freedreno: replace fixed array for globabl_bindings with dynamic array

Freedreno limits set_global_binding() to 16 resource entries
(MAX_GLOBAL_BUFFERS), however RustiCL can pass more global resources
(e.g. OpenCL CTS test api / min_max_constant_args requires passing of
17). Follow example of other drivers and use dynamic array for global
bindings.

Backport-to: 25.1
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@oss.qualcomm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35122>

src/gallium/drivers/freedreno/a4xx/fd4_compute.c

@@ -127,7 +127,7 @@ fd4_launch_grid(struct fd_context *ctx,
    fd4_emit_cs_state(ctx, ring, v);
    fd4_emit_cs_consts(v, ring, ctx, info);
 
-   u_foreach_bit (i, ctx->global_bindings.enabled_mask)
+   util_dynarray_foreach (&ctx->global_bindings, struct pipe_resource *, res)
       nglobal++;
 
    if (nglobal > 0) {
@@ -138,10 +138,8 @@ fd4_launch_grid(struct fd_context *ctx,
        * payload:
        */
       OUT_PKT3(ring, CP_NOP, 2 * nglobal);
-      u_foreach_bit (i, ctx->global_bindings.enabled_mask) {
-         struct pipe_resource *prsc = ctx->global_bindings.buf[i];
-         OUT_RELOC(ring, fd_resource(prsc)->bo, 0, 0, 0);
-      }
+      util_dynarray_foreach (&ctx->global_bindings, struct pipe_resource *, res)
+         OUT_RELOC(ring, fd_resource(*res)->bo, 0, 0, 0);
    }
 
    const unsigned *local_size =

src/gallium/drivers/freedreno/a5xx/fd5_compute.c

@@ -106,7 +106,7 @@ fd5_launch_grid(struct fd_context *ctx,
    fd5_emit_cs_state(ctx, ring, v);
    fd5_emit_cs_consts(v, ring, ctx, info);
 
-   u_foreach_bit (i, ctx->global_bindings.enabled_mask)
+   util_dynarray_foreach (&ctx->global_bindings, struct pipe_resource *, res)
       nglobal++;
 
    if (nglobal > 0) {
@@ -117,10 +117,8 @@ fd5_launch_grid(struct fd_context *ctx,
        * payload:
        */
       OUT_PKT7(ring, CP_NOP, 2 * nglobal);
-      u_foreach_bit (i, ctx->global_bindings.enabled_mask) {
-         struct pipe_resource *prsc = ctx->global_bindings.buf[i];
-         OUT_RELOC(ring, fd_resource(prsc)->bo, 0, 0, 0);
-      }
+      util_dynarray_foreach (&ctx->global_bindings, struct pipe_resource *, res)
+         OUT_RELOC(ring, fd_resource(*res)->bo, 0, 0, 0);
    }
 
    const unsigned *local_size =

src/gallium/drivers/freedreno/freedreno_context.c

@@ -401,6 +401,8 @@ fd_context_destroy(struct pipe_context *pctx)
       if (ctx->clear_rs_state[i])
          pctx->delete_rasterizer_state(pctx, ctx->clear_rs_state[i]);
 
+   util_dynarray_fini(&ctx->global_bindings);
+
    slab_destroy_child(&ctx->transfer_pool);
    slab_destroy_child(&ctx->transfer_pool_unsync);
 
@@ -678,6 +680,8 @@ fd_context_init(struct fd_context *ctx, struct pipe_screen *pscreen,
    slab_create_child(&ctx->transfer_pool, &screen->transfer_pool);
    slab_create_child(&ctx->transfer_pool_unsync, &screen->transfer_pool);
 
+   util_dynarray_init(&ctx->global_bindings, NULL);
+
    fd_draw_init(pctx);
    fd_resource_context_init(pctx);
    fd_query_context_init(pctx);

src/gallium/drivers/freedreno/freedreno_context.h

@@ -14,6 +14,7 @@
 #include "util/list.h"
 #include "util/slab.h"
 #include "util/u_blitter.h"
+#include "util/u_dynarray.h"
 #include "util/u_string.h"
 #include "util/u_threaded_context.h"
 #include "util/perf/u_trace.h"
@@ -110,12 +111,6 @@ struct fd_streamout_stateobj {
    unsigned verts_written;
 };
 
