Generating functions improperly

[ready4droid]

I am working on a Spir-V Decompiler and am looking at some binaries produced by glslang. It incorrectly generates opcode 54 (OpFunction) with only a 4 byte word count, followed by multiple other OpFunction opcodes. This is incorrect, because it should be immediately followed by 55, OpFunctionParameter, for each argument, and then followed by a 56 for OpFuncitonEnd. There is no generated OpFunctionParameter or OpFuncitonEnd in the created binary. Both source and binary are attached. Compiled using version 2.2.687

[ready4droid]

I stand corrected, it can't attach the binary file. Here is the source:
#version 110

attribute vec2 position;

varying vec2 texcoord;

void main()
{
gl_Position = vec4(position, 0.0, 1.0);
texcoord = position * vec2(0.5) + vec2(0.5);
}

[ready4droid]

After a bit more research, I realize it's not that it's generating functions improperly. The function that it is creating is really supposed to just be a name (OpName, opcode 5). So it's incorrectly generating opcodes.

Even OpExtInstImport is being generated as opcode 4 instead of opcode 11. Is there a new standard i'm missing where these things have changed?

[baldurk]

Building from head code the binary SPIR-V seems fine for me (OpExtInstImport first word highlighted with **s):

00000000    03 02 23 07   63 00 00 00  BB 00 1A 05  41 00 00 00  00 00 00 00  ..#.c.......A.......
00000014    03 00 03 00   02 00 00 00  A4 01 00 00  11 00 02 00  01 00 00 00  ....................
00000028   *0B 00 06 00*  01 00 00 00  47 4C 53 4C  2E 73 74 64  2E 34 35 30  ........GLSL.std.450
0000003C    00 00 00 00

OpExtInstImport was opcode 4 in rev 30 of SPIR-V, but glslang generates rev 31 which is opcode 11. Maybe you're building glslang from somewhere else, or running an older binary?

[ready4droid]

I downloaded the official windows executable from the https://www.khronos.org/opengles/sdk/tools/Reference-Compiler/ website. (https://cvs.khronos.org/svn/repos/ogl/trunk/ecosystem/public/sdk/tools/glslang/Install). I didn't realize it was so out of date, I will update tomorrow. Thanks for the help, I thought it was something weird because they were all formed properly just that the opcode # wasn't matching up with the reference api.

[johnkslang]

FYI, I updated the binaries at Khronos, but yes, they are generally updated less frequently than the source here at GitLab.

[ready4droid]

Thanks, I did get it to compile, but it didn't do so under vs 2010. I had to use cmake to generate new project files and download vs 2015.

[baldurk]

There is a poll in #42 about which VS version to support, currently vs2010 would almost work but needs some minor tweaks. I have a branch that isn't up to date where I try and keep 2010 working, since that's what I need.

[ready4droid]

Not a big deal, didn't realize the binary was that outdated and wasn't
having luck compiling. I got it now though. Slowly working on a software
Vulcan driver, of only the spec would be released. I mostly just copied
the mantle API until they do.
On Oct 6, 2015 4:41 PM, "Baldur Karlsson" notifications@github.com wrote:

There is a poll in #42 #42
about which VS version to support, currently vs2010 would almost work but
needs some minor tweaks. I have a branch that isn't up to date where I try
and keep 2010 working, since that's what I need.

—
Reply to this email directly or view it on GitHub
#78 (comment)
.