​MIPS: Improve Cvt_d_uw on mips32.

src/mips/macro-assembler-mips.cc

 
 // Copyright 2012 the V8 project authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include <limits.h>  // For LONG_MIN, LONG_MAX.
 
 #if V8_TARGET_ARCH_MIPS
 
 #include "src/base/bits.h"
@@ skipping 1249 matching lines @@
     and_(t8, rs, at);
     sll(t8, t8, pos);
     sll(at, at, pos);
     nor(at, at, zero_reg);
     and_(at, rt, at);
     or_(rt, t8, at);
   }
 }
 
 
-void MacroAssembler::Cvt_d_uw(FPURegister fd,
-                              FPURegister fs,
+void MacroAssembler::Cvt_d_uw(FPURegister fd, Register rs,
                               FPURegister scratch) {
-  // Move the data from fs to t8.
-  mfc1(t8, fs);
-  Cvt_d_uw(fd, t8, scratch);
+  // In FP64Mode we do convertion from long.
+  if (IsFp64Mode()) {
+    mtc1(rs, scratch);
+    cvt_d_l(fd, scratch);
+  } else {
+    // Convert rs to a FP value in fd.
+    DCHECK(!fd.is(scratch));
+    DCHECK(!rs.is(t9));

[ivica.bogosavljevic, 2015/11/23 08:56:45]

On one hand, I wished you weren't using register t9 in macro assembler, since
this ought to be available to others
On the other hand, I know that t9 is used in MacroAssembler elsewhere. You added
DCHECKs here, so it's ok.

If you could rewrite this without using t9, that would be nice, if not it's ok
like this too.

[paul.l..., 2015/11/23 15:57:40]

Well, I was going to point out that t9 (and t8) are reserved for macro-assembler
(https://chromium.googlesource.com/v8/v8/+/1d1557d8fded51c323e01b4ca5e36be2861...)
so the usage is OK.

But as I look at the code, it no longer uses t9, so Alan, just delete this
DCHECK() and everyone will be happy!

+    DCHECK(!rs.is(at));
+
+    Label msb_clear, conversion_done;
+    // For a value which is < 2^31, regard it as a signed positve word.
+    Branch(&msb_clear, ge, rs, Operand(zero_reg), USE_DELAY_SLOT);
+    mtc1(rs, fd);
+
+    lui(at, 0x41F0);  // FP value: 2^32.
+
+    // For unsigned inputs > 2^31, we convert to double as a signed int32,
+    // then add 2^32 to move it back to unsigned value in range 2^31..2^31-1.
+    mtc1(zero_reg, scratch);
+    Mthc1(at, scratch);
+
+    cvt_d_w(fd, fd);
+
+    Branch(USE_DELAY_SLOT, &conversion_done);
+    add_d(fd, fd, scratch);
+
+    bind(&msb_clear);
+    cvt_d_w(fd, fd);
+
+    bind(&conversion_done);
+  }
 }
 
 
-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 to the result (if needed).
-
-  DCHECK(!fd.is(scratch));
-  DCHECK(!rs.is(t9));
-  DCHECK(!rs.is(at));
-
-  // Save rs's MSB to t9.
-  Ext(t9, rs, 31, 1);
-  // Remove rs's MSB.
-  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, t9, Operand(zero_reg));
-
-  // Load 2^31 into f20 as its float representation.
-  li(at, 0x41E00000);
-  mtc1(zero_reg, scratch);
-  Mthc1(at, scratch);
-  // Add it to fd.
-  add_d(fd, fd, scratch);
-
-  bind(&conversion_done);
-}
-
-
 void MacroAssembler::Trunc_uw_d(FPURegister fd,
                                 FPURegister fs,
                                 FPURegister scratch) {
   Trunc_uw_d(fs, t8, scratch);
   mtc1(t8, fd);
 }
 
 
 void MacroAssembler::Trunc_w_d(FPURegister fd, FPURegister fs) {
   if (IsMipsArchVariant(kLoongson) && fd.is(fs)) {
@@ skipping 4518 matching lines @@
   if (mag.shift > 0) sra(result, result, mag.shift);
   srl(at, dividend, 31);
   Addu(result, result, Operand(at));
 }
 
 
 }  // namespace internal
 }  // namespace v8
 
 #endif  // V8_TARGET_ARCH_MIPS