[virglrenderer-devel] [PATCH] shader: Do not adjust y coordonate if an application render to a FBO

Gurchetan Singh gurchetansingh at chromium.org
Mon Jul 9 15:28:48 UTC 2018


On Fri, Jul 6, 2018 at 1:51 AM Elie Tournier <tournier.elie at gmail.com> wrote:
>
> On Wed, Jun 27, 2018 at 07:03:03PM -0700, Gurchetan Singh wrote:
> > On Wed, Jun 27, 2018 at 4:18 PM Elie Tournier <tournier.elie at gmail.com> wrote:
> > >
> > > I almost forget this patch.
> > >
> > > On Fri, May 25, 2018 at 05:03:51PM -0700, Gurchetan Singh wrote:
> > > > On Fri, May 25, 2018 at 7:41 AM Elie Tournier <tournier.elie at gmail.com>
> > > > wrote:
> > > >
> > > > > On Wed, May 16, 2018 at 10:34:29PM +0100, Elie Tournier wrote:
> > > > > > Fixes: 7f615cd "shader: Invert y coordonate if on gles host"
> > > > >
> > > > > Oups. s/coordonate/coordinate/g
> > > > > Ping?
> > > > > >
> > > > > > Signed-off-by: Elie Tournier <elie.tournier at collabora.com>
> > > > > > ---
> > > > > >  src/vrend_renderer.c |  3 +++
> > > > > >  src/vrend_shader.c   | 10 +++++++---
> > > > > >  2 files changed, 10 insertions(+), 3 deletions(-)
> > > > > >
> > > > > > diff --git a/src/vrend_renderer.c b/src/vrend_renderer.c
> > > > > > index 55b96db..70ea0c5 100644
> > > > > > --- a/src/vrend_renderer.c
> > > > > > +++ b/src/vrend_renderer.c
> > > > > > @@ -158,6 +158,7 @@ struct vrend_linked_shader_program {
> > > > > >
> > > > > >     GLuint *ubo_locs[PIPE_SHADER_TYPES];
> > > > > >     GLuint vs_ws_adjust_loc;
> > > > > > +   GLuint vs_ws_adjust_loc_gles;
> > > > >
> > > >
> > > > nit: should be fs_ws_adjust_loc_gles, since we only emit in the fragment
> > > > shader.
> > > True
> > > >
> > > >
> > > > > >
> > > > > >     GLuint fs_stipple_loc;
> > > > > >
> > > > > > @@ -975,6 +976,7 @@ static struct vrend_linked_shader_program
> > > > > *add_shader_program(struct vrend_conte
> > > > > >     else
> > > > > >        sprog->fs_stipple_loc = -1;
> > > > > >     sprog->vs_ws_adjust_loc = glGetUniformLocation(prog_id,
> > > > > "winsys_adjust_y");
> > > > > > +   sprog->vs_ws_adjust_loc_gles = glGetUniformLocation(prog_id,
> > > > > "winsys_adjust_y_gles");
> > > > > >     for (id = PIPE_SHADER_VERTEX; id <= last_shader; id++) {
> > > > > >        if (sprog->ss[id]->sel->sinfo.samplers_used_mask) {
> > > > > >           uint32_t mask = sprog->ss[id]->sel->sinfo.samplers_used_mask;
> > > > > > @@ -3028,6 +3030,7 @@ void vrend_draw_vbo(struct vrend_context *ctx,
> > > > > >        return;
> > > > > >     }
> > > > > >     glUniform1f(ctx->sub->prog->vs_ws_adjust_loc,
> > > > > ctx->sub->viewport_is_negative ? -1.0 : 1.0);
> > > > > > +   glUniform1f(ctx->sub->prog->vs_ws_adjust_loc_gles,
> > > > > ctx->sub->viewport_is_negative ? 0.0 : 1.0);
> > > > >
> > > >
> > > > It looks like vs_ws_adjust_loc and vs_ws_adjust_loc_gles seem to cancel
> > > > each other out on GLES (correct me if I'm wrong).  Is it possible to just
> > > > modify vs_ws_adjust_loc depending on some relevant variables?  i.e,
> > > >
> > > > bool invert = ctx->sub->viewport_is_negative ? true : false;
> > > > invert = invert ^ (use_gles && state->sprite_coord_mode
> > > > == PIPE_SPRITE_COORD_LOWER_LEFT)
> > > We have one on vertex shader and the other one on the fragment shader
> > > so they don't cancel each other.
> > > >
> > > > >
> > > > > >     if (ctx->sub->rs_state.clip_plane_enable) {
> > > > > >        for (i = 0 ; i < 8; i++) {
> > > > > > diff --git a/src/vrend_shader.c b/src/vrend_shader.c
> > > > > > index 1ce18de..05eadd7 100644
> > > > > > --- a/src/vrend_shader.c
> > > > > > +++ b/src/vrend_shader.c
> > > > > > @@ -53,7 +53,7 @@ struct vrend_shader_io {
> > > > > >     bool glsl_gl_in;
> > > > > >     bool override_no_wm;
> > > > > >     bool is_int;
> > > > > > -   char glsl_name[64];
> > > > > > +   char glsl_name[128];
> > > > >
> > > >
> > > > This needs a minor rebase, with the arb_gpu_shader5 having landed.
> > > Will do in the future patch.
> > > >
> > > >
> > > > > >  };
> > > > > >
> > > > > >  struct vrend_shader_sampler {
> > > > > > @@ -416,7 +416,7 @@ iter_declaration(struct tgsi_iterate_context *iter,
> > > > > >           if (iter->processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
> > > > > >              if (ctx->key->coord_replace & (1 << ctx->inputs[i].sid)) {
> > > > > >                 if (ctx->cfg->use_gles)
> > > > > > -                  name_prefix = "vec4(gl_PointCoord.x, 1.0 -
> > > > > gl_PointCoord.y, 0.0, 1.0)";
> > > > > > +                  name_prefix = "vec4(gl_PointCoord.x, mix(1.0 -
> > > > > gl_PointCoord.y, gl_PointCoord.y, winsys_adjust_y_gles), 0.0, 1.0)";
> > > If you prefer, we can use only one uniform winsys_adjust_y and clamp it like:
> > > name_prefix = "vec4(gl_PointCoord.x, mix(1.0 - gl_PointCoord.y, gl_PointCoord.y,
> > > clamp(winsys_adjust_y, 0.0, 1.0), 0.0, 1.0)";
> > >
> > > But I'm not a big fan of this solution. The expression is more complex and I
> > > think we have enough memory to store an other uniform.
> >
> > I prefer the one uniform approach.  Also, we don't have to emit the
> > uniform every time in the fragment shader -- only when we have point
> > primitives.
>
> I didn't find an easy way to know when we have point primitives.
> Did you have something in mind?

