<div dir="ltr"> <div class="gmail_extra"> <div class="gmail_quote">On Tue, Apr 19, 2016 at 10:15 AM, Rob Clark <<a href="mailto:robdclark@gmail.com" target="_blank">robdclark@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div class="HOEnZb"><div class="h5">On Tue, Apr 19, 2016 at 1:06 PM, Jason Ekstrand <<a href="mailto:jason@jlekstrand.net">jason@jlekstrand.net</a>> wrote: > > > On Tue, Apr 19, 2016 at 6:58 AM, Rob Clark <<a href="mailto:robdclark@gmail.com">robdclark@gmail.com</a>> wrote: >> >> From: Rob Clark <<a href="mailto:robclark@freedesktop.org">robclark@freedesktop.org</a>> >> >> Signed-off-by: Rob Clark <<a href="mailto:robclark@freedesktop.org">robclark@freedesktop.org</a>> >> --- >> src/compiler/nir/nir.h | 7 ++++++ >> src/compiler/nir/nir_lower_tex.c | 46 >> ++++++++++++++++++++++++++++++++++++++++ >> 2 files changed, 53 insertions(+) >> >> diff --git a/src/compiler/nir/nir.h b/src/compiler/nir/nir.h >> index bbbc208..2c0f705 100644 >> --- a/src/compiler/nir/nir.h >> +++ b/src/compiler/nir/nir.h >> @@ -2301,6 +2301,13 @@ typedef struct nir_lower_tex_options { >> * while 4 and 5 represent 0 and 1 respectively. >> */ >> uint8_t swizzles[32][4]; >> + >> + /** >> + * Bitmap of textures that need srgb to linear conversion. If >> + * (srgb_to_linear & (1 << texture_index)) then the rgb (xyz) >> + * components of the texture are lowered to linear. >> + */ >> + unsigned lower_srgb; > > > You have lower_srgb here and srgb_to_linear in the comment. Please pick > one. I think I prefer srgb_to_linear as it's more explicit. </div></div>oh, forgot to update the comment.. started as srgb_to_linear until I realized that lower_srgb was same # of chars as saturate_s/t/r and made a bunch of code line up neatly (ok, perhaps not the best reason.. but a reason.. :-P) <div><div class="h5"></div></div></blockquote><div> </div><div>Wow, that's sad... I guess I don't care too much beyond having it well documented, but I do thing srgb_to_linear is more self-documenting. Do what you'd like. </div><div> </div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div><div class="h5"> >> >> } nir_lower_tex_options; >> >> bool nir_lower_tex(nir_shader *shader, >> diff --git a/src/compiler/nir/nir_lower_tex.c >> b/src/compiler/nir/nir_lower_tex.c >> index 7740e58..6bae3b7 100644 >> --- a/src/compiler/nir/nir_lower_tex.c >> +++ b/src/compiler/nir/nir_lower_tex.c >> @@ -275,6 +275,45 @@ swizzle_result(nir_builder *b, nir_tex_instr *tex, >> const uint8_t swizzle[4]) >> swizzled->parent_instr); >> } >> >> +static void >> +linearize_srgb_result(nir_builder *b, nir_tex_instr *tex) >> +{ >> + assert(tex->dest.is_ssa); >> + assert(nir_tex_instr_dest_size(tex) == 4); >> + assert(nir_alu_type_get_base_type(tex->dest_type) == nir_type_float); >> + >> + b->cursor = nir_after_instr(&tex->instr); >> + >> + nir_ssa_def *components[4]; >> + >> + /* first three channels are rgb: */ >> + for (unsigned i = 0; i < 3; i++) { > > > Um... NIR is a vector IR. Why are we looping over channels? Just do a > nir_swizzle to pick off the first three, do the sRGB conversion, and nir_vec > them back together with the alpha channel. </div></div>oh, hmm, yeah.. used to thinking in scalar BR, -R <div class="HOEnZb"><div class="h5"> >> >> + /* Formula is: >> + * (comp <= 0.04045) ? >> + * (comp / 12.92) : >> + * pow((comp + 0.055) / 1.055, 2.4) >> + */ >> + nir_ssa_def *comp = nir_channel(b, &tex->dest.ssa, i); >> + nir_ssa_def *low = nir_fmul(b, comp, nir_imm_float(b, 1.0 / >> 12.92)); >> >> + nir_ssa_def *high = nir_fpow(b, >> + nir_fmul(b, >> + nir_fadd(b, >> + comp, >> + nir_imm_float(b, >> 0.055)), >> + nir_imm_float(b, 1.0 / >> 1.055)), >> + nir_imm_float(b, 2.4)); >> + nir_ssa_def *cond = nir_fge(b, nir_imm_float(b, 0.04045), comp); >> + components[i] = nir_bcsel(b, cond, low, high); >> + } >> + >> + /* alpha is untouched: */ >> + components[3] = nir_channel(b, &tex->dest.ssa, 3); >> + >> + nir_ssa_def *linearized = nir_vec(b, components, 4); >> + nir_ssa_def_rewrite_uses_after(&tex->dest.ssa, >> nir_src_for_ssa(linearized), >> + linearized->parent_instr); >> +} >> + >> static bool >> nir_lower_tex_block(nir_block *block, void *void_state) >> { >> @@ -323,6 +362,13 @@ nir_lower_tex_block(nir_block *block, void >> *void_state) >> swizzle_result(b, tex, options->swizzles[tex->texture_index]); >> state->progress = true; >> } >> + >> + /* should be after swizzle so we know which channels are rgb: */ >> + if (((1 << tex->texture_index) & options->lower_srgb) && >> + !nir_tex_instr_is_query(tex) && !tex->is_shadow) { >> + linearize_srgb_result(b, tex); >> + state->progress = true; >> + } >> } >> >> return true; >> -- >> 2.5.5 >> > </div></div></blockquote></div> </div></div>