gallium/u_threaded: allow drivers to change tc_call_set_vertex_buffers function

Move the execute function pointers to struct threaded_context, so that
drivers can change it. Also move struct tc_vertex_buffers into the header
file, so that drivers can implement their own function.

This allows drivers to inline pipe_context::set_vertex_buffers for TC.

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27492>

src/gallium/auxiliary/util/u_threaded_context.c

@@ -54,13 +54,6 @@
 
 #define TC_SENTINEL 0x5ca1ab1e
 
-enum tc_call_id {
-#define CALL(name) TC_CALL_##name,
-#include "u_threaded_context_calls.h"
-#undef CALL
-   TC_NUM_CALLS,
-};
-
 #if TC_DEBUG >= 3 || defined(TC_TRACE)
 static const char *tc_call_names[] = {
 #define CALL(name) #name,
@@ -75,10 +68,6 @@ static const char *tc_call_names[] = {
 #  define TC_TRACE_SCOPE(call_id)
 #endif
 
-typedef uint16_t (*tc_execute)(struct pipe_context *pipe, void *call);
-
-static const tc_execute execute_func[TC_NUM_CALLS];
-
 static void
 tc_buffer_subdata(struct pipe_context *_pipe,
                   struct pipe_resource *resource,
@@ -448,6 +437,8 @@ batch_execute(struct tc_batch *batch, struct pipe_context *pipe, uint64_t *last,
     * begin incrementing renderpass info on the first set_framebuffer_state call
     */
    bool first = !batch->first_set_fb;
+   const tc_execute *execute_func = batch->tc->execute_func;
+
    for (uint64_t *iter = batch->slots; iter != last;) {
       struct tc_call_base *call = (struct tc_call_base *)iter;
 
@@ -2151,12 +2142,6 @@ tc_set_shader_buffers(struct pipe_context *_pipe,
    tc->shader_buffers_writeable_mask[shader] |= writable_bitmask << start;
 }
 
-struct tc_vertex_buffers {
-   struct tc_call_base base;
-   uint8_t count;
-   struct pipe_vertex_buffer slot[0]; /* more will be allocated if needed */
-};
-
 static uint16_t
 tc_call_set_vertex_buffers(struct pipe_context *pipe, void *call)
 {
@@ -5156,12 +5141,6 @@ tc_destroy(struct pipe_context *_pipe)
    FREE(tc);
 }
 
-static const tc_execute execute_func[TC_NUM_CALLS] = {
-#define CALL(name) tc_call_##name,
-#include "u_threaded_context_calls.h"
-#undef CALL
-};
-
 void tc_driver_internal_flush_notify(struct threaded_context *tc)
 {
    /* Allow drivers to call this function even for internal contexts that
@@ -5419,6 +5398,10 @@ threaded_context_create(struct pipe_context *pipe,
    CTX_INIT(get_intel_perf_query_data);
 #undef CTX_INIT
 
+#define CALL(name) tc->execute_func[TC_CALL_##name] = tc_call_##name;
+#include "u_threaded_context_calls.h"
+#undef CALL
+
    if (out)
       *out = tc;
 

src/gallium/auxiliary/util/u_threaded_context.h

@@ -273,6 +273,13 @@ struct tc_unflushed_batch_token;
  */
 #define TC_MAX_SUBDATA_BYTES        320
 
+enum tc_call_id {
+#define CALL(name) TC_CALL_##name,
+#include "u_threaded_context_calls.h"
+#undef CALL
+   TC_NUM_CALLS,
+};
+
 enum tc_binding_type {
    TC_BINDING_VERTEX_BUFFER,
    TC_BINDING_STREAMOUT_BUFFER,
@@ -302,6 +309,8 @@ enum tc_binding_type {
    TC_BINDING_IMAGE_CS,
 };
 
+typedef uint16_t (*tc_execute)(struct pipe_context *pipe, void *call);
+
 typedef void (*tc_replace_buffer_storage_func)(struct pipe_context *ctx,
                                                struct pipe_resource *dst,
                                                struct pipe_resource *src,
@@ -558,6 +567,12 @@ struct threaded_context_options {
    void (*fs_parse)(void *state, struct tc_renderpass_info *info);
 };
 
+struct tc_vertex_buffers {
+   struct tc_call_base base;
+   uint8_t count;
+   struct pipe_vertex_buffer slot[0]; /* more will be allocated if needed */
+};
+
 struct threaded_context {
    struct pipe_context base;
    struct pipe_context *pipe;
@@ -652,6 +667,9 @@ struct threaded_context {
    struct tc_renderpass_info *renderpass_info_recording;
    /* accessed by driver thread */
    struct tc_renderpass_info *renderpass_info;
+
+   /* Callbacks that call pipe_context functions. */
+   tc_execute execute_func[TC_NUM_CALLS];
 };