pvr: Setup tile buffers.

Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20007>

3 files changed, 105 insertions, 6 deletions

diff --git a/src/imagination/vulkan/pvr_device.c b/src/imagination/vulkan/pvr_device.c
index 7fd96ac5e16..061cac44af1 100644
--- a/src/imagination/vulkan/pvr_device.c
+++ b/src/imagination/vulkan/pvr_device.c
@@ -1990,6 +1990,82 @@ err_free_nop_usc_bo:
    return result;
 }
 
+static void pvr_device_init_tile_buffer_state(struct pvr_device *device)
+{
+   simple_mtx_init(&device->tile_buffer_state.mtx, mtx_plain);
+
+   for (uint32_t i = 0; i < ARRAY_SIZE(device->tile_buffer_state.buffers); i++)
+      device->tile_buffer_state.buffers[i] = NULL;
+
+   device->tile_buffer_state.buffer_count = 0;
+}
+
+static void pvr_device_finish_tile_buffer_state(struct pvr_device *device)
+{
+   /* Destroy the mutex first to trigger asserts in case it's still locked so
+    * that we don't put things in an inconsistent state by freeing buffers that
+    * might be in use or attempt to free buffers while new buffers are being
+    * allocated.
+    */
+   simple_mtx_destroy(&device->tile_buffer_state.mtx);
+
+   for (uint32_t i = 0; i < device->tile_buffer_state.buffer_count; i++)
+      pvr_bo_free(device, device->tile_buffer_state.buffers[i]);
+}
+
+/**
+ * \brief Ensures that a certain amount of tile buffers are allocated.
+ *
+ * Make sure that \p capacity amount of tile buffers are allocated. If less were
+ * present, append new tile buffers of \p size_in_bytes each to reach the quota.
+ */
+VkResult pvr_device_tile_buffer_ensure_cap(struct pvr_device *device,
+                                           uint32_t capacity,
+                                           uint32_t size_in_bytes)
+{
+   const uint32_t cache_line_size =
+      rogue_get_slc_cache_line_size(&device->pdevice->dev_info);
+   uint32_t offset;
+   VkResult result;
+
+   simple_mtx_lock(&device->tile_buffer_state.mtx);
+
+   offset = device->tile_buffer_state.buffer_count;
+
+   /* Clamping in release and asserting in debug. */
+   assert(capacity <= ARRAY_SIZE(device->tile_buffer_state.buffers));
+   capacity = MIN2(capacity, ARRAY_SIZE(device->tile_buffer_state.buffers));
+
+   /* TODO: Implement bo multialloc? To reduce the amount of syscalls and
+    * allocations.
+    */
+   for (uint32_t i = 0; i < (capacity - offset); i++) {
+      result = pvr_bo_alloc(device,
+                            device->heaps.general_heap,
+                            size_in_bytes,
+                            cache_line_size,
+                            0,
+                            &device->tile_buffer_state.buffers[offset + i]);
+      if (result != VK_SUCCESS) {
+         for (uint32_t j = 0; j < i; j++)
+            pvr_bo_free(device, device->tile_buffer_state.buffers[offset + j]);
+
+         goto err_release_lock;
+      }
+   }
+
+   device->tile_buffer_state.buffer_count = capacity;
+
+   simple_mtx_unlock(&device->tile_buffer_state.mtx);
+
+   return VK_SUCCESS;
+
+err_release_lock:
+   simple_mtx_unlock(&device->tile_buffer_state.mtx);
+
+   return result;
+}
+
 static void pvr_device_init_default_sampler_state(struct pvr_device *device)
 {
    pvr_csb_pack (&device->input_attachment_sampler, TEXSTATE_SAMPLER, sampler) {
@@ -2100,9 +2176,11 @@ VkResult pvr_CreateDevice(VkPhysicalDevice physicalDevice,
    if (result != VK_SUCCESS)
       goto err_pvr_finish_compute_idfwdf;
 
+   pvr_device_init_tile_buffer_state(device);
+
    result = pvr_queues_create(device, pCreateInfo);
    if (result != VK_SUCCESS)
-      goto err_pvr_finish_graphics_static_clear;
+      goto err_pvr_finish_tile_buffer_state;
 
    pvr_device_init_default_sampler_state(device);
 
