radeonsi/gfx9: indirect buffers and all CP packets use TC L2

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

diff --git a/src/gallium/drivers/radeonsi/si_compute.c b/src/gallium/drivers/radeonsi/si_compute.c
index 79b107e96fc6f9dd5f5c75711971eb6e18b697f5..65f3261e0a1765d8b5b0af87ccbd265eae7f127f 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_compute.c
+++ b/src/gallium/drivers/radeonsi/si_compute.c
@@ -788,8 +788,9 @@ static void si_launch_grid(
        if (info->indirect) {
                r600_context_add_resource_size(ctx, info->indirect);
 
-               /* The hw doesn't read the indirect buffer via TC L2. */
-               if (r600_resource(info->indirect)->TC_L2_dirty) {
+               /* Indirect buffers use TC L2 on GFX9, but not older hw. */
+               if (sctx->b.chip_class <= VI &&
+                   r600_resource(info->indirect)->TC_L2_dirty) {
                        sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
                        r600_resource(info->indirect)->TC_L2_dirty = false;
                }

diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index fda615e30c41d2d10547f2c1611750f2f1ea2808..d2ba8df00dbf73a038340a88d0892ad523e9b66f 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -1010,8 +1010,10 @@ struct pipe_screen *radeonsi_screen_create(struct radeon_winsys *ws)
                (sscreen->b.debug_flags & DBG_MONOLITHIC_SHADERS) != 0;
 
        sscreen->b.barrier_flags.cp_to_L2 = SI_CONTEXT_INV_SMEM_L1 |
-                                           SI_CONTEXT_INV_VMEM_L1 |
-                                           SI_CONTEXT_INV_GLOBAL_L2;
+                                           SI_CONTEXT_INV_VMEM_L1;
+       if (sscreen->b.chip_class <= VI)
+               sscreen->b.barrier_flags.cp_to_L2 |= SI_CONTEXT_INV_GLOBAL_L2;
+
        sscreen->b.barrier_flags.compute_to_L2 = SI_CONTEXT_CS_PARTIAL_FLUSH;
 
        if (debug_get_bool_option("RADEON_DUMP_SHADERS", false))

diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index a0d790ac2a946780bc49aae6a5b4f2d6dd6a8a61..0f39ede0c7e19d6e1429d8d2ac27a77db801a360 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -4023,7 +4023,9 @@ static void si_memory_barrier(struct pipe_context *ctx, unsigned flags)
                                 SI_CONTEXT_WRITEBACK_GLOBAL_L2;
        }
 
-       if (flags & PIPE_BARRIER_INDIRECT_BUFFER)
+       /* Indirect buffers use TC L2 on GFX9, but not older hw. */
+       if (sctx->screen->b.chip_class <= VI &&
+           flags & PIPE_BARRIER_INDIRECT_BUFFER)
                sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
 }
 

diff --git a/src/gallium/drivers/radeonsi/si_state_draw.c b/src/gallium/drivers/radeonsi/si_state_draw.c
index d13c8b7086fa11172b1d7c34ee26737852eecfb8..2b000e7469615a26a4ede3ce155e524e9b672b7e 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_draw.c
+++ b/src/gallium/drivers/radeonsi/si_state_draw.c
@@ -1320,15 +1320,18 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
                /* Add the buffer size for memory checking in need_cs_space. */
                r600_context_add_resource_size(ctx, indirect->buffer);
 
-               if (r600_resource(indirect->buffer)->TC_L2_dirty) {
-                       sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
-                       r600_resource(indirect->buffer)->TC_L2_dirty = false;
-               }
+               /* Indirect buffers use TC L2 on GFX9, but not older hw. */
+               if (sctx->b.chip_class <= VI) {
+                       if (r600_resource(indirect->buffer)->TC_L2_dirty) {
+                               sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
+                               r600_resource(indirect->buffer)->TC_L2_dirty = false;
+                       }
 
-               if (indirect->indirect_draw_count &&
-                   r600_resource(indirect->indirect_draw_count)->TC_L2_dirty) {
-                       sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
-                       r600_resource(indirect->indirect_draw_count)->TC_L2_dirty = false;
+                       if (indirect->indirect_draw_count &&
+                           r600_resource(indirect->indirect_draw_count)->TC_L2_dirty) {
+                               sctx->b.flags |= SI_CONTEXT_WRITEBACK_GLOBAL_L2;
+                               r600_resource(indirect->indirect_draw_count)->TC_L2_dirty = false;
+                       }
                }
        }