panfrost: Precompile transform feedback program

This avoids the weird compiled_shader pointer inside of compiled_shader. Because
we don't have a nonempty vertex shader key, there will only ever be a single
transform feedback program per CSO.

Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19363>

3 files changed, 40 insertions, 30 deletions

diff --git a/src/gallium/drivers/panfrost/pan_cmdstream.c b/src/gallium/drivers/panfrost/pan_cmdstream.c
index 3bb6146ba69..1782ca17ce9 100644
--- a/src/gallium/drivers/panfrost/pan_cmdstream.c
+++ b/src/gallium/drivers/panfrost/pan_cmdstream.c
@@ -2018,8 +2018,11 @@ panfrost_emit_vertex_data(struct panfrost_batch *batch,
 
         unsigned count = vs->info.attribute_count;
 
-        if (vs->xfb)
-                count = MAX2(count, vs->xfb->info.attribute_count);
+        struct panfrost_compiled_shader *xfb =
+                ctx->uncompiled[PIPE_SHADER_VERTEX]->xfb;
+
+        if (xfb)
+                count = MAX2(count, xfb->info.attribute_count);
 
 #if PAN_ARCH <= 5
         /* Midgard needs vertexid/instanceid handled specially */
@@ -3546,14 +3549,14 @@ panfrost_launch_xfb(struct panfrost_batch *batch,
         struct panfrost_uncompiled_shader *vs_uncompiled = ctx->uncompiled[PIPE_SHADER_VERTEX];
         struct panfrost_compiled_shader *vs = ctx->prog[PIPE_SHADER_VERTEX];
 
-        vs->xfb->stream_output = vs->stream_output;
+        vs_uncompiled->xfb->stream_output = vs->stream_output;
 
         mali_ptr saved_rsd = batch->rsd[PIPE_SHADER_VERTEX];
         mali_ptr saved_ubo = batch->uniform_buffers[PIPE_SHADER_VERTEX];
         mali_ptr saved_push = batch->push_uniforms[PIPE_SHADER_VERTEX];
 
         ctx->uncompiled[PIPE_SHADER_VERTEX] = NULL; /* should not be read */
-        ctx->prog[PIPE_SHADER_VERTEX] = vs->xfb;
+        ctx->prog[PIPE_SHADER_VERTEX] = vs_uncompiled->xfb;
         batch->rsd[PIPE_SHADER_VERTEX] = panfrost_emit_compute_shader_meta(batch, PIPE_SHADER_VERTEX);
 
 #if PAN_ARCH >= 9
@@ -3734,7 +3737,7 @@ panfrost_direct_draw(struct panfrost_batch *batch,
         panfrost_update_shader_state(batch, PIPE_SHADER_FRAGMENT);
         panfrost_clean_state_3d(ctx);
 
-        if (vs->xfb) {
+        if (ctx->uncompiled[PIPE_SHADER_VERTEX]->xfb) {
 #if PAN_ARCH >= 9
                 mali_ptr attribs = 0, attrib_bufs = 0;
 #endif
diff --git a/src/gallium/drivers/panfrost/pan_context.h b/src/gallium/drivers/panfrost/pan_context.h
index 86cf7125ae9..65a78b6c4d4 100644
--- a/src/gallium/drivers/panfrost/pan_context.h
+++ b/src/gallium/drivers/panfrost/pan_context.h
@@ -294,9 +294,6 @@ struct panfrost_compiled_shader {
 
         struct pan_earlyzs_lut earlyzs;
 
-        /* Attached transform feedback program, if one exists */
-        struct panfrost_compiled_shader *xfb;
-
         /* Linked varyings, for non-separable programs */
         struct pan_linkage linkage;
 
@@ -322,6 +319,9 @@ struct panfrost_uncompiled_shader {
         /* Array of panfrost_compiled_shader */
         struct util_dynarray variants;
 
