freedreno/a6xx: Map inputs to VFD entries up front

Break this logic out of the loop in preperation for splitting the VFD
state emit loop up.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5064>

diff --git a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
index 42adf5df9a9a9a2d59c8f7251a124846bdcd5ed8..aad89e9d35c7f95a27b3baa817d62f7d09c87eb7 100644 (file)
--- a/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
+++ b/src/gallium/drivers/freedreno/a6xx/fd6_emit.c
@@ -555,58 +555,63 @@ static struct fd_ringbuffer *
 build_vbo_state(struct fd6_emit *emit, const struct ir3_shader_variant *vp)
 {
        const struct fd_vertex_state *vtx = emit->vtx;
-       int32_t i, j;
 
        struct fd_ringbuffer *ring = fd_submit_new_ringbuffer(emit->ctx->batch->submit,
                        4 * (10 * vp->inputs_count + 2), FD_RINGBUFFER_STREAMING);
 
-       for (i = 0, j = 0; i <= vp->inputs_count; i++) {
+       /* Determine which inputs need VFD state */
+       int32_t map[32];
+       int32_t cnt = 0;
+       for (int32_t i = 0; i <= vp->inputs_count; i++) {
                if (vp->inputs[i].sysval)
                        continue;
                if (vp->inputs[i].compmask) {
-                       struct pipe_vertex_element *elem = &vtx->vtx->pipe[i];
-                       const struct pipe_vertex_buffer *vb =
-                                       &vtx->vertexbuf.vb[elem->vertex_buffer_index];
-                       struct fd_resource *rsc = fd_resource(vb->buffer.resource);
-                       enum pipe_format pfmt = elem->src_format;
-                       enum a6xx_format fmt = fd6_pipe2vtx(pfmt);
-                       bool isint = util_format_is_pure_integer(pfmt);
-                       uint32_t off = vb->buffer_offset + elem->src_offset;
-                       uint32_t size = fd_bo_size(rsc->bo) - off;
-                       debug_assert(fmt != ~0);
+                       map[cnt++] = i;
+               }
+       }
+
+       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];
+               struct fd_resource *rsc = fd_resource(vb->buffer.resource);
+               enum pipe_format pfmt = elem->src_format;
+               enum a6xx_format fmt = fd6_pipe2vtx(pfmt);
+               bool isint = util_format_is_pure_integer(pfmt);
+               uint32_t off = vb->buffer_offset + elem->src_offset;
+               uint32_t size = fd_bo_size(rsc->bo) - off;
+               debug_assert(fmt != ~0);
 
 #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))
-                               continue;
+               /* see dEQP-GLES31.stress.vertex_attribute_binding.buffer_bounds.bind_vertex_buffer_offset_near_wrap_10
+                */
+               if (off > fd_bo_size(rsc->bo))
+                       continue;
 #endif
 
-                       OUT_PKT4(ring, REG_A6XX_VFD_FETCH(j), 4);
-                       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(j), 2);
-                       OUT_RING(ring, A6XX_VFD_DECODE_INSTR_IDX(j) |
-                                       A6XX_VFD_DECODE_INSTR_FORMAT(fmt) |
-                                       COND(elem->instance_divisor, A6XX_VFD_DECODE_INSTR_INSTANCED) |
-                                       A6XX_VFD_DECODE_INSTR_SWAP(fd6_pipe2swap(pfmt)) |
-                                       A6XX_VFD_DECODE_INSTR_UNK30 |
-                                       COND(!isint, A6XX_VFD_DECODE_INSTR_FLOAT));
-                       OUT_RING(ring, MAX2(1, elem->instance_divisor)); /* VFD_DECODE[j].STEP_RATE */
-
-                       OUT_PKT4(ring, REG_A6XX_VFD_DEST_CNTL(j), 1);
-                       OUT_RING(ring, A6XX_VFD_DEST_CNTL_INSTR_WRITEMASK(vp->inputs[i].compmask) |
-                                       A6XX_VFD_DEST_CNTL_INSTR_REGID(vp->inputs[i].regid));
-
-                       j++;
-               }
+               OUT_PKT4(ring, REG_A6XX_VFD_FETCH(j), 4);
+               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(j), 2);
+               OUT_RING(ring, A6XX_VFD_DECODE_INSTR_IDX(j) |
+                               A6XX_VFD_DECODE_INSTR_FORMAT(fmt) |
+                               COND(elem->instance_divisor, A6XX_VFD_DECODE_INSTR_INSTANCED) |
+                               A6XX_VFD_DECODE_INSTR_SWAP(fd6_pipe2swap(pfmt)) |
+                               A6XX_VFD_DECODE_INSTR_UNK30 |
+                               COND(!isint, A6XX_VFD_DECODE_INSTR_FLOAT));
+               OUT_RING(ring, MAX2(1, elem->instance_divisor)); /* VFD_DECODE[j].STEP_RATE */
+
+               OUT_PKT4(ring, REG_A6XX_VFD_DEST_CNTL(j), 1);
+               OUT_RING(ring, A6XX_VFD_DEST_CNTL_INSTR_WRITEMASK(vp->inputs[i].compmask) |
+                               A6XX_VFD_DEST_CNTL_INSTR_REGID(vp->inputs[i].regid));
        }
 
        OUT_PKT4(ring, REG_A6XX_VFD_CONTROL_0, 1);
-       OUT_RING(ring, A6XX_VFD_CONTROL_0_FETCH_CNT(j) |
-                       A6XX_VFD_CONTROL_0_DECODE_CNT(j));
+       OUT_RING(ring, A6XX_VFD_CONTROL_0_FETCH_CNT(cnt) |
+                       A6XX_VFD_CONTROL_0_DECODE_CNT(cnt));
 
        return ring;
 }