Connor Abbott <cwabbo...@gmail.com> writes: > On Fri, Jan 23, 2015 at 5:34 PM, Eric Anholt <e...@anholt.net> wrote: >> Connor Abbott <cwabbo...@gmail.com> writes: >> >>> Argh, nevermind, I was reading it wrong... >>> >>> On Thu, Jan 22, 2015 at 8:18 PM, Connor Abbott <cwabbo...@gmail.com> wrote: >>>> What happens if you have something like foo = vec3(foo.z, bar.x, >>>> foo.x)? I don't think emitting vector mov's for only the contiguous >>>> components is enough. >>>> >>>> On Thu, Jan 22, 2015 at 4:51 PM, Eric Anholt <e...@anholt.net> wrote: >>>>> +static unsigned >>>>> +insert_movs(nir_alu_instr *vec, unsigned start_channel, >>>>> + unsigned start_src_idx, void *mem_ctx) >>> >>> We need a comment explaining what this function does and what it >>> returns. Also, it only creates a single move so it should be called >>> insert_mov(). >> >> How about: >> >> /** >> * For a given writemask channel in the vec instruction, insert a MOV of all >> * the src values that come from the same reg to the destination of the vec >> * instruction. >> */ > > First, I think we need to be a little more clear on what it does, e.g. > "for a given starting channel/source, scans the rest of the sources of > the vec instruction to find all sources that come from the same reg > and emits a writemasked MOV from the reg to the destination of the vec > instruction." Also, you need to explain what the return value is (a > mask of all the channels/sources we inserted a MOV for).
Maybe: /** * For a given starting writemask channel and corresponding source index in * the vec instruction, insert a MOV to the vec instruction's dest of all the * writemask channels that get read from the same src reg. * * Returns the writemask of our MOV, so the parent loop calling this knows * which ones have been processed. */
signature.asc
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
