Add ADD and SUB to the predicable ARM instructions.

It is not my plan to duplicate the entire ARM instruction set with
predicated versions. We need a way of representing predicated
instructions in SSA form without requiring a separate opcode.

Then the pseudo-instructions can go away.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@162061 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/ARMBaseInstrInfo.cpp b/lib/Target/ARM/ARMBaseInstrInfo.cpp
index f5a3bcd11f3cd732d6f7d6fb21c2d57b371d5458..0261bb3ed7e8750011ffee6f98e09a03e6694a50 100644 (file)
--- a/lib/Target/ARM/ARMBaseInstrInfo.cpp
+++ b/lib/Target/ARM/ARMBaseInstrInfo.cpp
@@ -1613,6 +1613,26 @@ static unsigned canFoldIntoMOVCC(unsigned Reg, MachineInstr *&MI,
   case ARM::t2ORRri: return ARM::t2ORRCCri;
   case ARM::t2ORRrr: return ARM::t2ORRCCrr;
   case ARM::t2ORRrs: return ARM::t2ORRCCrs;
+
+  // ARM ADD/SUB
+  case ARM::ADDri:   return ARM::ADDCCri;
+  case ARM::ADDrr:   return ARM::ADDCCrr;
+  case ARM::ADDrsi:  return ARM::ADDCCrsi;
+  case ARM::ADDrsr:  return ARM::ADDCCrsr;
+  case ARM::SUBri:   return ARM::SUBCCri;
+  case ARM::SUBrr:   return ARM::SUBCCrr;
+  case ARM::SUBrsi:  return ARM::SUBCCrsi;
+  case ARM::SUBrsr:  return ARM::SUBCCrsr;
+
+  // Thumb2 ADD/SUB
+  case ARM::t2ADDri:   return ARM::t2ADDCCri;
+  case ARM::t2ADDri12: return ARM::t2ADDCCri12;
+  case ARM::t2ADDrr:   return ARM::t2ADDCCrr;
+  case ARM::t2ADDrs:   return ARM::t2ADDCCrs;
+  case ARM::t2SUBri:   return ARM::t2SUBCCri;
+  case ARM::t2SUBri12: return ARM::t2SUBCCri12;
+  case ARM::t2SUBrr:   return ARM::t2SUBCCrr;
+  case ARM::t2SUBrs:   return ARM::t2SUBCCrs;
   }
 }
 

diff --git a/lib/Target/ARM/ARMInstrInfo.td b/lib/Target/ARM/ARMInstrInfo.td
index db83a9feb65649c2eeb07fec276fc894364d6abf..992aba5803f6e066c2f060c5c80dbb273c827071 100644 (file)
--- a/lib/Target/ARM/ARMInstrInfo.td
+++ b/lib/Target/ARM/ARMInstrInfo.td
@@ -4023,6 +4023,10 @@ defm ORRCC : AsI1_bincc_irs<ORRri, ORRrr, ORRrsi, ORRrsr,
                             IIC_iBITi, IIC_iBITr, IIC_iBITsr>;
 defm EORCC : AsI1_bincc_irs<EORri, EORrr, EORrsi, EORrsr,
                             IIC_iBITi, IIC_iBITr, IIC_iBITsr>;
+defm ADDCC : AsI1_bincc_irs<ADDri, ADDrr, ADDrsi, ADDrsr,
+                            IIC_iBITi, IIC_iBITr, IIC_iBITsr>;
+defm SUBCC : AsI1_bincc_irs<SUBri, SUBrr, SUBrsi, SUBrsr,
+                            IIC_iBITi, IIC_iBITr, IIC_iBITsr>;
 
