Add variants of VST2, VST3 and VST4 with address register writeback, and

rewrite the existing VST3 and VST4 instructions to use the same classes as
the others.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99093 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 80 insertions, 30 deletions

diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index 19b3e08582..69b77f484a 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -616,21 +616,45 @@ def  VST2q8   : VST2Q<0b0000, "8">;
 def  VST2q16  : VST2Q<0b0100, "16">;
 def  VST2q32  : VST2Q<0b1000, "32">;
 
+// ...with address register writeback:
+class VST2DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
+  : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
+          (ins addrmode6:$addr, DPR:$src1, DPR:$src2),
+          IIC_VST, "vst2", Dt, "\\{$src1, $src2\\}, $addr",
+          "$addr.addr = $wb", []>;
+class VST2QWB<bits<4> op7_4, string Dt>
+  : NLdSt<0, 0b00, 0b0011, op7_4, (outs GPR:$wb),
+          (ins addrmode6:$addr,
+           DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
+          IIC_VST, "vst2", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
+          "$addr.addr = $wb", []>;
+
+def VST2d8_UPD  : VST2DWB<0b1000, 0b0000, "8">;
+def VST2d16_UPD : VST2DWB<0b1000, 0b0100, "16">;
+def VST2d32_UPD : VST2DWB<0b1000, 0b1000, "32">;
+def VST2d64_UPD : NLdSt<0,0b00,0b1010,0b1100, (outs GPR:$wb),
+                        (ins addrmode6:$addr,
+                         DPR:$src1, DPR:$src2), IIC_VST,
+                        "vst1", "64", "\\{$src1, $src2\\}, $addr",
+                        "$addr.addr = $wb", []>;
+
+def VST2q8_UPD  : VST2QWB<0b0000, "8">;
+def VST2q16_UPD : VST2QWB<0b0100, "16">;
+def VST2q32_UPD : VST2QWB<0b1000, "32">;
+
 // ...with double-spaced registers (for disassembly only):
 def VST2b8      : VST2D<0b1001, 0b0000, "8">;
 def VST2b16     : VST2D<0b1001, 0b0100, "16">;
 def VST2b32     : VST2D<0b1001, 0b1000, "32">;
+def VST2b8_UPD  : VST2DWB<0b1001, 0b0000, "8">;
+def VST2b16_UPD : VST2DWB<0b1001, 0b0100, "16">;
+def VST2b32_UPD : VST2DWB<0b1001, 0b1000, "32">;
 
 //   VST3     : Vector Store (multiple 3-element structures)
 class VST3D<bits<4> op11_8, bits<4> op7_4, string Dt>
   : NLdSt<0, 0b00, op11_8, op7_4, (outs),
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
           "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr", "", []>;
-class VST3WB<bits<4> op7_4, string Dt>
-  : NLdSt<0,0b00,0b0101,op7_4, (outs GPR:$wb),
-          (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
-          "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr",
-          "$addr.addr = $wb", []>;
 
 def  VST3d8   : VST3D<0b0100, 0b0000, "8">;
 def  VST3d16  : VST3D<0b0100, 0b0100, "16">;
@@ -640,20 +664,35 @@ def  VST3d64  : NLdSt<0,0b00,0b0110,0b1100, (outs),
                       IIC_VST,
                       "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr", "", []>;
 
-// ...with double-spaced registers:
+// ...with address register writeback:
+class VST3DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
+  : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
+          (ins addrmode6:$addr,
+           DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
+          "vst3", Dt, "\\{$src1, $src2, $src3\\}, $addr",
+          "$addr.addr = $wb", []>;
+
+def VST3d8_UPD  : VST3DWB<0b0100, 0b0000, "8">;
+def VST3d16_UPD : VST3DWB<0b0100, 0b0100, "16">;
+def VST3d32_UPD : VST3DWB<0b0100, 0b1000, "32">;
+def VST3d64_UPD : NLdSt<0,0b00,0b0110,0b1100, (outs GPR:$wb),
+                      (ins addrmode6:$addr,
+                       DPR:$src1, DPR:$src2, DPR:$src3), IIC_VST,
+                      "vst1", "64", "\\{$src1, $src2, $src3\\}, $addr",
+                      "$addr.addr = $wb", []>;
+
+// ...with double-spaced registers (non-updating versions for disassembly only):
 def VST3q8      : VST3D<0b0101, 0b0000, "8">;
 def VST3q16     : VST3D<0b0101, 0b0100, "16">;
 def VST3q32     : VST3D<0b0101, 0b1000, "32">;
+def VST3q8_UPD  : VST3DWB<0b0101, 0b0000, "8">;
+def VST3q16_UPD : VST3DWB<0b0101, 0b0100, "16">;
+def VST3q32_UPD : VST3DWB<0b0101, 0b1000, "32">;
 
