pan/v9+: Change texel buffer limits

Increase texel buffer size limit and lower uniform texel buffer
alignment limit.

Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37007>

src/gallium/drivers/panfrost/pan_cmdstream.c

@@ -1091,7 +1091,8 @@ panfrost_upload_txs_sysval(struct panfrost_batch *batch,
    if (tex->target == PIPE_BUFFER) {
       assert(dim == 1);
       unsigned buf_size = tex->u.buf.size / util_format_get_blocksize(tex->format);
-      uniform->i[0] = MIN2(buf_size, PAN_MAX_TEXEL_BUFFER_ELEMENTS);
+      uniform->i[0] =
+         MIN2(buf_size, pan_get_max_texel_buffer_elements(PAN_ARCH));
       return;
    }
 
@@ -1725,7 +1726,7 @@ panfrost_create_sampler_view_bo(struct panfrost_sampler_view *so,
          .format = format,
          .width_el =
             MIN2(so->base.u.buf.size / util_format_get_blocksize(format),
-                 PAN_MAX_TEXEL_BUFFER_ELEMENTS),
+                 pan_get_max_texel_buffer_elements(PAN_ARCH)),
          .base = prsrc->plane.base + so->base.u.buf.offset,
       };
 

src/gallium/drivers/panfrost/pan_device.h

@@ -66,8 +66,6 @@ extern "C" {
  * Gallium interface limits us to 32k at the moment */
 #define PAN_MAX_MIP_LEVELS 16
 
-#define PAN_MAX_TEXEL_BUFFER_ELEMENTS 65536
-
 /* How many power-of-two levels in the BO cache do we want? 2^12
  * minimum chosen as it is the page size that all allocations are
  * rounded to */
@@ -246,6 +244,19 @@ pan_gpu_time_to_ns(struct panfrost_device *dev, uint64_t gpu_time)
    return (gpu_time * NSEC_PER_SEC) / dev->kmod.props.timestamp_frequency;
 }
 
+static inline uint32_t
+pan_get_max_texel_buffer_elements(unsigned arch)
+{
+   if (arch >= 11)
+      /* TODO 1<<27 can be made larger for v11+ with a refactor of the buffer
+       * path away from using image logic. */
+      return 1 << 27;
+   else if (arch >= 9)
+      return 1 << 27;
+   else
+      return 65536;
+}
+
 #if defined(__cplusplus)
 } // extern "C"
 #endif

src/gallium/drivers/panfrost/pan_screen.c

@@ -782,7 +782,8 @@ panfrost_init_screen_caps(struct panfrost_screen *screen)
    caps->texture_border_color_quirk = dev->arch == 7 || dev->arch >= 10 ?
       PIPE_QUIRK_TEXTURE_BORDER_COLOR_SWIZZLE_FREEDRENO : 0;
 
-   caps->max_texel_buffer_elements = PAN_MAX_TEXEL_BUFFER_ELEMENTS;
+   caps->max_texel_buffer_elements =
+      pan_get_max_texel_buffer_elements(dev->arch);
 
    /* Must be at least 64 for correct behaviour */
    caps->texture_buffer_offset_alignment = 64;

src/panfrost/vulkan/panvk_vX_buffer_view.c

@@ -49,9 +49,10 @@ panvk_per_arch(CreateBufferView)(VkDevice _device,
 
 #if PAN_ARCH >= 9
    tex_usage_mask |= VK_BUFFER_USAGE_STORAGE_TEXEL_BUFFER_BIT;
-#endif
-
+#else
+   /* This alignment constraint only applies when TextureDescriptors are used. */
    assert(!(address & 63));
+#endif
 
    if (buffer->vk.usage & tex_usage_mask) {
       struct pan_buffer_view bview = {

src/panfrost/vulkan/panvk_vX_physical_device.c

@@ -635,11 +635,12 @@ panvk_per_arch(get_physical_device_properties)(
       .maxImageDimension3D = PAN_ARCH <= 10 ? (1 << 9) : (1 << 14),
       .maxImageDimensionCube = PAN_ARCH <= 10 ? (1 << 14) - 1 : (1 << 16),
       .maxImageArrayLayers = (1 << 16),
-      /* Currently limited by the 1D texture size, which is 2^16.
-       * TODO: If we expose buffer views as 2D textures, we can increase the
-       * limit.
-       */
-      .maxTexelBufferElements = (1 << 16),
+      /* Currently Bifrost is limited by the 1D texture size, which is 2^16,
+         while pre-v11 is limited to 2^27 elements of 16 byte formats due to
+         size fields of 32 bits. */
+      .maxTexelBufferElements = PAN_ARCH >= 11  ? PANVK_MAX_BUFFER_SIZE
+                                : PAN_ARCH >= 9 ? (1 << 27)
+                                                : (1 << 16),
       /* Each uniform entry is 16-byte and the number of entries is encoded in a
        * 12-bit field, with the minus(1) modifier, which gives 2^20.
        */
@@ -965,8 +966,8 @@ panvk_per_arch(get_physical_device_properties)(
       .integerDotProductAccumulatingSaturating64BitMixedSignednessAccelerated = false,
       .storageTexelBufferOffsetAlignmentBytes = 64,
       .storageTexelBufferOffsetSingleTexelAlignment = false,
-      .uniformTexelBufferOffsetAlignmentBytes = 64,
-      .uniformTexelBufferOffsetSingleTexelAlignment = false,
+      .uniformTexelBufferOffsetAlignmentBytes = PAN_ARCH >= 9 ? 4 : 64,
+      .uniformTexelBufferOffsetSingleTexelAlignment = PAN_ARCH >= 9,
       .maxBufferSize = PANVK_MAX_BUFFER_SIZE,
 
       /* Vulkan 1.4 properties */