freedreno/a6xx: Decouple VFD_FETCH and VFD_DECODE

We used to output a VFD_FETCH entry for each VFD_DECODE, but we can
instead output just one VFD_FETCH per VBO and point multiple
VFD_DECODE entries at the same VFD_FETCH entry.  There's typically
fewer VBOs than vertex elements so this is a small win in itselfs, but
more importantly, the VFD_DECODE state now only depends on program
state.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5140>

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
index 5f5198f265260fa5bb131dfb94c3b7792b424820..741f4b8e6d3ea9263e82a028a9311cfdd9e99068 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
@@ -568,37 +568,29 @@ build_vbo_state(struct fd6_emit *emit, const struct ir3_shader_variant *vp)
        }
 
        struct fd_ringbuffer *ring = fd_submit_new_ringbuffer(emit->ctx->batch->submit,
-                       4 * (5 + cnt * 7), FD_RINGBUFFER_STREAMING);
+                       4 * (5 + cnt * 3 + vtx->vertexbuf.count * 4), FD_RINGBUFFER_STREAMING);
 
        OUT_PKT4(ring, REG_A6XX_VFD_CONTROL_0, 1);
-       OUT_RING(ring, A6XX_VFD_CONTROL_0_FETCH_CNT(cnt) |
+       OUT_RING(ring, A6XX_VFD_CONTROL_0_FETCH_CNT(vtx->vertexbuf.count) |
                        A6XX_VFD_CONTROL_0_DECODE_CNT(cnt));
 
-       OUT_PKT4(ring, REG_A6XX_VFD_FETCH(0), 4 * cnt);
-       for (int32_t j = 0; j < cnt; j++) {
-               int32_t i = map[j];
-               struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
-               const struct pipe_vertex_buffer *vb =
-                       &vtx->vertexbuf.vb[elem->vertex_buffer_index];
+       OUT_PKT4(ring, REG_A6XX_VFD_FETCH(0), 4 * vtx->vertexbuf.count);
+       for (int32_t j = 0; j < vtx->vertexbuf.count; j++) {
+               const struct pipe_vertex_buffer *vb = &vtx->vertexbuf.vb[j];
                struct fd_resource *rsc = fd_resource(vb->buffer.resource);
-               uint32_t off = vb->buffer_offset;
-               uint32_t size = fd_bo_size(rsc->bo) - off;
-
-#ifdef DEBUG
-               /* see dEQP-GLES31.stress.vertex_attribute_binding.buffer_bounds.bind_vertex_buffer_offset_near_wrap_10
-                */
-               if (off > fd_bo_size(rsc->bo)) {
+               if (rsc == NULL) {
                        OUT_RING(ring, 0);
                        OUT_RING(ring, 0);
                        OUT_RING(ring, 0);
                        OUT_RING(ring, 0);
-                       continue;
-               }
-#endif
+               } else {
+                       uint32_t off = vb->buffer_offset;
+                       uint32_t size = fd_bo_size(rsc->bo) - off;
 
-               OUT_RELOC(ring, rsc->bo, off, 0, 0);
-               OUT_RING(ring, size);           /* VFD_FETCH[j].SIZE */
-               OUT_RING(ring, vb->stride);     /* VFD_FETCH[j].STRIDE */
+                       OUT_RELOC(ring, rsc->bo, off, 0, 0);
+                       OUT_RING(ring, size);           /* VFD_FETCH[j].SIZE */
+                       OUT_RING(ring, vb->stride);     /* VFD_FETCH[j].STRIDE */
+               }
        }
 
        OUT_PKT4(ring, REG_A6XX_VFD_DECODE(0), 2 * cnt);
@@ -610,7 +602,7 @@ build_vbo_state(struct fd6_emit *emit, const struct ir3_shader_variant *vp)
                bool isint = util_format_is_pure_integer(pfmt);
                debug_assert(fmt != FMT6_NONE);
 
-               OUT_RING(ring, A6XX_VFD_DECODE_INSTR_IDX(j) |
+               OUT_RING(ring, A6XX_VFD_DECODE_INSTR_IDX(elem->vertex_buffer_index) |
                                A6XX_VFD_DECODE_INSTR_OFFSET(elem->src_offset) |
                                A6XX_VFD_DECODE_INSTR_FORMAT(fmt) |
                                COND(elem->instance_divisor, A6XX_VFD_DECODE_INSTR_INSTANCED) |