austriancoder changed the topic of #etnaviv to: #etnaviv - the home of the reverse-engineered Vivante GPU driver - Logs https://freenode.irclog.whitequark.org/etnaviv
<marex> austriancoder: I think I've got it ... the TXP in TGSI divides the texture xyz by w, if I remove the MUL/RCP then the pointsprite example magically works, and so does neverball
<marex> (i.e. if I make TXP behave as TGSI_OPCODE_TEX)
srk has quit [Ping timeout: 240 seconds]
srk has joined #etnaviv
srk has quit [Ping timeout: 240 seconds]
srk has joined #etnaviv
pcercuei has joined #etnaviv
<marex> austriancoder: any idea why disabling that RCP/MUL in TXP would help with this texcoord ?
<marex> mntmn: uh, did your glClear problem ever get solved ?
<austriancoder> marex: no idea.. I can start looking into the pointsprite myself if that would help you
<marex> austriancoder: well yes and no
<marex> austriancoder: I am exploring the boundaries of my frustration, but that little progress with TXP this morning helped a bit
<marex> austriancoder: I wonder whether the fixed pipeline to glsl part of mesa isn't the problem
<marex> maybe it shouldn't generate TXP for GLES1 in the first place
<austriancoder> marex: okay.. then I will NOT look into that problem and continue with other stuff
<marex> austriancoder: but I could use the possibility to ask you questions
<austriancoder> marex: sure.. but I might not have answers
<marex> austriancoder: you probably have more answers than I have questions
<marex> austriancoder: Im now trying to figure out how to dump the IR in src/mesa/main/ff_fragment_shader.cpp
<marex> austriancoder: basically how to get the generated VS source before it is compiled
<austriancoder> marex: there is an env var for that
<mntmn> marex: yes, glClear got solved by something unexpected
<austriancoder> marex: should be MESA_GLSL=dump
<marex> austriancoder: ah, thanks
<mntmn> this patch fixes the glClear corruption for me
<marex> (assign (xyz) (var_ref compiler_temp) (swiz xyz (expression vec4 * (tex vec4 (var_ref sampler_0) (swiz xy (var_ref gl_in_TexCoord0) ) 0 (swiz w (var_ref gl_in_TexCoord0) ) () )(var_ref gl_Color) ) ))
<marex> well ... that swiz w shouldn't be there I think
<marex> oh, I think I've got it
<marex> mntmn: that should "fix" the star texture issue
<marex> austriancoder: ^ the last change to ff_fragment_shader.cpp seems to do the trick
<marex> but now to figure out how to fix it properly
<marex> mntmn: ^
<marex> flto: ^
srk has quit [Read error: Connection reset by peer]
srk has joined #etnaviv
<mntmn> marex: mind that i'm not using GLES1 but desktop gl version of neverball
<mntmn> marex: huh, does the shader compiler use s-expressions? :3
<marex> mntmn: it should work either way
<marex> I think we need to handle sampler state normalized_coords better , maybe
<mntmn> marex: i can try it out soon!
<mntmn> thanks for the heads up
<marex> I need a break before my head explodes from all this
pcercuei_ has joined #etnaviv
pcercuei has quit [Ping timeout: 240 seconds]
<marex> yep, works even on the mp1
cphealy has quit [Remote host closed the connection]
pcercuei_ is now known as pcercuei
cphealy has joined #etnaviv
<mntmn> marex: i'll try your patch now to see if it makes neverball look different
<mntmn> marex: big success, neverball suddenly has a lot of star sprites!
<mntmn> on imx8m
<marex> mntmn: awesome
<cphealy> marex: Nice work, up until this point in etnaviv history, I've never seen Neverball star sprites work!
<marex> cphealy: the patch is obviously wrong though
<mntmn> marex: what's wrong with it?
<marex> mntmn: hacking common code for one thing ?
<mntmn> well, ok
<marex> I think it has to do with sampler_state normalized_coords
JohnnyonFlame has joined #etnaviv
pcercuei has quit [Quit: dodo]