MIPS: Fixed and optimized MacroAssembler::Trunc_uw_d, Cvt_d_uw, Ext, Ins.

src/mips/macro-assembler-mips.cc

 // Copyright 2011 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
@@ skipping 739 matching lines @@
   }
   addiu(sp, sp, 4 * NumSaved);
 }
 
 
 void MacroAssembler::Ext(Register rt,
                          Register rs,
                          uint16_t pos,
                          uint16_t size) {
   ASSERT(pos < 32);
-  ASSERT(pos + size < 32);
+  ASSERT(pos + size < 33);
 
   if (mips32r2) {
     ext_(rt, rs, pos, size);
   } else {
     // Move rs to rt and shift it left then right to get the
     // desired bitfield on the right side and zeroes on the left.
-    sll(rt, rs, 32 - (pos + size));
-    srl(rt, rt, 32 - size);
+    int shift_left = 32 - (pos + size);
+    if (shift_left > 0) {
+      sll(rt, rs, shift_left);
+    }
+
+    int shift_right = 32 - size;
+    if (shift_right > 0) {
+      srl(rt, rt, shift_right);
+    }
   }
 }
 
 
 void MacroAssembler::Ins(Register rt,
                          Register rs,
                          uint16_t pos,
                          uint16_t size) {
   ASSERT(pos < 32);
   ASSERT(pos + size < 32);
@@ skipping 21 matching lines @@
     // t8 now contains the chunk from rs on the left and zeroes.
     srl(t8, t8, 32 - size - pos);
     // t8 now contains the original chunk from rs in
     // the middle (proper position).
     or_(rt, rt, t8);
     // rt now contains the result of the ins instruction in R2 mode.
   }
 }
 
 
-void MacroAssembler::Cvt_d_uw(FPURegister fd, FPURegister fs) {
-  // Move the data from fs to t4.
-  mfc1(t4, fs);
-  return Cvt_d_uw(fd, t4);
+void MacroAssembler::Cvt_d_uw(FPURegister fd,
+                              FPURegister fs,
+                              FPURegister scratch) {
+  // Move the data from fs to t8.
+  mfc1(t8, fs);
+  Cvt_d_uw(fd, t8, scratch);
 }
 
 
-void MacroAssembler::Cvt_d_uw(FPURegister fd, Register rs) {
+void MacroAssembler::Cvt_d_uw(FPURegister fd,
+                              Register rs,
+                              FPURegister scratch) {
   // Convert rs to a FP value in fd (and fd + 1).
   // We do this by converting rs minus the MSB to avoid sign conversion,
-  // then adding 2^31-1 and 1 to the result.
+  // then adding 2^31 to the result (if needed).
 
-  ASSERT(!fd.is(f20));
+  ASSERT(!fd.is(scratch));
   ASSERT(!rs.is(t9));
-  ASSERT(!rs.is(t8));
+  ASSERT(!rs.is(at));
 
-  // Save rs's MSB to t8.
-  And(t8, rs, 0x80000000);
+  // Save rs's MSB to t9.
+  Ext(t9, rs, 31, 1);
   // Remove rs's MSB.
-  And(t9, rs, 0x7FFFFFFF);
-  // Move t9 to fd.
-  mtc1(t9, fd);
+  Ext(at, rs, 0, 31);
+  // Move the result to fd.
+  mtc1(at, fd);
 
   // Convert fd to a real FP value.
   cvt_d_w(fd, fd);
 
   Label conversion_done;
 
   // If rs's MSB was 0, it's done.
   // Otherwise we need to add that to the FP register.
-  Branch(&conversion_done, eq, t8, Operand(zero_reg));
+  Branch(&conversion_done, eq, t9, Operand(zero_reg));
 
-  // First load 2^31 - 1 into f20.
-  Or(t9, zero_reg, 0x7FFFFFFF);
-  mtc1(t9, f20);
+  // Load 2^31 into f20 as its float representation.
+  li(at, 0x41E00000);
+  mtc1(at, FPURegister::from_code(scratch.code() + 1));
+  mtc1(zero_reg, scratch);
+  // Add it to fd.
+  add_d(fd, fd, scratch);
 
-  // Convert it to FP and add it to fd.
-  cvt_d_w(f20, f20);
-  add_d(fd, fd, f20);
-  // Now add 1.
-  Or(t9, zero_reg, 1);
-  mtc1(t9, f20);
-
-  cvt_d_w(f20, f20);
-  add_d(fd, fd, f20);
   bind(&conversion_done);
 }
 
 
-void MacroAssembler::Trunc_uw_d(FPURegister fd, FPURegister fs) {
-  Trunc_uw_d(fs, t4);
-  mtc1(t4, fd);
+void MacroAssembler::Trunc_uw_d(FPURegister fd,
+                                FPURegister fs,
+                                FPURegister scratch) {
+  Trunc_uw_d(fs, t8, scratch);
+  mtc1(t8, fd);
 }
 
 
-void MacroAssembler::Trunc_uw_d(FPURegister fd, Register rs) {
-  ASSERT(!fd.is(f22));
-  ASSERT(!rs.is(t8));
+void MacroAssembler::Trunc_uw_d(FPURegister fd,
+                                Register rs,
+                                FPURegister scratch) {
+  ASSERT(!fd.is(scratch));
+  ASSERT(!rs.is(at));
 
-  // Load 2^31 into f22.
-  Or(t8, zero_reg, 0x80000000);
-  Cvt_d_uw(f22, t8);
-
-  // Test if f22 > fd.
-  c(OLT, D, fd, f22);
+  // Load 2^31 into scratch as its float representation.
+  li(at, 0x41E00000);
+  mtc1(at, FPURegister::from_code(scratch.code() + 1));
+  mtc1(zero_reg, scratch);
+  // Test if scratch > fd.
+  c(OLT, D, fd, scratch);
 
   Label simple_convert;
   // If fd < 2^31 we can convert it normally.
   bc1t(&simple_convert);
 
   // First we subtract 2^31 from fd, then trunc it to rs
   // and add 2^31 to rs.
-
-  sub_d(f22, fd, f22);
-  trunc_w_d(f22, f22);
-  mfc1(rs, f22);
-  or_(rs, rs, t8);
+  sub_d(scratch, fd, scratch);
+  trunc_w_d(scratch, scratch);
+  mfc1(rs, scratch);
+  Or(rs, rs, 1 << 31);
 
   Label done;
   Branch(&done);
   // Simple conversion.
   bind(&simple_convert);
-  trunc_w_d(f22, fd);
-  mfc1(rs, f22);
+  trunc_w_d(scratch, fd);
+  mfc1(rs, scratch);
 
   bind(&done);
 }
 
 
 // Tries to get a signed int32 out of a double precision floating point heap
 // number. Rounds towards 0. Branch to 'not_int32' if the double is out of the
 // 32bits signed integer range.
 // This method implementation differs from the ARM version for performance
 // reasons.
@@ skipping 3385 matching lines @@
        opcode == BGTZL);
   opcode = (cond == eq) ? BEQ : BNE;
   instr = (instr & ~kOpcodeMask) | opcode;
   masm_.emit(instr);
 }
 
 
 } }  // namespace v8::internal
 
 #endif  // V8_TARGET_ARCH_MIPS