LGTM
http://codereview.chromium.org/2813046/diff/1/2 File src/arm/assembler-arm.cc (right): http://codereview.chromium.org/2813046/diff/1/2#newcode1804 src/arm/assembler-arm.cc:1804: static uint64_t DoubleAsInt(double d) { Please add a comment here: // Only works for little endian floating point formats. // We don't support VFP on the mixed endian floating point platform. Also, both places where you use this you immediately separate out the low and high bits. Perhaps it should take two uint32_t*s and put the answer there. http://codereview.chromium.org/2813046/diff/1/2#newcode1847 src/arm/assembler-arm.cc:1847: if (((hi ^ (hi << 1)) & (1 << 30)) == 0) { It would be more consistent to use 0x40000000 here instead of 1 << 30. http://codereview.chromium.org/2813046/diff/1/2#newcode1886 src/arm/assembler-arm.cc:1886: // registers of D register dst. Would it be faster to take a scratch register as an argument and use that with ip to do the move in one? http://codereview.chromium.org/2813046/diff/1/2#newcode1894 src/arm/assembler-arm.cc:1894: stemp.code_ = (dst.code_ / 2) + 1; Please introduce a new variable here instead of updating the one you have. That's confusing. Even better would be to add a couple of methods to DwVfpRegister that return the low and high half equivalent registers. http://codereview.chromium.org/2813046/diff/1/5 File src/arm/disasm-arm.cc (right): http://codereview.chromium.org/2813046/diff/1/5#newcode440 src/arm/disasm-arm.cc:440: "#%d", imm); This isn't very helpful. Can we output the floating point number instead of its encoding in the instruction or would that violate some ARM disassembly standard? http://codereview.chromium.org/2813046/show -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev
