[ARM] Improve smul* and smla* isel for Thumb2

Added (sra (shl x, 16), 16) to the sext_16_node PatLeaf for ARM to
simplify some pattern matching. This has allowed several patterns
for smul* and smla* to be removed as well as making it easier to add
the matching for the corresponding instructions for Thumb2 targets.
Also added two Pat classes that are predicated on Thumb2 with the
hasDSP flag and UseMulOps flags. Updated the smul codegen test with
the wider range of patterns plus the ThumbV6 and ThumbV6T2 targets.

Differential Revision: https://reviews.llvm.org/D22908



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@277450 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 41 insertions, 25 deletions

diff --git a/lib/Target/ARM/ARMInstrFormats.td b/lib/Target/ARM/ARMInstrFormats.td
index 37a83f70a1f..25f92df7e11 100644
--- a/lib/Target/ARM/ARMInstrFormats.td
+++ b/lib/Target/ARM/ARMInstrFormats.td
@@ -999,6 +999,12 @@ class VFPPat<dag pattern, dag result> : Pat<pattern, result> {
 class VFPNoNEONPat<dag pattern, dag result> : Pat<pattern, result> {
   list<Predicate> Predicates = [HasVFP2, DontUseNEONForFP];
 }
+class Thumb2DSPPat<dag pattern, dag result> : Pat<pattern, result> {
+  list<Predicate> Predicates = [IsThumb2, HasDSP];
+}
+class Thumb2DSPMulPat<dag pattern, dag result> : Pat<pattern, result> {
+  list<Predicate> Predicates = [IsThumb2, UseMulOps, HasDSP];
+}
 //===----------------------------------------------------------------------===//
 // Thumb Instruction Format Definitions.
 //
diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index c9735f3ec27..b7abc48fe08 100644
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -358,7 +358,23 @@ def imm16_31 : ImmLeaf<i32, [{
 
 // sext_16_node predicate - True if the SDNode is sign-extended 16 or more bits.
 def sext_16_node : PatLeaf<(i32 GPR:$a), [{
-  return CurDAG->ComputeNumSignBits(SDValue(N,0)) >= 17;
+  if (CurDAG->ComputeNumSignBits(SDValue(N,0)) >= 17)
+    return true;
+
+  if (N->getOpcode() != ISD::SRA)
+    return false;
+  if (N->getOperand(0).getOpcode() != ISD::SHL)
+    return false;
+
+  auto *ShiftVal = dyn_cast<ConstantSDNode>(N->getOperand(1));
+  if (!ShiftVal || ShiftVal->getZExtValue() != 16)
+    return false;
+
+  ShiftVal = dyn_cast<ConstantSDNode>(N->getOperand(0)->getOperand(1));
+  if (!ShiftVal || ShiftVal->getZExtValue() != 16)
+    return false;
+
+  return true;
 }]>;
 
 /// Split a 32-bit immediate into two 16 bit parts.
@@ -5492,45 +5508,22 @@ def : ARMPat<(extloadi8  addrmodepc:$addr), (PICLDRB addrmodepc:$addr)>;
 def : ARMPat<(extloadi16 addrmodepc:$addr), (PICLDRH addrmodepc:$addr)>;
 
 // smul* and smla*
-def : ARMV5TEPat<(mul (sra (shl GPR:$a, (i32 16)), (i32 16)),
-                      (sra (shl GPR:$b, (i32 16)), (i32 16))),
-                 (SMULBB GPR:$a, GPR:$b)>;
 def : ARMV5TEPat<(mul sext_16_node:$a, sext_16_node:$b),
                  (SMULBB GPR:$a, GPR:$b)>;
-def : ARMV5TEPat<(mul (sra (shl GPR:$a, (i32 16)), (i32 16)),
-                      (sra GPR:$b, (i32 16))),
-                 (SMULBT GPR:$a, GPR:$b)>;
 def : ARMV5TEPat<(mul sext_16_node:$a, (sra GPR:$b, (i32 16))),
                  (SMULBT GPR:$a, GPR:$b)>;
-def : ARMV5TEPat<(mul (sra GPR:$a, (i32 16)),
-                      (sra (shl GPR:$b, (i32 16)), (i32 16))),
-                 (SMULTB GPR:$a, GPR:$b)>;
 def : ARMV5TEPat<(mul (sra GPR:$a, (i32 16)), sext_16_node:$b),
                 (SMULTB GPR:$a, GPR:$b)>;
-
-def : ARMV5MOPat<(add GPR:$acc,
-                      (mul (sra (shl GPR:$a, (i32 16)), (i32 16)),
-                           (sra (shl GPR:$b, (i32 16)), (i32 16)))),
-                 (SMLABB GPR:$a, GPR:$b, GPR:$acc)>;
 def : ARMV5MOPat<(add GPR:$acc,
                       (mul sext_16_node:$a, sext_16_node:$b)),
                  (SMLABB GPR:$a, GPR:$b, GPR:$acc)>;
 def : ARMV5MOPat<(add GPR:$acc,
-                      (mul (sra (shl GPR:$a, (i32 16)), (i32 16)),
-                           (sra GPR:$b, (i32 16)))),
-                 (SMLABT GPR:$a, GPR:$b, GPR:$acc)>;
-def : ARMV5MOPat<(add GPR:$acc,
                       (mul sext_16_node:$a, (sra GPR:$b, (i32 16)))),
                  (SMLABT GPR:$a, GPR:$b, GPR:$acc)>;
 def : ARMV5MOPat<(add GPR:$acc,
-                      (mul (sra GPR:$a, (i32 16)),
-                           (sra (shl GPR:$b, (i32 16)), (i32 16)))),
-                 (SMLATB GPR:$a, GPR:$b, GPR:$acc)>;
-def : ARMV5MOPat<(add GPR:$acc,
                       (mul (sra GPR:$a, (i32 16)), sext_16_node:$b)),
                  (SMLATB GPR:$a, GPR:$b, GPR:$acc)>;
 