 } // neverHasSideEffects
 

diff --git a/lib/Target/ARM/ARMInstrThumb2.td b/lib/Target/ARM/ARMInstrThumb2.td
index 2761b50f4410ef17efdf00e600cef5f8e2984b99..8ecf0091d8b6a2e53c8b731e9cbd3a15a363e9a7 100644 (file)
--- a/lib/Target/ARM/ARMInstrThumb2.td
+++ b/lib/Target/ARM/ARMInstrThumb2.td
@@ -757,6 +757,33 @@ multiclass T2I_bin_ii12rs<bits<3> op23_21, string opc, PatFrag opnode,
      let Inst{24} = 1;
      let Inst{23-21} = op23_21;
    }
+
+   // Predicated versions.
+   def CCri : t2PseudoExpand<(outs GPRnopc:$Rd),
+                             (ins GPRnopc:$Rfalse, GPRnopc:$Rn, t2_so_imm:$imm,
+                                  pred:$p, cc_out:$s), 4, IIC_iALUi, [],
+                             (!cast<Instruction>(NAME#ri) GPRnopc:$Rd,
+                              GPRnopc:$Rn, t2_so_imm:$imm, pred:$p, cc_out:$s)>,
+              RegConstraint<"$Rfalse = $Rd">;
+   def CCri12 : t2PseudoExpand<(outs GPRnopc:$Rd),
+                             (ins GPRnopc:$Rfalse, GPR:$Rn, imm0_4095:$imm,
+                                  pred:$p),
+                             4, IIC_iALUi, [],
+                             (!cast<Instruction>(NAME#ri12) GPRnopc:$Rd,
+                              GPR:$Rn, imm0_4095:$imm, pred:$p)>,
+                RegConstraint<"$Rfalse = $Rd">;
+   def CCrr : t2PseudoExpand<(outs GPRnopc:$Rd),
+                             (ins GPRnopc:$Rfalse, GPRnopc:$Rn, rGPR:$Rm,
+                                  pred:$p, cc_out:$s), 4, IIC_iALUr, [],
+                             (!cast<Instruction>(NAME#rr) GPRnopc:$Rd,
+                              GPRnopc:$Rn, rGPR:$Rm, pred:$p, cc_out:$s)>,
+              RegConstraint<"$Rfalse = $Rd">;
+   def CCrs : t2PseudoExpand<(outs GPRnopc:$Rd),
+                             (ins GPRnopc:$Rfalse, GPRnopc:$Rn, t2_so_reg:$Rm,
+                                  pred:$p, cc_out:$s), 4, IIC_iALUsi, [],
+                             (!cast<Instruction>(NAME#rs) GPRnopc:$Rd,
+                              GPRnopc:$Rn, t2_so_reg:$Rm, pred:$p, cc_out:$s)>,
+              RegConstraint<"$Rfalse = $Rd">;
 }
 
 /// T2I_adde_sube_irs - Defines a set of (op reg, {so_imm|r|so_reg}) patterns

diff --git a/test/CodeGen/ARM/select.ll b/test/CodeGen/ARM/select.ll
index 418d4f31ee2bbb96be9089de8326d80962df3594..557556662892142c32269c5615d2b25c046b8f87 100644 (file)
--- a/test/CodeGen/ARM/select.ll
+++ b/test/CodeGen/ARM/select.ll
@@ -76,12 +76,11 @@ define double @f7(double %a, double %b) {
 ; block generated, odds are good that we have close to the ideal code for this:
 ;
 ; CHECK-NEON:      _f8:
+; CHECK-NEON:      movw    [[R3:r[0-9]+]], #1123
 ; CHECK-NEON:      adr     [[R2:r[0-9]+]], LCPI7_0
-; CHECK-NEON-NEXT: movw    [[R3:r[0-9]+]], #1123
-; CHECK-NEON-NEXT: adds    {{r.*}}, [[R2]], #4
 ; CHECK-NEON-NEXT: cmp     r0, [[R3]]
-; CHECK-NEON-NEXT: it      ne
-; CHECK-NEON-NEXT: movne   {{r.*}}, [[R2]]
+; CHECK-NEON-NEXT: it      eq
+; CHECK-NEON-NEXT: addeq.w {{r.*}}, [[R2]]
 ; CHECK-NEON-NEXT: ldr
 ; CHECK-NEON:      bx
 

diff --git a/test/CodeGen/ARM/select_xform.ll b/test/CodeGen/ARM/select_xform.ll
index d98d7a628e29f0112230aecafd32cf8de97ee7f1..26f7cb68901fb2df7cfa4a99af6d3bc7666a2cfe 100644 (file)
--- a/test/CodeGen/ARM/select_xform.ll
+++ b/test/CodeGen/ARM/select_xform.ll
@@ -4,13 +4,13 @@
 
 define i32 @t1(i32 %a, i32 %b, i32 %c) nounwind {
 ; ARM: t1:
-; ARM: sub r0, r1, #-2147483647
-; ARM: movgt r0, r1
+; ARM: suble r1, r1, #-2147483647
+; ARM: mov r0, r1
 
