anv/gen12: Write GFX_AUX_TABLE base address register

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>

diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
index 06c6b467acf6634e0d36d7b8f7afcb1f41c2de96..0274fe8b3a8926f30cdb88f030577d01a943b79d 100644 (file)
--- a/src/intel/vulkan/anv_genX.h
+++ b/src/intel/vulkan/anv_genX.h
@@ -54,6 +54,8 @@ void genX(flush_pipeline_select_gpgpu)(struct anv_cmd_buffer *cmd_buffer);
 void genX(cmd_buffer_config_l3)(struct anv_cmd_buffer *cmd_buffer,
                                 const struct gen_l3_config *cfg);
 
+void genX(cmd_buffer_aux_map_state)(struct anv_cmd_buffer *cmd_buffer);
+
 void genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer);
 void genX(cmd_buffer_flush_dynamic_state)(struct anv_cmd_buffer *cmd_buffer);
 

diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index 33614bb942671e5c7965f6c320a4cfe1820b4faa..7e83db043464a79595418378cae4e898d82b6a1c 100644 (file)
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -2414,6 +2414,7 @@ struct anv_cmd_state {
    /* PIPELINE_SELECT.PipelineSelection */
    uint32_t                                     current_pipeline;
    const struct gen_l3_config *                 current_l3_config;
+   uint32_t                                     last_aux_map_state;
 
    struct anv_cmd_graphics_state                gfx;
    struct anv_cmd_compute_state                 compute;

diff --git a/src/intel/vulkan/genX_blorp_exec.c b/src/intel/vulkan/genX_blorp_exec.c
index 239bfb474330579772a93b117be634eefdf2100c..9c754f7318e76eb3583626920e04d21a85f91a14 100644 (file)
--- a/src/intel/vulkan/genX_blorp_exec.c
+++ b/src/intel/vulkan/genX_blorp_exec.c
@@ -255,6 +255,10 @@ genX(blorp_exec)(struct blorp_batch *batch,
 
    genX(flush_pipeline_select_3d)(cmd_buffer);
 
+#if GEN_GEN >= 12
+   genX(cmd_buffer_aux_map_state)(cmd_buffer);
+#endif
+
    genX(cmd_buffer_emit_gen7_depth_flush)(cmd_buffer);
 
    /* BLORP doesn't do anything fancy with depth such as discards, so we want

diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c
index ff9c6c79eb9af626917fa03f7a21782b21a8191b..ced99939f754c8b64d23b2219a9b4a3ffc49cc74 100644 (file)
--- a/src/intel/vulkan/genX_cmd_buffer.c
+++ b/src/intel/vulkan/genX_cmd_buffer.c
@@ -29,6 +29,7 @@
 #include "vk_util.h"
 #include "util/fast_idiv_by_const.h"
 
+#include "common/gen_aux_map.h"
 #include "common/gen_l3_config.h"
 #include "genxml/gen_macros.h"
 #include "genxml/genX_pack.h"
@@ -2660,6 +2661,34 @@ cmd_buffer_flush_push_constants(struct anv_cmd_buffer *cmd_buffer,
    cmd_buffer->state.push_constants_dirty &= ~flushed;
 }
 
+#if GEN_GEN >= 12
+void
+genX(cmd_buffer_aux_map_state)(struct anv_cmd_buffer *cmd_buffer)
+{
+   void *aux_map_ctx = cmd_buffer->device->aux_map_ctx;
+   if (!aux_map_ctx)
+      return;
+   uint32_t aux_map_state_num = gen_aux_map_get_state_num(aux_map_ctx);
+   if (cmd_buffer->state.last_aux_map_state != aux_map_state_num) {
+      /* If the aux-map state number increased, then we need to rewrite the
+       * register. Rewriting the register is used to both set the aux-map
+       * translation table address, and also to invalidate any previously
+       * cached translations.
+       */
+      uint64_t base_addr = gen_aux_map_get_base(aux_map_ctx);
+      anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
+         lri.RegisterOffset = GENX(GFX_AUX_TABLE_BASE_ADDR_num);
+         lri.DataDWord = base_addr & 0xffffffff;
+      }
+      anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_IMM), lri) {
+         lri.RegisterOffset = GENX(GFX_AUX_TABLE_BASE_ADDR_num) + 4;
+         lri.DataDWord = base_addr >> 32;
+      }
+      cmd_buffer->state.last_aux_map_state = aux_map_state_num;
+   }
+}
+#endif
+
 void
 genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer)
 {
@@ -2678,6 +2707,10 @@ genX(cmd_buffer_flush_state)(struct anv_cmd_buffer *cmd_buffer)
 
    genX(flush_pipeline_select_3d)(cmd_buffer);
 
+#if GEN_GEN >= 12
+   genX(cmd_buffer_aux_map_state)(cmd_buffer);
+#endif
+
    if (vb_emit) {
       const uint32_t num_buffers = __builtin_popcount(vb_emit);
       const uint32_t num_dwords = 1 + num_buffers * 4;
@@ -3544,6 +3577,10 @@ genX(cmd_buffer_flush_compute_state)(struct anv_cmd_buffer *cmd_buffer)
 
    genX(flush_pipeline_select_gpgpu)(cmd_buffer);
 
+#if GEN_GEN >= 12
+   genX(cmd_buffer_aux_map_state)(cmd_buffer);
+#endif
+
    if (cmd_buffer->state.compute.pipeline_dirty) {
       /* From the Sky Lake PRM Vol 2a, MEDIA_VFE_STATE:
        *