[Mesa-dev] [PATCH v2] mesa/st: translate SO info in glsl_to_nir() case

[Karol Herbst]

From: Rob Clark <robdclark at gmail.com>

This was handled for VS, but not for GS.

Fixes for gallium drivers using nir:
spec at arb_gpu_shader5@arb_gpu_shader5-xfb-streams-without-invocations
spec at arb_gpu_shader5@arb_gpu_shader5-xfb-streams*
spec at arb_transform_feedback3@arb_transform_feedback3-ext_interleaved_two_bufs_gs*
spec at ext_transform_feedback@geometry-shaders-basic
spec at ext_transform_feedback@* use_gs
spec at glsl-1.50@execution at geometry@primitive-id*
spec at glsl-1.50@execution at geometry@tri-strip-ordering-with-prim-restart gl_triangle_strip *
spec at glsl-1.50@transform-feedback-builtins
spec at glsl-1.50@transform-feedback-type-and-size

v2: we don't need this for TCP

Signed-off-by: Rob Clark <robdclark at gmail.com>
---
 src/mesa/state_tracker/st_program.c | 57 ++++++++++++++++++++++++++++++++++---
 1 file changed, 53 insertions(+), 4 deletions(-)

diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index 05e6042f42..2a1a695948 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -1421,6 +1421,50 @@ st_translate_program_common(struct st_context *st,
    }
 }
 
+/**
+ * Update stream-output info for GS/TCS/TES.  Normally this is done in
+ * st_translate_program_common() but that is not called for glsl_to_nir
+ * case.
+ */
+static void
+st_translate_program_stream_output(struct gl_program *prog,
+                                   struct pipe_stream_output_info *stream_output)
+{
+   if (!prog->sh.LinkedTransformFeedback)
+      return;
+
+   ubyte outputMapping[VARYING_SLOT_TESS_MAX];
+   GLuint attr;
+   uint num_outputs = 0;
+
+   memset(outputMapping, 0, sizeof(outputMapping));
+
+   /*
+    * Determine number of outputs, the (default) output register
+    * mapping and the semantic information for each output.
+    */
+   for (attr = 0; attr < VARYING_SLOT_MAX; attr++) {
+      if (prog->info.outputs_written & BITFIELD64_BIT(attr)) {
+         GLuint slot = num_outputs++;
+
+         outputMapping[attr] = slot;
+      }
+   }
+
+   /* Also add patch outputs. */
+   for (attr = 0; attr < 32; attr++) {
+      if (prog->info.patch_outputs_written & (1u << attr)) {
+         GLuint slot = num_outputs++;
+         GLuint patch_attr = VARYING_SLOT_PATCH0 + attr;
+
+         outputMapping[patch_attr] = slot;
+      }
+   }
+
+   st_translate_stream_output_info2(prog->sh.LinkedTransformFeedback,
+                                    outputMapping,
+                                    stream_output);
+}
 
 /**
  * Translate a geometry program to create a new variant.
@@ -1432,8 +1476,10 @@ st_translate_geometry_program(struct st_context *st,
    struct ureg_program *ureg;
 
    /* We have already compiled to NIR so just return */
-   if (stgp->shader_program)
+   if (stgp->shader_program) {
+      st_translate_program_stream_output(&stgp->Base, &stgp->tgsi.stream_output);
       return true;
+   }
 
    ureg = ureg_create_with_screen(PIPE_SHADER_GEOMETRY, st->pipe->screen);
    if (ureg == NULL)
@@ -1489,6 +1535,7 @@ st_get_basic_variant(struct st_context *st,
 	    tgsi.ir.nir = nir_shader_clone(NULL, prog->tgsi.ir.nir);
 	    st_finalize_nir(st, &prog->Base, prog->shader_program,
                             tgsi.ir.nir);
+            tgsi.stream_output = prog->tgsi.stream_output;
 	 } else
 	    tgsi = prog->tgsi;
          /* fill in new variant */
@@ -1529,7 +1576,7 @@ st_translate_tessctrl_program(struct st_context *st,
 {
    struct ureg_program *ureg;
 
-   /* We have already compiler to NIR so just return */
+   /* We have already compiled to NIR so just return */
    if (sttcp->shader_program)
       return true;
 
@@ -1558,9 +1605,11 @@ st_translate_tesseval_program(struct st_context *st,
 {
    struct ureg_program *ureg;
 
-   /* We have already compiler to NIR so just return */
-   if (sttep->shader_program)
+   /* We have already compiled to NIR so just return */
+   if (sttep->shader_program) {
+      st_translate_program_stream_output(&sttep->Base, &sttep->tgsi.stream_output);
       return true;
+   }
 
    ureg = ureg_create_with_screen(PIPE_SHADER_TESS_EVAL, st->pipe->screen);
    if (ureg == NULL)
-- 
2.14.3