 ; T2: t1:
 ; T2: mvn r0, #-2147483648
-; T2: add r0, r1
-; T2: movgt r0, r1
+; T2: addle.w r1, r1
+; T2: mov r0, r1
   %tmp1 = icmp sgt i32 %c, 10
   %tmp2 = select i1 %tmp1, i32 0, i32 2147483647
   %tmp3 = add i32 %tmp2, %b
@@ -19,12 +19,12 @@ define i32 @t1(i32 %a, i32 %b, i32 %c) nounwind {
 
 define i32 @t2(i32 %a, i32 %b, i32 %c, i32 %d) nounwind {
 ; ARM: t2:
-; ARM: sub r0, r1, #10
-; ARM: movgt r0, r1
+; ARM: suble r1, r1, #10
+; ARM: mov r0, r1
 
 ; T2: t2:
-; T2: sub.w r0, r1, #10
-; T2: movgt r0, r1
+; T2: suble.w r1, r1, #10
+; T2: mov r0, r1
   %tmp1 = icmp sgt i32 %c, 10
   %tmp2 = select i1 %tmp1, i32 0, i32 10
   %tmp3 = sub i32 %b, %tmp2
@@ -164,3 +164,18 @@ define i32 @t11(i32 %a, i32 %b) nounwind {
   %tmp1 = select i1 %cond, i32 %x, i32 %a
   ret i32 %tmp1
 }
+
+; Fold ADDri12 into movcc
+define i32 @t12(i32 %a, i32 %b) nounwind {
+; ARM: t12:
+; ARM: cmp r0, r1
+; ARM: addge r0, r1,
+
+; T2: t12:
+; T2: cmp r0, r1
+; T2: addwge r0, r1, #3000
+  %x = add i32 %b, 3000
+  %cond = icmp slt i32 %a, %b
+  %tmp1 = select i1 %cond, i32 %a, i32 %x
+  ret i32 %tmp1
+}

diff --git a/test/CodeGen/Thumb2/thumb2-select_xform.ll b/test/CodeGen/Thumb2/thumb2-select_xform.ll
index 74729fd4150f2333bcb6416c0e9b3d04a8347d27..ead198f21624e42d21b36b555a027ff4ae774177 100644 (file)
--- a/test/CodeGen/Thumb2/thumb2-select_xform.ll
+++ b/test/CodeGen/Thumb2/thumb2-select_xform.ll
@@ -4,9 +4,9 @@ define i32 @t1(i32 %a, i32 %b, i32 %c) nounwind {
 ; CHECK: t1
 ; CHECK: mvn r0, #-2147483648
 ; CHECK: cmp r2, #10
-; CHECK: add r0, r1
-; CHECK: it  gt
-; CHECK: movgt r0, r1
+; CHECK: it  le
+; CHECK: addle.w r1, r1, r0
+; CHECK: mov r0, r1
         %tmp1 = icmp sgt i32 %c, 10
         %tmp2 = select i1 %tmp1, i32 0, i32 2147483647
         %tmp3 = add i32 %tmp2, %b
@@ -15,10 +15,10 @@ define i32 @t1(i32 %a, i32 %b, i32 %c) nounwind {
 
 define i32 @t2(i32 %a, i32 %b, i32 %c) nounwind {
 ; CHECK: t2
-; CHECK: add.w r0, r1, #-2147483648
 ; CHECK: cmp r2, #10
-; CHECK: it  gt
-; CHECK: movgt r0, r1
+; CHECK: it  le
+; CHECK: addle.w r1, r1, #-2147483648
+; CHECK: mov r0, r1
 
         %tmp1 = icmp sgt i32 %c, 10
         %tmp2 = select i1 %tmp1, i32 0, i32 2147483648
@@ -28,10 +28,10 @@ define i32 @t2(i32 %a, i32 %b, i32 %c) nounwind {
 
 define i32 @t3(i32 %a, i32 %b, i32 %c, i32 %d) nounwind {
 ; CHECK: t3
-; CHECK: sub.w r0, r1, #10
 ; CHECK: cmp r2, #10
-; CHECK: it  gt
-; CHECK: movgt r0, r1
+; CHECK: it  le
+; CHECK: suble.w r1, r1, #10
+; CHECK: mov r0, r1
         %tmp1 = icmp sgt i32 %c, 10
         %tmp2 = select i1 %tmp1, i32 0, i32 10
         %tmp3 = sub i32 %b, %tmp2