[mips] Use less instruction to load zero into FPR by li.s / li.d pseudos

[atanasyan]

Summary

If li.s or li.d loads zero into a FPR, it's not necessary to load zero into at GPR register and then move its value into a floating point register. We can use as a source register the zero / $0 one.

[atanasyan]

Rebased against the master branch.

[atanasyan]

Rebased against the master branch.

[mstojanovic]

LGTM

llvm/lib/Target/Mips/AsmParser/MipsAsmParser.cpp

bool MipsAsmParser::expandLoadSingleImmToFPR(MCInst &Inst, SMLoc IDLoc,
 
   unsigned FirstReg = Inst.getOperand(0).getReg();
   uint64_t ImmOp64 = Inst.getOperand(1).getImm();
 
   ImmOp64 = convertIntToDoubleImm(ImmOp64);
 
   uint32_t ImmOp32 = covertDoubleImmToSingleImm(ImmOp64);
 
-  unsigned TmpReg = getATReg(IDLoc);
+  unsigned TmpReg = Mips::ZERO;
+  if (ImmOp32 != 0) {
+    TmpReg = getATReg(IDLoc);
     if (!TmpReg)
       return true;
+  }
 
   if (Lo_32(ImmOp64) == 0) {
-    if (loadImmediate(ImmOp32, TmpReg, Mips::NoRegister, true, true, IDLoc, Out,
-                      STI))
+    if (TmpReg != Mips::ZERO && loadImmediate(ImmOp32, TmpReg, Mips::NoRegister,
+                                              true, false, IDLoc, Out, STI))
       return true;
     TOut.emitRR(Mips::MTC1, FirstReg, TmpReg, IDLoc, STI);
     return false;
   }
 
   MCSection *CS = getStreamer().getCurrentSectionOnly();
   // FIXME: Enhance this expansion to use the .lit4 & .lit8 sections
   // where appropriate.
bool MipsAsmParser::expandLoadDoubleImmToFPR(MCInst &Inst, bool Is64FPU,
   unsigned FirstReg = Inst.getOperand(0).getReg();
   uint64_t ImmOp64 = Inst.getOperand(1).getImm();
 
   ImmOp64 = convertIntToDoubleImm(ImmOp64);
 
   uint32_t LoImmOp64 = Lo_32(ImmOp64);
   uint32_t HiImmOp64 = Hi_32(ImmOp64);
 
-  unsigned TmpReg = getATReg(IDLoc);
+  unsigned TmpReg = Mips::ZERO;
+  if (ImmOp64 != 0) {
+    TmpReg = getATReg(IDLoc);
     if (!TmpReg)
       return true;
+  }
 
   if ((LoImmOp64 == 0) &&
       !((HiImmOp64 & 0xffff0000) && (HiImmOp64 & 0x0000ffff))) {
-    // FIXME: In the case where the constant is zero, we can load the
-    // register directly from the zero register.
-
     if (isABI_N32() || isABI_N64()) {
-      if (loadImmediate(ImmOp64, TmpReg, Mips::NoRegister, false, false, IDLoc,
+      if (TmpReg != Mips::ZERO &&
+          loadImmediate(ImmOp64, TmpReg, Mips::NoRegister, false, false, IDLoc,
                         Out, STI))
         return true;
       TOut.emitRR(Mips::DMTC1, FirstReg, TmpReg, IDLoc, STI);
       return false;
     }
 
-    if (loadImmediate(HiImmOp64, TmpReg, Mips::NoRegister, true, false, IDLoc,
+    if (TmpReg != Mips::ZERO &&
+        loadImmediate(HiImmOp64, TmpReg, Mips::NoRegister, true, false, IDLoc,
                       Out, STI))
       return true;
 
     if (hasMips32r2()) {
       TOut.emitRR(Mips::MTC1, FirstReg, Mips::ZERO, IDLoc, STI);
       TOut.emitRRR(Mips::MTHC1_D32, FirstReg, FirstReg, TmpReg, IDLoc, STI);
     } else {
       TOut.emitRR(Mips::MTC1, nextReg(FirstReg), TmpReg, IDLoc, STI);

llvm/test/MC/Mips/macro-li.d.s

 # N64-PIC:                                            #   fixup A - offset: 0, value: %got([[LABEL]]), kind: fixup_Mips_GOT
 # N64-PIC:         daddiu  $1, $1, %lo([[LABEL]])     # encoding: [A,A,0x21,0x64]
 # N64-PIC:                                            #   fixup A - offset: 0, value: %lo([[LABEL]]), kind: fixup_Mips_LO16
 # O32:             lw      $4, 0($1)                  # encoding: [0x00,0x00,0x24,0x8c]
 # O32:             lw      $5, 4($1)                  # encoding: [0x04,0x00,0x25,0x8c]
 # N32-N64:         ld      $4, 0($1)                  # encoding: [0x00,0x00,0x24,0xdc]
 