+        /* Compiled transform feedback program, if one is required */
+        struct panfrost_compiled_shader *xfb;
+
         /* On vertex shaders, bit mask of special desktop-only varyings to link
          * with the fragment shader. Used on Valhall to implement separable
          * shaders for desktop GL.
diff --git a/src/gallium/drivers/panfrost/pan_shader.c b/src/gallium/drivers/panfrost/pan_shader.c
index 7ed210e9e46..af4a53b5a19 100644
--- a/src/gallium/drivers/panfrost/pan_shader.c
+++ b/src/gallium/drivers/panfrost/pan_shader.c
@@ -68,19 +68,6 @@ panfrost_shader_compile(struct pipe_screen *pscreen,
 
         nir_shader *s = nir_shader_clone(NULL, ir);
 
-        if (s->xfb_info && !s->info.internal) {
-                /* Create compute shader doing transform feedback */
-                nir_shader *xfb = nir_shader_clone(NULL, s);
-                xfb->info.name = ralloc_asprintf(xfb, "%s@xfb", xfb->info.name);
-                xfb->info.internal = true;
-
-                state->xfb = calloc(1, sizeof(struct panfrost_compiled_shader));
-                panfrost_shader_compile(pscreen, shader_pool, desc_pool, xfb, dbg, state->xfb, 0, fixed_varying_mask);
-
-                /* Main shader no longer uses XFB */
-                s->info.has_transform_feedback_varyings = false;
-        }
-
         struct panfrost_compile_inputs inputs = {
                 .debug = dbg,
                 .gpu_id = dev->gpu_id,
@@ -357,15 +344,35 @@ panfrost_create_shader_state(
                         ~VARYING_BIT_POS & ~VARYING_BIT_PSIZ;
         }
 
+        /* If this shader uses transform feedback, compile the transform
+         * feedback program. This is a special shader variant.
+         */
+        struct panfrost_context *ctx = pan_context(pctx);
+        struct util_debug_callback *dbg = &ctx->base.debug;
+
+        if (so->nir->xfb_info) {
+                nir_shader *xfb = nir_shader_clone(NULL, so->nir);
+                xfb->info.name = ralloc_asprintf(xfb, "%s@xfb", xfb->info.name);
+                xfb->info.internal = true;
+
+                so->xfb = calloc(1, sizeof(struct panfrost_compiled_shader));
+                panfrost_shader_compile(pctx->screen, &ctx->shaders,
+                                        &ctx->descs, xfb, dbg, so->xfb, 0,
+                                        so->fixed_varying_mask);
+
+                /* Since transform feedback is handled via the transform
+                 * feedback program, the original program no longer uses XFB
+                 */
+                so->nir->info.has_transform_feedback_varyings = false;
+        }
+
         /* Precompile for shader-db if we need to */
         if (unlikely(dev->debug & PAN_DBG_PRECOMPILE)) {
-                struct panfrost_context *ctx = pan_context(pctx);
-
                 struct panfrost_compiled_shader state = { 0 };
 
                 panfrost_shader_compile(pctx->screen,
                                         &ctx->shaders, &ctx->descs,
-                                        so->nir, &ctx->base.debug, &state, 0,
+                                        so->nir, dbg, &state, 0,
                                         so->fixed_varying_mask);
         }
 
@@ -385,13 +392,13 @@ panfrost_delete_shader_state(
                 panfrost_bo_unreference(so->bin.bo);
                 panfrost_bo_unreference(so->state.bo);
                 panfrost_bo_unreference(so->linkage.bo);
+        }
 
-                if (so->xfb) {
-                        panfrost_bo_unreference(so->xfb->bin.bo);
-                        panfrost_bo_unreference(so->xfb->state.bo);
-                        panfrost_bo_unreference(so->xfb->linkage.bo);
-                        free(so->xfb);
-                }
+        if (cso->xfb) {
+                panfrost_bo_unreference(cso->xfb->bin.bo);
+                panfrost_bo_unreference(cso->xfb->state.bo);
+                panfrost_bo_unreference(cso->xfb->linkage.bo);
+                free(cso->xfb);
         }
 
         simple_mtx_destroy(&cso->lock);