Re: [patch 1/2] AMD bdver2 processors - BMI

[Richard Henderson <rth at redhat dot com>]

On 11/02/2010 05:27 AM, Quentin Neill wrote:
> Actually BSF indexes the lowest set bit, TZCNT counts zeros, so if I'm
> thinking clearly BSF(X)==TZCNT(X)+1

Nope.  Looking carefully the differences are:

		BSF			TZCNT
Result for 0	Undefined		Operand-Width
ZF		Set iff 0 input		Set iff 0 output
CF		Undefined		Set iff 0 input

otherwise the actual value result is identical.


>> /* The value at zero is only defined for the BMI instruction
>>   TZCNT, not the BSF insn in the original isa.  */
>> #define CTZ_DEFINED_VALUE_AT_ZERO(MODE, VALUE) \
>>  ((VALUE) = GET_MODE_BITSIZE (MODE), TARGET_BMI)
> 
> I take it his goes in defaults.h?  How do I test this?

This goes in i386.h.  Test this by seeing that __builtin_ctz
with a constant operand 0 gets folded properly.  Note that 
there's a CLZ_DEFINED_VALUE_AT_ZERO as well, which would be
used for the LZCNT instruction.

>> Note that this has already been done for lzcnt as present
>> in the ABM extension.
> 
> I believe the LZCNT instruction is not a part of the BMI extension,
> I've removed it from the list of mnemonics.

It is according rev 2.06 of the manual you gave me.

# Support for the LZCNT instruction is indicated by ECX bit 5 (LZCNT) as returned by CPUID
# function 8000_0001h or EBX bit 3 (BMI) as returned by CPUID function 0000_0007h.

While this manual appears to have renamed the bit, gcc's cpuid.h
records 0x80000001 bit 5 as ABM, so it *is* the same instruction.
You simply need to adjust the conditions under which we generate it.


r~