[Beignet] Program compilation crash on Ivy Bridge
Zhigang Gong
zhigang.gong at linux.intel.com
Mon Nov 17 17:52:11 PST 2014
The following error log is from LLVM internal. It's likely a llvm/clang 3.4's bug.
The kernel works fine with both beignet 0.9.3 and 1.0.0 by using llvm/clang 3.5.
The recommended llvm version for beignet is LLVM 3.5.
OpenCL platform name: Intel Gen OCL Driver
OpenCL version: OpenCL 1.2 beignet 0.9.3
OpenCL device: Intel(R) HD Graphics IvyBridge GT2
Running OpenCL 1.2 version
going to compile CL prog
Call parameter type does not match function signature!
i128 8404990
float %8 = call i32 @_Z6islessff(float %7, i128 8404990) #7
Both operands to a binary operator are not of the same type!
%18 = fmul ppc_fp128 0xM0000000000FF40060000000000000000, float %17
Invalid insertelement operands!
%20 = insertelement <4 x float> %19, ppc_fp128 %18, i32 1
Invalid insertelement operands!
%26 = insertelement <4 x float> %25, ppc_fp128 0xM00000000000000000000000000000000, i32 3
Broken module found, compilation aborted!
Aborted
> -----Original Message-----
> From: Beignet [mailto:beignet-bounces at lists.freedesktop.org] On Behalf Of
> Nicolas Bourdaud
> Sent: Monday, November 17, 2014 7:27 PM
> To: beignet at lists.freedesktop.org
> Subject: [Beignet] Program compilation crash on Ivy Bridge
>
> Hi!
>
> In beignet 0.9.3, beignet aborts in the middle on compilation with compiling for
> ivybridge microarchitecture, while it compiles and runs fine on Haswell.
>
> I have provided a test program (test.cl) that shows the problem and the
> generate log on haswell (haswell.log) and ivybridge (ivybridge.log). I have
> tested on different ivybridge CPU (HD4000, HD2500) and the issue is the
> same. I have also tested on different haswell CPU (only HD4600) and it always
> goes fine.
>
> For information beignet has been tested for:
> - ivybridge using linux 3.14 compiled with llvm 3.4.
> - haswell using linux 3.14 compiled with llvm 3.4.
> - haswell using linux 3.16 compiled with llvm 3.5 In all case, the program
> compiled and runs correctly with haswell, not with ivybridge.
>
> I could give a try with beignet 1.0.0, but maybe you the problem would be
> obvious to you.
>
> Any idea about the problem?
>
> Cheers,
>
> Nicolas Bourdaud
More information about the Beignet
mailing list