[Mesa-dev] [PATCH 4/5] radv: Use correct clear words for HTILE.

[Bas Nieuwenhuizen]

Yeah, I got your message on IRC and found this one too. Looking at
things I suspect it is correct and just unhides a bug somewhere else
though, so I was a bit reluctant to roll back. However, since I'm
probably not able to fix this before the weekend, I just send a change
that reverts the regressing part of the change.

- Bas

On Tue, May 30, 2017 at 9:15 PM, Fredrik Höglund <fredrik at kde.org> wrote:
> On Monday 22 May 2017, Bas Nieuwenhuizen wrote:
>> Did some RE'ing what several HTILE words give when read from a descriptor
>> with HTILE compression enabled.
>>
>> Seems to align with -pro usage for D16 too.
>>
>> Signed-off-by: Bas Nieuwenhuizen <basni at google.com>
>> ---
>>  src/amd/vulkan/radv_cmd_buffer.c | 17 +++++++++++++----
>>  1 file changed, 13 insertions(+), 4 deletions(-)
>>
>> diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
>> index d0101259fb1..eff0fd68c9f 100644
>> --- a/src/amd/vulkan/radv_cmd_buffer.c
>> +++ b/src/amd/vulkan/radv_cmd_buffer.c
>> @@ -2971,10 +2971,16 @@ void radv_CmdEndRenderPass(
>>       cmd_buffer->state.framebuffer = NULL;
>>  }
>>
>> -
>> +/*
>> + * For HTILE we have the following interesting clear words:
>> + *   0x0000030f: Uncompressed.
>> + *   0xfffffff0: Clear depth to 1.0
>> + *   0x00000000: Clear depth to 0.0
>> + */
>>  static void radv_initialize_htile(struct radv_cmd_buffer *cmd_buffer,
>>                                    struct radv_image *image,
>> -                                  const VkImageSubresourceRange *range)
>> +                                  const VkImageSubresourceRange *range,
>> +                                  uint32_t clear_word)
>>  {
>>       assert(range->baseMipLevel == 0);
>>       assert(range->levelCount == 1 || range->levelCount == VK_REMAINING_ARRAY_LAYERS);
>> @@ -2986,7 +2992,7 @@ static void radv_initialize_htile(struct radv_cmd_buffer *cmd_buffer,
>>       cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_DB |
>>                                       RADV_CMD_FLAG_FLUSH_AND_INV_DB_META;
>>
>> -     radv_fill_buffer(cmd_buffer, image->bo, offset, size, 0xffffffff);
>> +     radv_fill_buffer(cmd_buffer, image->bo, offset, size, clear_word);
>>
>>       cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_DB_META |
>>                                       RADV_CMD_FLAG_CS_PARTIAL_FLUSH |
>> @@ -3013,7 +3019,10 @@ static void radv_handle_depth_image_transition(struct radv_cmd_buffer *cmd_buffe
>>       } else if (src_layout == VK_IMAGE_LAYOUT_UNDEFINED &&
>>                  radv_layout_has_htile(image, dst_layout, dst_queue_mask)) {
>>               /* TODO: merge with the clear if applicable */
>> -             radv_initialize_htile(cmd_buffer, image, range);
>> +             radv_initialize_htile(cmd_buffer, image, range, 0);
>> +     } else if (!radv_layout_is_htile_compressed(image, src_layout, src_queue_mask) &&
>> +                radv_layout_is_htile_compressed(image, dst_layout, dst_queue_mask)) {
>> +             radv_initialize_htile(cmd_buffer, image, range, 0x0000030f);
>
> This change broke Mad Max on Polaris 10.
>
>>       } else if (radv_layout_is_htile_compressed(image, src_layout, src_queue_mask) &&
>>                  !radv_layout_is_htile_compressed(image, dst_layout, dst_queue_mask)) {
>>               VkImageSubresourceRange local_range = *range;
>>
>