v3d: Add SSBO/atomic counters support.

So far I assume that all the buffers get written.  If they weren't, you'd
probably be using UBOs instead.

6 files changed, 102 insertions, 1 deletions

diff --git a/src/gallium/drivers/v3d/v3d_context.c b/src/gallium/drivers/v3d/v3d_context.c
index b9eaf7e67ec..104096d5248 100644
--- a/src/gallium/drivers/v3d/v3d_context.c
+++ b/src/gallium/drivers/v3d/v3d_context.c
@@ -66,6 +66,16 @@ v3d_pipe_flush(struct pipe_context *pctx, struct pipe_fence_handle **fence,
 }
 
 static void
+v3d_memory_barrier(struct pipe_context *pctx, unsigned int flags)
+{
+        struct v3d_context *v3d = v3d_context(pctx);
+
+        /* We only need to flush jobs writing to SSBOs/images. */
+        perf_debug("Flushing all jobs for glMemoryBarrier(), could do better");
+        v3d_flush(pctx);
+}
+
+static void
 v3d_set_debug_callback(struct pipe_context *pctx,
                        const struct pipe_debug_callback *cb)
 {
@@ -172,6 +182,7 @@ v3d_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
         pctx->priv = priv;
         pctx->destroy = v3d_context_destroy;
         pctx->flush = v3d_pipe_flush;
+        pctx->memory_barrier = v3d_memory_barrier;
         pctx->set_debug_callback = v3d_set_debug_callback;
         pctx->invalidate_resource = v3d_invalidate_resource;
         pctx->get_sample_position = v3d_get_sample_position;
diff --git a/src/gallium/drivers/v3d/v3d_context.h b/src/gallium/drivers/v3d/v3d_context.h
index 25be07c2437..686ecfa8024 100644
--- a/src/gallium/drivers/v3d/v3d_context.h
+++ b/src/gallium/drivers/v3d/v3d_context.h
@@ -82,6 +82,7 @@ void v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo);
 #define VC5_DIRTY_OQ            (1 << 28)
 #define VC5_DIRTY_CENTROID_FLAGS (1 << 29)
 #define VC5_DIRTY_NOPERSPECTIVE_FLAGS (1 << 30)
+#define VC5_DIRTY_SSBO          (1 << 31)
 
 #define VC5_MAX_FS_INPUTS 64
 
@@ -203,6 +204,11 @@ struct v3d_streamout_stateobj {
         unsigned num_targets;
 };
 
+struct v3d_ssbo_stateobj {
+        struct pipe_shader_buffer sb[PIPE_MAX_SHADER_BUFFERS];
+        uint32_t enabled_mask;
+};
+
 /* Hash table key for v3d->jobs */
 struct v3d_job_key {
         struct pipe_surface *cbufs[4];
@@ -433,6 +439,7 @@ struct v3d_context {
         struct pipe_poly_stipple stipple;
         struct pipe_clip_state clip;
         struct pipe_viewport_state viewport;
+        struct v3d_ssbo_stateobj ssbo[PIPE_SHADER_TYPES];
         struct v3d_constbuf_stateobj constbuf[PIPE_SHADER_TYPES];
         struct v3d_texture_stateobj tex[PIPE_SHADER_TYPES];
         struct v3d_vertexbuf_stateobj vertexbuf;
diff --git a/src/gallium/drivers/v3d/v3d_screen.c b/src/gallium/drivers/v3d/v3d_screen.c
index 5d949514d09..dd5eda87e8f 100644
--- a/src/gallium/drivers/v3d/v3d_screen.c
+++ b/src/gallium/drivers/v3d/v3d_screen.c
@@ -299,8 +299,11 @@ v3d_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
         case PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS:
         case PIPE_SHADER_CAP_MAX_SAMPLER_VIEWS:
         case PIPE_SHADER_CAP_MAX_SHADER_IMAGES:
-        case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
                 return VC5_MAX_TEXTURE_SAMPLERS;
+
+        case PIPE_SHADER_CAP_MAX_SHADER_BUFFERS:
+                return PIPE_MAX_SHADER_BUFFERS;
+
         case PIPE_SHADER_CAP_PREFERRED_IR:
                 return PIPE_SHADER_IR_NIR;
         case PIPE_SHADER_CAP_SUPPORTED_IRS:
diff --git a/src/gallium/drivers/v3d/v3d_uniforms.c b/src/gallium/drivers/v3d/v3d_uniforms.c
index bc680af0956..5d907726349 100644
--- a/src/gallium/drivers/v3d/v3d_uniforms.c
+++ b/src/gallium/drivers/v3d/v3d_uniforms.c
@@ -276,6 +276,21 @@ v3d_write_uniforms(struct v3d_context *v3d, struct v3d_compiled_shader *shader,
                         }
                         break;
 
