radeonsi/gfx10: implement streamout-related queries

[mesa.git] / src / gallium / drivers / radeonsi / si_query.c

diff --git a/src/gallium/drivers/radeonsi/si_query.c b/src/gallium/drivers/radeonsi/si_query.c
index 1d541f194f35623531c0503419350b6b5a5e7213..394bf7ff12477a79d670bc9c424134a358c182e3 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_query.c
+++ b/src/gallium/drivers/radeonsi/si_query.c
@@ -32,8 +32,6 @@
 #include "util/u_suballoc.h"
 #include "amd/common/sid.h"
 
-#define SI_MAX_STREAMS 4
-
 static const struct si_query_ops query_hw_ops;
 
 struct si_hw_query_params {
@@ -58,12 +56,12 @@ struct si_query_sw {
        struct pipe_fence_handle *fence;
 };
 
-static void si_query_sw_destroy(struct si_screen *sscreen,
+static void si_query_sw_destroy(struct si_context *sctx,
                                struct si_query *squery)
 {
        struct si_query_sw *query = (struct si_query_sw *)squery;
 
-       sscreen->b.fence_reference(&sscreen->b, &query->fence, NULL);
+       sctx->b.screen->fence_reference(sctx->b.screen, &query->fence, NULL);
        FREE(query);
 }
 
@@ -624,12 +622,11 @@ bool si_query_buffer_alloc(struct si_context *sctx, struct si_query_buffer *buff
 }
 
 
-void si_query_hw_destroy(struct si_screen *sscreen,
-                        struct si_query *squery)
+void si_query_hw_destroy(struct si_context *sctx, struct si_query *squery)
 {
        struct si_query_hw *query = (struct si_query_hw *)squery;
 
-       si_query_buffer_destroy(sscreen, &query->buffer);
+       si_query_buffer_destroy(sctx->screen, &query->buffer);
        si_resource_reference(&query->workaround_buf, NULL);
        FREE(squery);
 }
@@ -1016,6 +1013,12 @@ static void si_emit_query_predication(struct si_context *ctx)
        if (!query)
                return;
 
+       if (ctx->chip_class == GFX10 &&
+           (query->b.type == PIPE_QUERY_SO_OVERFLOW_PREDICATE ||
+            query->b.type == PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE)) {
+               assert(!"not implemented");
+       }
+
        invert = ctx->render_cond_invert;
        flag_wait = ctx->render_cond_mode == PIPE_RENDER_COND_WAIT ||
                    ctx->render_cond_mode == PIPE_RENDER_COND_BY_REGION_WAIT;
@@ -1097,6 +1100,14 @@ static struct pipe_query *si_create_query(struct pipe_context *ctx, unsigned que
             query_type != SI_QUERY_TIME_ELAPSED_SDMA))
                return si_query_sw_create(query_type);
 
+       if (sscreen->info.chip_class >= GFX10 &&
+           (query_type == PIPE_QUERY_PRIMITIVES_EMITTED ||
+            query_type == PIPE_QUERY_PRIMITIVES_GENERATED ||
+            query_type == PIPE_QUERY_SO_STATISTICS ||
+            query_type == PIPE_QUERY_SO_OVERFLOW_PREDICATE ||
+            query_type == PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE))
+               return gfx10_sh_query_create(sscreen, query_type, index);
+
        return si_query_hw_create(sscreen, query_type, index);
 }
 
@@ -1105,7 +1116,7 @@ static void si_destroy_query(struct pipe_context *ctx, struct pipe_query *query)
        struct si_context *sctx = (struct si_context *)ctx;
        struct si_query *squery = (struct si_query *)query;
 
-       squery->ops->destroy(sctx->screen, squery);
+       squery->ops->destroy(sctx, squery);
 }
 
 static boolean si_begin_query(struct pipe_context *ctx,
@@ -1461,20 +1472,6 @@ bool si_query_hw_get_result(struct si_context *sctx,
        return true;
 }
 
-static void si_restore_qbo_state(struct si_context *sctx,
-                                struct si_qbo_state *st)
-{
-       sctx->b.bind_compute_state(&sctx->b, st->saved_compute);
-
-       sctx->b.set_constant_buffer(&sctx->b, PIPE_SHADER_COMPUTE, 0, &st->saved_const0);
-       pipe_resource_reference(&st->saved_const0.buffer, NULL);
-
-       sctx->b.set_shader_buffers(&sctx->b, PIPE_SHADER_COMPUTE, 0, 3, st->saved_ssbo,
-                                  st->saved_ssbo_writable_mask);
-       for (unsigned i = 0; i < 3; ++i)
-               pipe_resource_reference(&st->saved_ssbo[i].buffer, NULL);
-}
-
 static void si_query_hw_get_result_resource(struct si_context *sctx,
                                               struct si_query *squery,
                                               bool wait,
@@ -1819,7 +1816,7 @@ static struct pipe_driver_query_info si_driver_query_list[] = {
 static unsigned si_get_num_queries(struct si_screen *sscreen)
 {
        /* amdgpu */
-       if (sscreen->info.drm_major == 3) {
+       if (sscreen->info.is_amdgpu) {
                if (sscreen->info.chip_class >= GFX8)
                        return ARRAY_SIZE(si_driver_query_list);
                else
@@ -1920,10 +1917,11 @@ void si_init_query_functions(struct si_context *sctx)
        sctx->b.end_query = si_end_query;
        sctx->b.get_query_result = si_get_query_result;
        sctx->b.get_query_result_resource = si_get_query_result_resource;
-       sctx->atoms.s.render_cond.emit = si_emit_query_predication;
 
-       if (((struct si_screen*)sctx->b.screen)->info.num_render_backends > 0)
-           sctx->b.render_condition = si_render_condition;
+       if (sctx->has_graphics) {
+               sctx->atoms.s.render_cond.emit = si_emit_query_predication;
+               sctx->b.render_condition = si_render_condition;
+       }
 
        LIST_INITHEAD(&sctx->active_queries);
 }