You could just keep some variable that emits winsys_adjust_y in the
fragment shader, which you set when ctx->key->coord_replace & (1 <<
ctx->inputs[i].sid).

BTW, which test(s) does this patch fix?

> >
> > > > > >                 else
> > > > > >                    name_prefix = "vec4(gl_PointCoord, 0.0, 1.0)";
> > > > > >                 ctx->inputs[i].glsl_predefined_no_emit = true;
> > > > > > @@ -444,7 +444,7 @@ iter_declaration(struct tgsi_iterate_context *iter,
> > > > > >        }
> > > > > >
> > > > > >        if (ctx->inputs[i].glsl_no_index)
> > > > > > -         snprintf(ctx->inputs[i].glsl_name, 64, "%s", name_prefix);
> > > > > > +         snprintf(ctx->inputs[i].glsl_name, 128, "%s", name_prefix);
> > > > > >        else {
> > > > > >           if (ctx->inputs[i].name == TGSI_SEMANTIC_FOG)
> > > > > >              snprintf(ctx->inputs[i].glsl_name, 64, "%s_f%d",
> > > > > name_prefix, ctx->inputs[i].sid);
> > > > > > @@ -2439,6 +2439,10 @@ static char *emit_ios(struct dump_ctx *ctx, char
> > > > > *glsl_hdr)
> > > > > >        bcolor_emitted[0] = bcolor_emitted[1] = false;
> > > > > >     }
> > > > > >     if (ctx->prog_type == TGSI_PROCESSOR_FRAGMENT) {
> > > > > > +      if (ctx->cfg->use_gles) {
> > > > > > +         snprintf(buf, 255, "uniform float winsys_adjust_y_gles;\n");
> > > > > > +         STRCAT_WITH_RET(glsl_hdr, buf);
> > > > > > +      }
> > > > > >        if (fs_emit_layout(ctx)) {
> > > > > >           bool upper_left = !(ctx->fs_coord_origin ^
> > > > > ctx->key->invert_fs_origin);
> > > > > >           char comma = (upper_left && ctx->fs_pixel_center) ? ',' : ' ';
> > > > > > --
> > > > > > 2.17.0
> > > > > >
> > > > > _______________________________________________
> > > > > virglrenderer-devel mailing list
> > > > > virglrenderer-devel at lists.freedesktop.org
> > > > > https://lists.freedesktop.org/mailman/listinfo/virglrenderer-devel
> > > > >


More information about the virglrenderer-devel mailing list