+                case QUNIFORM_SSBO_OFFSET: {
+                        struct pipe_shader_buffer *sb =
+                                &v3d->ssbo[stage].sb[data];
+
+                        cl_aligned_reloc(&job->indirect, &uniforms,
+                                         v3d_resource(sb->buffer)->bo,
+                                         sb->buffer_offset);
+                        break;
+                }
+
+                case QUNIFORM_GET_BUFFER_SIZE:
+                        cl_aligned_u32(&uniforms,
+                                       v3d->ssbo[stage].sb[data].buffer_size);
+                        break;
+
                 case QUNIFORM_TEXTURE_FIRST_LEVEL:
                         cl_aligned_f(&uniforms,
                                      texstate->textures[data]->u.tex.first_level);
@@ -362,6 +377,11 @@ v3d_set_shader_uniform_dirty_flags(struct v3d_compiled_shader *shader)
                         dirty |= VC5_DIRTY_FRAGTEX | VC5_DIRTY_VERTTEX;
                         break;
 
+                case QUNIFORM_SSBO_OFFSET:
+                case QUNIFORM_GET_BUFFER_SIZE:
+                        dirty |= VC5_DIRTY_SSBO;
+                        break;
+
                 case QUNIFORM_ALPHA_REF:
                         dirty |= VC5_DIRTY_ZSA;
                         break;
diff --git a/src/gallium/drivers/v3d/v3dx_draw.c b/src/gallium/drivers/v3d/v3dx_draw.c
index 46e629d0c64..7f111bbe75f 100644
--- a/src/gallium/drivers/v3d/v3dx_draw.c
+++ b/src/gallium/drivers/v3d/v3dx_draw.c
@@ -478,6 +478,17 @@ v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
                 job->submit.in_sync_bcl = v3d->out_sync;
         }
 
+        /* Mark SSBOs as being written.  We don't actually know which ones are
+         * read vs written, so just assume the worst
+         */
+        for (int s = 0; s < PIPE_SHADER_TYPES; s++) {
+                foreach_bit(i, v3d->ssbo[s].enabled_mask) {
+                        v3d_job_add_write_resource(job,
+                                                   v3d->ssbo[s].sb[i].buffer);
+                        job->tmu_dirty_rcl = true;
+                }
+        }
+
         /* Get space to emit our draw call into the BCL, using a branch to
          * jump to a new BO if necessary.
          */
diff --git a/src/gallium/drivers/v3d/v3dx_state.c b/src/gallium/drivers/v3d/v3dx_state.c
index 95a2dc30c8b..7ea5dabbe1f 100644
--- a/src/gallium/drivers/v3d/v3dx_state.c
+++ b/src/gallium/drivers/v3d/v3dx_state.c
@@ -986,6 +986,53 @@ v3d_set_stream_output_targets(struct pipe_context *pctx,
         ctx->dirty |= VC5_DIRTY_STREAMOUT;
 }
 
+static void
+v3d_set_shader_buffers(struct pipe_context *pctx,
+                       enum pipe_shader_type shader,
+                       unsigned start, unsigned count,
+                       const struct pipe_shader_buffer *buffers)
+{
+        struct v3d_context *v3d = v3d_context(pctx);
+        struct v3d_ssbo_stateobj *so = &v3d->ssbo[shader];
+        unsigned mask = 0;
+
+        if (buffers) {
+                for (unsigned i = 0; i < count; i++) {
+                        unsigned n = i + start;
+                        struct pipe_shader_buffer *buf = &so->sb[n];
+
+                        if ((buf->buffer == buffers[i].buffer) &&
+                            (buf->buffer_offset == buffers[i].buffer_offset) &&
+                            (buf->buffer_size == buffers[i].buffer_size))
+                                continue;
+
+                        mask |= 1 << n;
+
+                        buf->buffer_offset = buffers[i].buffer_offset;
+                        buf->buffer_size = buffers[i].buffer_size;
+                        pipe_resource_reference(&buf->buffer, buffers[i].buffer);
+
+                        if (buf->buffer)
+                                so->enabled_mask |= 1 << n;
+                        else
+                                so->enabled_mask &= ~(1 << n);
+                }
+        } else {
+                mask = ((1 << count) - 1) << start;
+
+                for (unsigned i = 0; i < count; i++) {
+                        unsigned n = i + start;
+                        struct pipe_shader_buffer *buf = &so->sb[n];
+
+                        pipe_resource_reference(&buf->buffer, NULL);
+                }
+
+                so->enabled_mask &= ~mask;
+        }
+
+        v3d->dirty |= VC5_DIRTY_SSBO;
+}
+
 void
 v3dX(state_init)(struct pipe_context *pctx)
 {
@@ -1025,6 +1072,8 @@ v3dX(state_init)(struct pipe_context *pctx)
         pctx->sampler_view_destroy = v3d_sampler_view_destroy;
         pctx->set_sampler_views = v3d_set_sampler_views;
 
+        pctx->set_shader_buffers = v3d_set_shader_buffers;
+
         pctx->create_stream_output_target = v3d_create_stream_output_target;
         pctx->stream_output_target_destroy = v3d_stream_output_target_destroy;
         pctx->set_stream_output_targets = v3d_set_stream_output_targets;