panvk: Prepare panvk_draw_prepare_varyings for JM indirect

On JM hardware, we need to allocate a buffer depending on vertex count.

As a result, for indirect and indexed draw we allocate a large buffer with
alloc on fault set.

The size of that buffer is calculated assuming a max of 2 millions vertices
and 18 attributes per vertex (16 user attributes, 2 specials)

Signed-off-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Olivia Lee <olivia.lee@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35724>

src/panfrost/vulkan/jm/panvk_vX_cmd_draw.c

@@ -76,6 +76,7 @@ struct panvk_draw_data {
    struct {
       uint64_t attribs;
       uint64_t attrib_bufs;
+      uint64_t varying_bufs;
    } indirect_info;
 };
 
@@ -441,6 +442,7 @@ static VkResult
 panvk_draw_prepare_varyings(struct panvk_cmd_buffer *cmdbuf,
                             struct panvk_draw_data *draw)
 {
+   struct panvk_device *dev = to_panvk_device(cmdbuf->vk.base.device);
    const struct panvk_shader_variant *vs =
       panvk_shader_hw_variant(cmdbuf->state.gfx.vs.shader);
    const struct panvk_shader_link *link = &cmdbuf->state.gfx.link;
@@ -455,18 +457,44 @@ panvk_draw_prepare_varyings(struct panvk_cmd_buffer *cmdbuf,
    bool writes_point_size =
       vs->info.vs.writes_point_size &&
       ia->primitive_topology == VK_PRIMITIVE_TOPOLOGY_POINT_LIST;
-   unsigned vertex_count =
-      draw->padded_vertex_count * draw->info.instance.count;
    uint64_t psiz_buf = 0;
 
-   for (unsigned i = 0; i < PANVK_VARY_BUF_MAX; i++) {
-      unsigned buf_size = vertex_count * link->buf_strides[i];
-      uint64_t buf_addr =
-         buf_size ? panvk_cmd_alloc_dev_mem(cmdbuf, varying, buf_size, 64).gpu
-                  : 0;
-      if (buf_size && !buf_addr)
+   if (is_indirect_draw(draw) &&
+       !cmdbuf->state.gfx.vs.indirect_varying_bufs_infos) {
+      struct pan_ptr bufs_info_storage = panvk_cmd_alloc_dev_mem(
+         cmdbuf, desc, sizeof(struct libpan_draw_helper_varying_buf_info), 8);
+
+      if (!bufs_info_storage.gpu)
          return VK_ERROR_OUT_OF_DEVICE_MEMORY;
 
+      cmdbuf->state.gfx.vs.indirect_varying_bufs_infos = bufs_info_storage.gpu;
+
+      struct libpan_draw_helper_varying_buf_info *vary_bufs_info =
+         bufs_info_storage.cpu;
+      vary_bufs_info->address = dev->indirect_varying_buffer->addr.dev;
+      vary_bufs_info->size = PANVK_JM_MAX_PER_VTX_ATTRIBUTES_INDIRECT_SIZE *
+                             PANVK_JM_MAX_VERTICES_INDIRECT;
+      vary_bufs_info->offset = 0;
+   }
+
+   for (unsigned i = 0; i < PANVK_VARY_BUF_MAX; i++) {
+      uint32_t buf_size;
+      uint64_t buf_addr;
+      if (is_indirect_draw(draw)) {
+         buf_addr = dev->indirect_varying_buffer->addr.dev;
+         buf_size = 0;
+      } else {
+         buf_size = draw->padded_vertex_count * draw->info.instance.count *
+                    link->buf_strides[i];
+         buf_addr =
+            buf_size
+               ? panvk_cmd_alloc_dev_mem(cmdbuf, varying, buf_size, 64).gpu
+               : 0;
+
+         if (buf_size && !buf_addr)
+            return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+      }
+
       pan_pack(&buf_descs[i], ATTRIBUTE_BUFFER, cfg) {
          cfg.stride = link->buf_strides[i];
          cfg.size = buf_size;
@@ -493,6 +521,8 @@ panvk_draw_prepare_varyings(struct panvk_cmd_buffer *cmdbuf,
       draw->line_width = 1.0f;
 
    draw->varying_bufs = bufs.gpu;
+   draw->indirect_info.varying_bufs =
+      cmdbuf->state.gfx.vs.indirect_varying_bufs_infos;
    draw->vs.varyings = panvk_priv_mem_dev_addr(link->vs.attribs);
    draw->fs.varyings = panvk_priv_mem_dev_addr(link->fs.attribs);
    return VK_SUCCESS;