@@ -2126,7 +2204,8 @@ VkResult pvr_CreateDevice(VkPhysicalDevice physicalDevice,
 
    return VK_SUCCESS;
 
-err_pvr_finish_graphics_static_clear:
+err_pvr_finish_tile_buffer_state:
+   pvr_device_finish_tile_buffer_state(device);
    pvr_device_finish_graphics_static_clear_state(device);
 
 err_pvr_finish_compute_idfwdf:
@@ -2169,6 +2248,7 @@ void pvr_DestroyDevice(VkDevice _device,
    PVR_FROM_HANDLE(pvr_device, device, _device);
 
    pvr_queues_destroy(device);
+   pvr_device_finish_tile_buffer_state(device);
    pvr_device_finish_graphics_static_clear_state(device);
    pvr_device_finish_compute_idfwdf_state(device);
    pvr_bo_free(device, device->pds_compute_fence_program.pvr_bo);
diff --git a/src/imagination/vulkan/pvr_pass.c b/src/imagination/vulkan/pvr_pass.c
index f3dda723005..f9702df5ecf 100644
--- a/src/imagination/vulkan/pvr_pass.c
+++ b/src/imagination/vulkan/pvr_pass.c
@@ -348,9 +348,9 @@ static void pvr_load_op_destroy(struct pvr_device *device,
 
 #define PVR_SPM_LOAD_IN_BUFFERS_COUNT(dev_info)              \
    ({                                                        \
-      int __ret = 7U;                                        \
+      int __ret = PVR_MAX_TILE_BUFFER_COUNT;                 \
       if (PVR_HAS_FEATURE(dev_info, eight_output_registers)) \
-         __ret = 3U;                                         \
+         __ret -= 4U;                                        \
       __ret;                                                 \
    })
 
@@ -590,8 +590,14 @@ VkResult pvr_CreateRenderPass2(VkDevice _device,
          &pass->hw_setup->renders[i];
       struct pvr_load_op *load_op = NULL;
 
-      if (hw_render->tile_buffers_count)
-         pvr_finishme("Set up tile buffer table");
+      if (hw_render->tile_buffers_count) {
+         result = pvr_device_tile_buffer_ensure_cap(
+            device,
+            hw_render->tile_buffers_count,
+            hw_render->eot_setup.tile_buffer_size);
+         if (result != VK_SUCCESS)
+            goto err_free_pass;
+      }
 
       assert(!hw_render->load_op);
 
diff --git a/src/imagination/vulkan/pvr_private.h b/src/imagination/vulkan/pvr_private.h
index 7cd98a9a458..7b217313866 100644
--- a/src/imagination/vulkan/pvr_private.h
+++ b/src/imagination/vulkan/pvr_private.h
@@ -52,6 +52,7 @@
 #include "util/format/u_format.h"
 #include "util/log.h"
 #include "util/macros.h"
+#include "util/simple_mtx.h"
 #include "util/u_dynarray.h"
 #include "vk_buffer.h"
 #include "vk_command_buffer.h"
@@ -398,6 +399,14 @@ struct pvr_device {
       uint32_t large_clear_vdm_words[PVR_CLEAR_VDM_STATE_DWORD_COUNT];
    } static_clear_state;
 
+   struct {
+      simple_mtx_t mtx;
+
+#define PVR_MAX_TILE_BUFFER_COUNT 7U
+      struct pvr_bo *buffers[PVR_MAX_TILE_BUFFER_COUNT];
+      uint32_t buffer_count;
+   } tile_buffer_state;
+
    VkPhysicalDeviceFeatures features;
 
    struct pvr_bo_store *bo_store;
@@ -1521,6 +1530,10 @@ VkResult pvr_pds_unitex_state_program_create_and_upload(
    uint32_t uniform_kicks,
    struct pvr_pds_upload *const pds_upload_out);
 
+VkResult pvr_device_tile_buffer_ensure_cap(struct pvr_device *device,
+                                           uint32_t capacity,
+                                           uint32_t size_in_bytes);
+
 #define PVR_FROM_HANDLE(__pvr_type, __name, __handle) \
    VK_FROM_HANDLE(__pvr_type, __name, __handle)