-
 // Pre-v7 uses MCR for synchronization barriers.
 def : ARMPat<(ARMMemBarrierMCR GPR:$zero), (MCR 15, 0, GPR:$zero, 7, 10, 5)>,
          Requires<[IsARM, HasV6]>;
diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td
index 8f59a5b8cfc..d908b8a906b 100644
--- a/lib/Target/ARM/ARMInstrThumb2.td
+++ b/lib/Target/ARM/ARMInstrThumb2.td
@@ -2640,7 +2640,15 @@ def t2SMULTT : T2ThreeRegSMUL<0b001, 0b11, "smultt",
 def t2SMULWB : T2ThreeRegSMUL<0b011, 0b00, "smulwb", []>;
 def t2SMULWT : T2ThreeRegSMUL<0b011, 0b01, "smulwt", []>;
 
-class T2FourRegSMLA<bits<3> op22_20, bits<2> op5_4, string opc, list<dag> pattern>
+def : Thumb2DSPPat<(mul sext_16_node:$Rm, sext_16_node:$Rn),
+                   (t2SMULBB rGPR:$Rm, rGPR:$Rn)>;
+def : Thumb2DSPPat<(mul sext_16_node:$Rn, (sra rGPR:$Rm, (i32 16))),
+                   (t2SMULBT rGPR:$Rn, rGPR:$Rm)>;
+def : Thumb2DSPPat<(mul (sra rGPR:$Rn, (i32 16)), sext_16_node:$Rm),
+                   (t2SMULTB rGPR:$Rn, rGPR:$Rm)>;
+
+class T2FourRegSMLA<bits<3> op22_20, bits<2> op5_4, string opc,
+                    list<dag> pattern>
   : T2FourReg<(outs rGPR:$Rd), (ins rGPR:$Rn, rGPR:$Rm, rGPR:$Ra), IIC_iMUL16,
                opc, "\t$Rd, $Rn, $Rm, $Ra", pattern>,
     Requires<[IsThumb2, HasDSP, UseMulOps]> {
@@ -2667,6 +2675,15 @@ def t2SMLATT : T2FourRegSMLA<0b001, 0b11, "smlatt",
 def t2SMLAWB : T2FourRegSMLA<0b011, 0b00, "smlawb", []>;
 def t2SMLAWT : T2FourRegSMLA<0b011, 0b01, "smlawt", []>;
 
+def : Thumb2DSPMulPat<(add rGPR:$Ra, (mul sext_16_node:$Rn, sext_16_node:$Rm)),
+                      (t2SMLABB rGPR:$Rn, rGPR:$Rm, rGPR:$Ra)>;
+def : Thumb2DSPMulPat<(add rGPR:$Ra,
+                        (mul sext_16_node:$Rn, (sra rGPR:$Rm, (i32 16)))),
+                      (t2SMLABT rGPR:$Rn, rGPR:$Rm, rGPR:$Ra)>;
+def : Thumb2DSPMulPat<(add rGPR:$Ra,
+                        (mul (sra rGPR:$Rn, (i32 16)), sext_16_node:$Rm)),
+                      (t2SMLATB rGPR:$Rn, rGPR:$Rm, rGPR:$Ra)>;
+
 class T2SMLAL<bits<3> op22_20, bits<4> op7_4, string opc, list<dag> pattern>
   : T2FourReg_mac<1, op22_20, op7_4,
                   (outs rGPR:$Ra, rGPR:$Rd),