-// vst3 to double-spaced even registers.
-def VST3q8_UPD  : VST3WB<0b0000, "8">;
-def VST3q16_UPD : VST3WB<0b0100, "16">;
-def VST3q32_UPD : VST3WB<0b1000, "32">;
-
-// vst3 to double-spaced odd registers.
-def VST3q8odd_UPD  : VST3WB<0b0000, "8">;
-def VST3q16odd_UPD : VST3WB<0b0100, "16">;
-def VST3q32odd_UPD : VST3WB<0b1000, "32">;
+// ...alternate versions to be allocated odd register numbers:
+def VST3q8odd_UPD  : VST3DWB<0b0101, 0b0000, "8">;
+def VST3q16odd_UPD : VST3DWB<0b0101, 0b0100, "16">;
+def VST3q32odd_UPD : VST3DWB<0b0101, 0b1000, "32">;
 
 //   VST4     : Vector Store (multiple 4-element structures)
 class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
@@ -661,11 +700,6 @@ class VST4D<bits<4> op11_8, bits<4> op7_4, string Dt>
           (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
           IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
           "", []>;
-class VST4WB<bits<4> op7_4, string Dt>
-  : NLdSt<0,0b00,0b0001,op7_4, (outs GPR:$wb),
-          (ins addrmode6:$addr, DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4),
-          IIC_VST, "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
-          "$addr.addr = $wb", []>;
 
 def  VST4d8   : VST4D<0b0000, 0b0000, "8">;
 def  VST4d16  : VST4D<0b0000, 0b0100, "16">;
@@ -676,20 +710,36 @@ def  VST4d64  : NLdSt<0,0b00,0b0010,0b1100, (outs),
                       "vst1", "64", "\\{$src1, $src2, $src3, $src4\\}, $addr",
                       "", []>;
 
-// ...with double-spaced registers:
+// ...with address register writeback:
+class VST4DWB<bits<4> op11_8, bits<4> op7_4, string Dt>
+  : NLdSt<0, 0b00, op11_8, op7_4, (outs GPR:$wb),
+          (ins addrmode6:$addr,
+           DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
+           "vst4", Dt, "\\{$src1, $src2, $src3, $src4\\}, $addr",
+          "$addr.addr = $wb", []>;
+
+def VST4d8_UPD  : VST4DWB<0b0000, 0b0000, "8">;
+def VST4d16_UPD : VST4DWB<0b0000, 0b0100, "16">;
+def VST4d32_UPD : VST4DWB<0b0000, 0b1000, "32">;
+def VST4d64_UPD : NLdSt<0,0b00,0b0010,0b1100, (outs GPR:$wb),
+                      (ins addrmode6:$addr,
+                       DPR:$src1, DPR:$src2, DPR:$src3, DPR:$src4), IIC_VST,
+                      "vst1", "64",
+                      "\\{$src1, $src2, $src3, $src4\\}, $addr",
+                      "$addr.addr = $wb", []>;
+
+// ...with double-spaced registers (non-updating versions for disassembly only):
 def VST4q8      : VST4D<0b0001, 0b0000, "8">;
 def VST4q16     : VST4D<0b0001, 0b0100, "16">;
 def VST4q32     : VST4D<0b0001, 0b1000, "32">;
+def VST4q8_UPD  : VST4DWB<0b0001, 0b0000, "8">;
+def VST4q16_UPD : VST4DWB<0b0001, 0b0100, "16">;
+def VST4q32_UPD : VST4DWB<0b0001, 0b1000, "32">;
 
-// vst4 to double-spaced even registers.
-def VST4q8_UPD  : VST4WB<0b0000, "8">;
-def VST4q16_UPD : VST4WB<0b0100, "16">;
-def VST4q32_UPD : VST4WB<0b1000, "32">;
-
-// vst4 to double-spaced odd registers.
-def VST4q8odd_UPD  : VST4WB<0b0000, "8">;
-def VST4q16odd_UPD : VST4WB<0b0100, "16">;
-def VST4q32odd_UPD : VST4WB<0b1000, "32">;
+// ...alternate versions to be allocated odd register numbers:
+def VST4q8odd_UPD  : VST4DWB<0b0001, 0b0000, "8">;
+def VST4q16odd_UPD : VST4DWB<0b0001, 0b0100, "16">;
+def VST4q32odd_UPD : VST4DWB<0b0001, 0b1000, "32">;
 
 //   VST1LN   : Vector Store (single element from one lane)
 //   FIXME: Not yet implemented.