[Mesa-dev] [Mesa-stable] [PATCH] ac: fix the number of coordinates for ac_image_get_lod and arrays

[Nicolai Hähnle]

On 23.04.2018 21:45, Samuel Pitoiset wrote:
> 
> 
> On 04/23/2018 08:42 PM, Marek Olšák wrote:
>> On Mon, Apr 23, 2018 at 1:12 PM, Samuel Pitoiset 
>> <samuel.pitoiset at gmail.com <mailto:samuel.pitoiset at gmail.com>> wrote:
>>
>>
>>
>>     On 04/23/2018 06:55 PM, Nicolai Hähnle wrote:
>>
>>         On 23.04.2018 17:52, Samuel Pitoiset wrote:
>>
>>             This fixes crashes for the following CTS:
>>             dEQP-VK.glsl.texture_functions.query.texturequerylod.*
>>
>>             Fixes: 625dcbbc456 ("amd/common: pass address components
>>             individually to
>>             ac_build_image_intrinsic")
>>             Cc: 18.1 <mesa-stable at lists.freedesktop.org
>>             <mailto:mesa-stable at lists.freedesktop.org>>
>>             Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com
>>             <mailto:samuel.pitoiset at gmail.com>>
>>             ---
>>                src/amd/common/ac_llvm_build.c | 13 +++++++++++++
>>                1 file changed, 13 insertions(+)
>>
>>             diff --git a/src/amd/common/ac_llvm_build.c
>>             b/src/amd/common/ac_llvm_build.c
>>             index 02739f9da9c..d5bad3eeea3 100644
>>             --- a/src/amd/common/ac_llvm_build.c
>>             +++ b/src/amd/common/ac_llvm_build.c
>>             @@ -1521,6 +1521,19 @@ LLVMValueRef
>>             ac_build_image_opcode(struct ac_llvm_context *ctx,
>>                    LLVMValueRef addr;
>>                    unsigned num_addr = 0;
>>             +    if (a->opcode == ac_image_get_lod) {
>>             +        switch (a->dim) {
>>             +        case ac_image_1darray:
>>             +            num_coords = 1;
>>             +            break;
>>             +        case ac_image_2darray:
>>             +            num_coords = 2;
>>
>>
>>         This is probably also needed for cube maps, isn't it?
>>
>>
>>     cubes should have 3 coordinates for textureQueryLod(), no?
>>
>>
>> Cubemaps are the same as 2D arrays.
> 
> Okay. CTS doesn't seem to have tests for cubemaps but that makes sense 
> anyway. I'm going to push the patch with the cubemaps fix.

You know, now I'm not so sure anymore :}

The coordinates sent for cubemaps are s/t/layer like for 2D arrays, like 
Marek said and I was thinking yesterday, but there's special handling 
for cubemap corners. We really need a test for that though, because the 
docs aren't too clear either.

Cheers,
Nicolai

> 
>>
>> Marek
>>


-- 
Lerne, wie die Welt wirklich ist,
Aber vergiss niemals, wie sie sein sollte.