[Mesa-dev] [RFC PATCH 2/2] main/format: skip format conversion if src and dst format are equal

[Karol Herbst]

On Thu, Aug 17, 2017 at 10:52 AM, Erik Faye-Lund <kusmabite at gmail.com> wrote:
> On Wed, Aug 16, 2017 at 8:32 PM, Karol Herbst <karolherbst at gmail.com> wrote:
>> Fixes 'KHR-GL45.copy_image.functional' on Nouveau
>>
>> Signed-off-by: Karol Herbst <karolherbst at gmail.com>
>> ---
>>  src/mesa/main/format_utils.c | 6 +++++-
>>  1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/mesa/main/format_utils.c b/src/mesa/main/format_utils.c
>> index d16d69c379..a3ddaebb2e 100644
>> --- a/src/mesa/main/format_utils.c
>> +++ b/src/mesa/main/format_utils.c
>> @@ -485,7 +485,11 @@ _mesa_format_convert(void *void_dst, uint32_t dst_format, size_t dst_stride,
>>
>>     assert(src_integer == dst_integer);
>>
>> -   if (src_integer && dst_integer) {
>> +   /* do a simply memcpy if applicable */
>> +   if (dst_format == src_format && dst_stride == src_stride &&
>> +       !dst_format_is_mesa_array_format && !rebase_swizzle) {
>> +      memcpy(dst, src, src_stride * height);
>> +   } else if (src_integer && dst_integer) {
>
> This sounds like it just makes the bug harder to trigger. What if the
> format is the same, but the pitch isn't the same? Won't that cause the
> same issue, just without the CTS picking it up?

Okay, I forgot to mention that we have a few formats where we never
are able to reproduce the original data. Think about RGB9_E5_EXT and
0xc0000000 as the raw data. We lose the exponent while converting to
RGBA and can't reproduce it anymore when converting back to
RGB9_E5_EXT.

Maybe somebody knows a much better way how to solve this issue? Or
maybe it's simple CTS which is wrong by trying to read out the GPU
memory with glGetTexImage?