, 1, 1, 1, 2]>,
+  //
+  // VLD3u
+  InstrItinData<IIC_VLD3u,    [InstrStage<1, [A9_DRegsN],   0, Required>,
+                               InstrStage<10, [A9_DRegsVFP], 0, Reserved>,
+                               InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
+                               InstrStage<1, [A9_MUX0], 0>,
+                               InstrStage<4, [A9_NPipe]>],
+                              [4, 4, 5, 2, 1]>,
+  //
+  // VLD3lnu
+  InstrItinData<IIC_VLD3lnu,  [InstrStage<1, [A9_DRegsN],   0, Required>,
+                               InstrStage<11, [A9_DRegsVFP], 0, Reserved>,
+                               InstrStage<1, [A9_Issue0, A9_Issue1], 0>,
+                               InstrStage<1, [A9_MUX0], 0>,
+                               InstrStage<5, [A9_NPipe]>],
+                              [5, 5, 6, 2, 1, 1, 1, 1, 1, 2]>,
   //
   // VLD4
   // FIXME: We don't model this instruction properly