Mesa (main): anv: Make use_softpin compile-time in genX code

[GitLab Mirror]

Module: Mesa
Branch: main
Commit: a63e97e09a899d4117bfae52c017ad0d283572af
URL:    http://cgit.freedesktop.org/mesa/mesa/commit/?id=a63e97e09a899d4117bfae52c017ad0d283572af

Author: Jason Ekstrand <jason at jlekstrand.net>
Date:   Mon Jun  7 19:53:42 2021 -0500

anv: Make use_softpin compile-time in genX code

Whenever we have the GFX_VERx10 macro available, we can make use_softpin
a compile-time thing for everything but Broadwell and Cherryview.  This
should save us some CPU cycles especially on SKL+.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin at intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/11236>

---

 src/intel/vulkan/anv_private.h     | 29 +++++++++++++++++++++++++----
 src/intel/vulkan/genX_cmd_buffer.c | 12 ++++++------
 src/intel/vulkan/genX_query.c      |  2 +-
 3 files changed, 32 insertions(+), 11 deletions(-)

diff --git a/src/intel/vulkan/anv_private.h b/src/intel/vulkan/anv_private.h
index 059319fe609..0a955b09eae 100644
--- a/src/intel/vulkan/anv_private.h
+++ b/src/intel/vulkan/anv_private.h
@@ -1258,6 +1258,26 @@ struct anv_device {
     struct intel_debug_block_frame              *debug_frame_desc;
 };
 
+static inline bool
+anv_use_softpin(const struct anv_physical_device *pdevice)
+{
+#if defined(GFX_VERx10) && GFX_VERx10 >= 90
+   /* Sky Lake and later always uses softpin */
+   assert(pdevice->use_softpin);
+   return true;
+#elif defined(GFX_VERx10) && GFX_VERx10 < 80
+   /* Haswell and earlier never use softpin */
+   assert(!pdevice->use_softpin);
+   return false;
+#else
+   /* If we don't have a GFX_VERx10 #define, we need to look at the physical
+    * device.  Also, for GFX version 8, we need to look at the physical
+    * device because Broadwell softpins but Cherryview doesn't.
+    */
+   return pdevice->use_softpin;
+#endif
+}
+
 static inline struct anv_instance *
 anv_device_instance_or_null(const struct anv_device *device)
 {
@@ -1267,15 +1287,16 @@ anv_device_instance_or_null(const struct anv_device *device)
 static inline struct anv_state_pool *
 anv_binding_table_pool(struct anv_device *device)
 {
-   if (device->physical->use_softpin)
+   if (anv_use_softpin(device->physical))
       return &device->binding_table_pool;
    else
       return &device->surface_state_pool;
 }
 
 static inline struct anv_state
-anv_binding_table_pool_alloc(struct anv_device *device) {
-   if (device->physical->use_softpin)
+anv_binding_table_pool_alloc(struct anv_device *device)
+{
+   if (anv_use_softpin(device->physical))
       return anv_state_pool_alloc(&device->binding_table_pool,
                                   device->binding_table_pool.block_size, 0);
    else
@@ -3010,7 +3031,7 @@ struct anv_cmd_buffer {
 static inline bool
 anv_cmd_buffer_is_chainable(struct anv_cmd_buffer *cmd_buffer)
 {
-   return cmd_buffer->device->physical->use_softpin &&
+   return anv_use_softpin(cmd_buffer->device->physical) &&
       !(cmd_buffer->usage_flags & VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT);
 }
 
diff --git a/src/intel/vulkan/genX_cmd_buffer.c b/src/intel/vulkan/genX_cmd_buffer.c
index 60604a14eeb..0af322c199c 100644
--- a/src/intel/vulkan/genX_cmd_buffer.c
+++ b/src/intel/vulkan/genX_cmd_buffer.c
@@ -132,7 +132,7 @@ genX(cmd_buffer_emit_state_base_address)(struct anv_cmd_buffer *cmd_buffer)
        */
       sba.GeneralStateBufferSize       = 0xfffff;
       sba.IndirectObjectBufferSize     = 0xfffff;
-      if (device->physical->use_softpin) {
+      if (anv_use_softpin(device->physical)) {
          /* With softpin, we use fixed addresses so we actually know how big
           * our base addresses are.
           */
@@ -165,7 +165,7 @@ genX(cmd_buffer_emit_state_base_address)(struct anv_cmd_buffer *cmd_buffer)
       sba.InstructionAccessUpperBoundModifyEnable = true;
 #  endif
 #  if (GFX_VER >= 9)
-      if (cmd_buffer->device->physical->use_softpin) {
+      if (anv_use_softpin(device->physical)) {
          sba.BindlessSurfaceStateBaseAddress = (struct anv_address) {
             .bo = device->surface_state_pool.block_pool.bo,
             .offset = 0,
@@ -523,7 +523,7 @@ anv_image_init_aux_tt(struct anv_cmd_buffer *cmd_buffer,
          aux_entry_map = intel_aux_map_get_entry(cmd_buffer->device->aux_map_ctx,
                                                  address, &aux_entry_addr64);
 
-         assert(cmd_buffer->device->physical->use_softpin);
+         assert(anv_use_softpin(cmd_buffer->device->physical));
          struct anv_address aux_entry_address = {
             .bo = NULL,
             .offset = aux_entry_addr64,
@@ -2542,7 +2542,7 @@ emit_binding_table(struct anv_cmd_buffer *cmd_buffer,
     * softpin then we always keep all user-allocated memory objects resident.
     */
    const bool need_client_mem_relocs =
-      !cmd_buffer->device->physical->use_softpin;
+      !anv_use_softpin(cmd_buffer->device->physical);
    struct anv_push_constants *push = &pipe_state->push_constants;
 
    for (uint32_t s = 0; s < map->surface_count; s++) {
@@ -5045,7 +5045,7 @@ genX(cmd_buffer_set_binding_for_gfx8_vb_flush)(struct anv_cmd_buffer *cmd_buffer
                                                uint32_t vb_size)
 {
    if (GFX_VER < 8 || GFX_VER > 9 ||
-       !cmd_buffer->device->physical->use_softpin)
+       !anv_use_softpin(cmd_buffer->device->physical))
       return;
 
    struct anv_vb_cache_range *bound, *dirty;
@@ -5093,7 +5093,7 @@ genX(cmd_buffer_update_dirty_vbs_for_gfx8_vb_flush)(struct anv_cmd_buffer *cmd_b
                                                     uint64_t vb_used)
 {
    if (GFX_VER < 8 || GFX_VER > 9 ||
-       !cmd_buffer->device->physical->use_softpin)
+       !anv_use_softpin(cmd_buffer->device->physical))
       return;
 
    if (access_type == RANDOM) {
diff --git a/src/intel/vulkan/genX_query.c b/src/intel/vulkan/genX_query.c
index bad36925f52..26db49d51db 100644
--- a/src/intel/vulkan/genX_query.c
+++ b/src/intel/vulkan/genX_query.c
@@ -198,7 +198,7 @@ VkResult genX(CreateQueryPool)(
    if (pdevice->supports_48bit_addresses)
       bo_flags |= EXEC_OBJECT_SUPPORTS_48B_ADDRESS;
 
-   if (pdevice->use_softpin)
+   if (anv_use_softpin(pdevice))
       bo_flags |= EXEC_OBJECT_PINNED;
 
    if (pdevice->has_exec_async)