src/panfrost/vulkan/panvk_cmd_draw.h

@@ -151,6 +151,7 @@ struct panvk_cmd_graphics_state {
       uint64_t attrib_bufs;
       uint64_t indirect_attribs_infos;
       uint64_t indirect_attrib_bufs_infos;
+      uint64_t indirect_varying_bufs_infos;
       bool previous_draw_was_indirect;
 #endif
    } vs;

src/panfrost/vulkan/panvk_device.h

@@ -27,6 +27,18 @@
 #include "util/u_printf.h"
 #include "util/vma.h"
 
+/* On JM hardware, we need to allocate a buffer depending on vertex count.
+ *
+ * As a result, for indirect and indexed draw we allocate a large buffer with
+ * alloc on fault set.
+ *
+ * The size of that buffer is calculated assuming a max of 2 millions vertices
+ * and 18 attributes per vertex (16 user attributes, 2 specials)
+ */
+
+#define PANVK_JM_MAX_VERTICES_INDIRECT                (2000000)
+#define PANVK_JM_MAX_PER_VTX_ATTRIBUTES_INDIRECT_SIZE (18 * 4)
+
 struct panvk_precomp_cache;
 struct panvk_device_draw_context;
 
@@ -55,6 +67,7 @@ struct panvk_device {
    } kmod;
 
    struct panvk_priv_bo *tiler_heap;
+   struct panvk_priv_bo *indirect_varying_buffer;
    struct panvk_priv_bo *sample_positions;
 
    struct {

src/panfrost/vulkan/panvk_vX_cmd_meta.c

@@ -174,6 +174,7 @@ panvk_per_arch(cmd_meta_gfx_end)(
    cmdbuf->state.gfx.vs.attrib_bufs = 0;
    cmdbuf->state.gfx.vs.indirect_attribs_infos = 0;
    cmdbuf->state.gfx.vs.indirect_attrib_bufs_infos = 0;
+   cmdbuf->state.gfx.vs.indirect_varying_bufs_infos = 0;
    cmdbuf->state.gfx.fs.rsd = 0;
 #else
    cmdbuf->state.gfx.fs.desc.res_table = 0;

src/panfrost/vulkan/panvk_vX_device.c

@@ -426,6 +426,14 @@ panvk_per_arch(create_device)(struct panvk_physical_device *physical_device,
       VK_SYSTEM_ALLOCATION_SCOPE_DEVICE, &device->tiler_heap);
    if (result != VK_SUCCESS)
       goto err_free_priv_bos;
+
+   result = panvk_priv_bo_create(
+      device,
+      PANVK_JM_MAX_VERTICES_INDIRECT * PANVK_JM_MAX_PER_VTX_ATTRIBUTES_INDIRECT_SIZE,
+      PAN_KMOD_BO_FLAG_NO_MMAP | PAN_KMOD_BO_FLAG_ALLOC_ON_FAULT,
+      VK_SYSTEM_ALLOCATION_SCOPE_DEVICE, &device->indirect_varying_buffer);
+   if (result != VK_SUCCESS)
+      goto err_free_priv_bos;
 #endif
 
    result = panvk_priv_bo_create(
@@ -533,6 +541,7 @@ err_free_priv_bos:
    panvk_priv_bo_unref(device->printf.bo);
    panvk_priv_bo_unref(device->tiler_oom.handlers_bo);
    panvk_priv_bo_unref(device->sample_positions);
+   panvk_priv_bo_unref(device->indirect_varying_buffer);
    panvk_priv_bo_unref(device->tiler_heap);
    panvk_device_cleanup_mempools(device);
    vk_free(&device->vk.alloc, device->dump_region_size);
@@ -578,6 +587,7 @@ panvk_per_arch(destroy_device)(struct panvk_device *device,
    u_printf_destroy(&device->printf.ctx);
    panvk_priv_bo_unref(device->printf.bo);
    panvk_priv_bo_unref(device->tiler_oom.handlers_bo);
+   panvk_priv_bo_unref(device->indirect_varying_buffer);
    panvk_priv_bo_unref(device->tiler_heap);
    panvk_priv_bo_unref(device->sample_positions);
    panvk_device_cleanup_mempools(device);