-#define MAX_GLOBAL_BUFFERS 16
-struct fd_global_bindings_stateobj {
-   struct pipe_resource *buf[MAX_GLOBAL_BUFFERS];
-   uint32_t enabled_mask;
-};
-
 /* group together the vertex and vertexbuf state.. for ease of passing
  * around, and because various internal operations (gmem<->mem, etc)
  * need their own vertex state:
@@ -507,7 +502,7 @@ struct fd_context {
    struct fd_shaderbuf_stateobj shaderbuf[PIPE_SHADER_TYPES] dt;
    struct fd_shaderimg_stateobj shaderimg[PIPE_SHADER_TYPES] dt;
    struct fd_streamout_stateobj streamout dt;
-   struct fd_global_bindings_stateobj global_bindings dt;
+   struct util_dynarray global_bindings dt;
    struct pipe_clip_state ucp dt;
 
    struct pipe_query *cond_query dt;

src/gallium/drivers/freedreno/freedreno_draw.c

@@ -596,8 +596,9 @@ fd_launch_grid(struct pipe_context *pctx,
    /* For global buffers, we don't really know if read or written, so assume
     * the worst:
     */
-   u_foreach_bit (i, ctx->global_bindings.enabled_mask)
-      resource_written(batch, ctx->global_bindings.buf[i]);
+   util_dynarray_foreach (&ctx->global_bindings, struct pipe_resource *, res) {
+      resource_written(batch, *res);
+   }
 
    if (info->indirect)
       resource_read(batch, info->indirect);

src/gallium/drivers/freedreno/freedreno_state.c

@@ -725,40 +725,35 @@ fd_set_global_binding(struct pipe_context *pctx, unsigned first, unsigned count,
                       struct pipe_resource **prscs, uint32_t **handles) in_dt
 {
    struct fd_context *ctx = fd_context(pctx);
-   struct fd_global_bindings_stateobj *so = &ctx->global_bindings;
-   unsigned mask = 0;
+   unsigned old_size = util_dynarray_num_elements(&ctx->global_bindings, *prscs);
 
-   if (prscs) {
-      for (unsigned i = 0; i < count; i++) {
-         unsigned n = i + first;
+   if (old_size < first + count) {
+      /* we are screwed no matter what */
+      if (!util_dynarray_grow(&ctx->global_bindings, *prscs,
+                              (first + count) - old_size))
+         unreachable("out of memory");
 
-         mask |= BIT(n);
+      for (unsigned i = old_size; i < first + count; i++)
+         *util_dynarray_element(&ctx->global_bindings,
+                                struct pipe_resource *, i) = NULL;
+   }
 
-         pipe_resource_reference(&so->buf[n], prscs[i]);
+   for (unsigned i = first; i < first + count; ++i) {
+      struct pipe_resource **res = util_dynarray_element(&ctx->global_bindings,
+                                                         struct pipe_resource *,
+                                                         first + i);
+      if (prscs && prscs[i]) {
+         pipe_resource_reference(res, prscs[i]);
 
-         if (so->buf[n]) {
-            struct fd_resource *rsc = fd_resource(so->buf[n]);
-            uint32_t offset = *handles[i];
-            uint64_t iova = fd_bo_get_iova(rsc->bo) + offset;
+         struct fd_resource *rsc = fd_resource(prscs[i]);
+         uint32_t offset = *handles[i];
+         uint64_t iova = fd_bo_get_iova(rsc->bo) + offset;
 
-            /* Yes, really, despite what the type implies: */
-            memcpy(handles[i], &iova, sizeof(iova));
-         }
-
-         if (prscs[i])
-            so->enabled_mask |= BIT(n);
-         else
-            so->enabled_mask &= ~BIT(n);
+         /* Yes, really, despite what the type implies: */
+         memcpy(handles[i], &iova, sizeof(iova));
+      } else {
+         pipe_resource_reference(res, NULL);
       }
-   } else {
-      mask = (BIT(count) - 1) << first;
-
-      for (unsigned i = 0; i < count; i++) {
-         unsigned n = i + first;
-         pipe_resource_reference(&so->buf[n], NULL);
-      }
-
-      so->enabled_mask &= ~mask;
    }
 }