[Mesa-dev] [PATCH] glsl: Allow token pasting in ESSL

Ian Romanick idr at freedesktop.org
Tue Aug 4 11:39:49 PDT 2015


On 07/22/2015 07:52 PM, Ryan Houdek wrote:
> All vendors I've tested allow token pasting in their ESSL shaders.
> At least one application expects this to be supported
> even though it is against spec.

I'm not very excited about deviating from the spec.  Do you know if the
Khronos reference compiler allows this behavior for ESSL shaders?  I
have submitted a spec bug report to Khronos.  We've discussed this
specific issue quite a few time in the past, and, IIRC, none of the
mobile vendors wanted to change the spec... to match what their
compilers seem to actually do.  I'll try again. :(  I should have some
information this week.

When we have deviated from the spec in the past, it has been under the
control of a driconf option.  If the option is enabled, we emit a
warning.  If the option is not enabled (default) we continue to generate
the error.  Before I suggest you write more code, let me talk to the
other vendors in Khronos.

> This would be the Dolphin GameCube/Wii emulator that requires this.

Have you contacted the Dolphin authors to see if they'd change their
code to follow the spec?  I know they've been very receptive to such
requests in the past.  degasus hangs out on #dri-devel quite a lot.

> The vendors I've tested that is known to support this is:
> ARM Mali
> Qualcomm Adreno
> IMGTec PowerVR
> Nvidia Tegra K1/X1
> 
> The vendors that I don't have access to testing:
> Intel Android
> Vivante
> ---
>  src/glsl/glcpp/glcpp-lex.l | 2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/src/glsl/glcpp/glcpp-lex.l b/src/glsl/glcpp/glcpp-lex.l
> index fa9aa50..3e2ad23 100644
> --- a/src/glsl/glcpp/glcpp-lex.l
> +++ b/src/glsl/glcpp/glcpp-lex.l
> @@ -494,8 +494,6 @@ HEXADECIMAL_INTEGER	0[xX][0-9a-fA-F]+[uU]?
>  
>  "##" {
>  	if (! parser->skipping) {
> -		if (parser->is_gles)
> -			glcpp_error(yylloc, yyextra, "Token pasting (##) is illegal in GLES");
>  		RETURN_TOKEN (PASTE);
>  	}
>  }
> 



More information about the mesa-dev mailing list