 li.d  $f4, 0
-# O32:            addiu   $1, $zero, 0       # encoding: [0x00,0x00,0x01,0x24]
-# O32:            mtc1    $1, $f5            # encoding: [0x00,0x28,0x81,0x44]
+# O32:            mtc1    $zero, $f5         # encoding: [0x00,0x28,0x80,0x44]
 # O32:            mtc1    $zero, $f4         # encoding: [0x00,0x20,0x80,0x44]
-# CHECK-MIPS32r2: addiu   $1, $zero, 0       # encoding: [0x00,0x00,0x01,0x24]
 # CHECK-MIPS32r2: mtc1    $zero, $f4         # encoding: [0x00,0x20,0x80,0x44]
-# CHECK-MIPS32r2: mthc1   $1, $f4            # encoding: [0x00,0x20,0xe1,0x44]
-# N32-N64:        addiu   $1, $zero, 0       # encoding: [0x00,0x00,0x01,0x24]
-# N32-N64:        dmtc1   $1, $f4            # encoding: [0x00,0x20,0xa1,0x44]
+# CHECK-MIPS32r2: mthc1   $zero, $f4         # encoding: [0x00,0x20,0xe0,0x44]
+# N32-N64:        dmtc1   $zero, $f4         # encoding: [0x00,0x20,0xa0,0x44]
 
 li.d  $f4, 0.0
-# O32:            addiu   $1, $zero, 0       # encoding: [0x00,0x00,0x01,0x24]
-# O32:            mtc1    $1, $f5            # encoding: [0x00,0x28,0x81,0x44]
+# O32:            mtc1    $zero, $f5         # encoding: [0x00,0x28,0x80,0x44]
 # O32:            mtc1    $zero, $f4         # encoding: [0x00,0x20,0x80,0x44]
-# CHECK-MIPS32r2: addiu   $1, $zero, 0       # encoding: [0x00,0x00,0x01,0x24]
 # CHECK-MIPS32r2: mtc1    $zero, $f4         # encoding: [0x00,0x20,0x80,0x44]
-# CHECK-MIPS32r2: mthc1   $1, $f4            # encoding: [0x00,0x20,0xe1,0x44]
-# N32-N64:        addiu   $1, $zero, 0       # encoding: [0x00,0x00,0x01,0x24]
-# N32-N64:        dmtc1   $1, $f4            # encoding: [0x00,0x20,0xa1,0x44]
+# CHECK-MIPS32r2: mthc1   $zero, $f4         # encoding: [0x00,0x20,0xe0,0x44]
+# N32-N64:        dmtc1   $zero, $f4         # encoding: [0x00,0x20,0xa0,0x44]
 
 li.d  $f4, 1.12345
 # ALL:  .section  .rodata,"a",@progbits
 # ALL:  [[LABEL:\$tmp[0-9]+]]:
 # ALL:  .4byte  1072822694
 # ALL:  .4byte  3037400872
 # ALL:  .text
 # O32-N32-PIC:     lw      $1, %got([[LABEL]])($gp)   # encoding: [A,A,0x81,0x8f]

llvm/test/MC/Mips/macro-li.s.s

 # ALL:   lui     $4, 20535                   # encoding: [0x37,0x50,0x04,0x3c]
 # ALL:   ori     $4, $4, 63239               # encoding: [0x07,0xf7,0x84,0x34]
 
 li.s  $4, 12345678910123456789.12345678910
 # ALL:   lui     $4, 24363                   # encoding: [0x2b,0x5f,0x04,0x3c]
 # ALL:   ori     $4, $4, 21674               # encoding: [0xaa,0x54,0x84,0x34]
 
 li.s  $f4, 0
-# ALL:   addiu   $1, $zero, 0                # encoding: [0x00,0x00,0x01,0x24]
-# ALL:   mtc1    $1, $f4                     # encoding: [0x00,0x20,0x81,0x44]
+# ALL:   mtc1    $zero, $f4                  # encoding: [0x00,0x20,0x80,0x44]
 
 li.s  $f4, 0.0
-# ALL:   addiu   $1, $zero, 0                # encoding: [0x00,0x00,0x01,0x24]
-# ALL:   mtc1    $1, $f4                     # encoding: [0x00,0x20,0x81,0x44]
+# ALL:   mtc1    $zero, $f4                  # encoding: [0x00,0x20,0x80,0x44]
 
 li.s  $f4, 1.12345
 # ALL:  .section  .rodata,"a",@progbits
 # ALL:  [[LABEL:\$tmp[0-9]+]]:
 # ALL:  .4byte  1066388790
 # ALL:  .text
 # O32-N32-PIC:     lw      $1, %got([[LABEL]])($gp)   # encoding: [A,A,0x81,0x8f]
 # O32-N32-PIC:                                        #   fixup A - offset: 0, value: %got([[LABEL]]), kind: fixup_Mips_GOT