Hi,
On 07.10.2014 17:53, ext Daniel Sanders wrote:
> Hi,
>
> I'm debugging a problem with the shuffle() and shuffle2() functions
> on
MIPS. I think I've found (part of) the problem but if I'm correct, then
I don't see how it can be working for other targets.
The macros that make up the kernel are admittedly a bit twisty, and
shuffle is no exception. I've wrecked my brain several times on these.
>
> Starting with test/kernel/test_shuffle.cc, I've reduced the testcase down to
> n=2 and m=16 and changed stimuli in testcase() so that it only contains 13's.
> My testcase produces this output:
> Error in shuffle short 16 => short 2 :[1, 1] = shuffle( [0, 1, 2, 3, 4, 5, 6,
> 7, 8, 9, 10, 11, 12, 13, 14, 15], [13, 13]);
> element 0 should be 13 (mask 13), got 1
> element 1 should be 13 (mask 13), got 1
> Error in shuffle2 short 16 => short 2 :[1, 1] = shuffle2( [0, 1, 2, 3, 4, 5,
> 6, 7, 8, 9, 10, 11, 12, 13, 14, 15], [16, 17, 18, 19, 20, 21, 22, 23, 24, 25,
> 26, 27, 28, 29, 30, 31], [13, 13]);
> As you can see, the mask is selecting element 1 instead of the 13 that was
> expected.
Ok, this is clearly an incorrect output.
>
> lib/kernel/shuffle.cl contains the following code:
> #define _CL_IMPLEMENT_SHUFFLE(ELTYPE, MTYPE, N, M) \
> ELTYPE##N __attribute__ ((overloadable)) \
> shuffle(ELTYPE##M in, MTYPE##N mask) \
> { \
> MTYPE msize = M==3 ? 4 : M; \
> ELTYPE##N out; \
> for (int i=0; i<N; ++i) { \
> MTYPE m = mask[i] & (msize-1); \
> out[i] = in[m]; \
> } \
> return out; \
> }
>
> #define _CL_IMPLEMENT_SHUFFLE_M(ELTYPE, MTYPE, M) \
> _CL_IMPLEMENT_SHUFFLE(ELTYPE, MTYPE, M, 2) \
> ...
>
> Hasn't this swapped M and N so that the mask is being limited to 0-N (0-1)
> instead of 0-M (0-15)?
The naming here follows the OCL spec - Out and Mask are vectors of size
(0-N), the inputs are of size (0-M). The content of mask values must be
limited to the element count of the inputs.
Looking at the above - the variable M in _CL_IMPLEMENT_SHUFFLE_M *might*
be poorly named - N might be a better name here... But that should not
affect the functionality (?), as all permutations of N and M are generated.
Or is there some other thing I overlooked?
Note also, that the spec doesn't define shuffle for vectors of 3 - this
extension went in unnoticed, as some other OCL implementation had it too.
kalle
------------------------------------------------------------------------------
Meet PCI DSS 3.0 Compliance Requirements with EventLog Analyzer
Achieve PCI DSS 3.0 Compliant Status with Out-of-the-box PCI DSS Reports
Are you Audit-Ready for PCI DSS 3.0 Compliance? Download White paper
Comply to PCI DSS 3.0 Requirement 10 and 11.5 with EventLog Analyzer
http://pubads.g.doubleclick.net/gampad/clk?id=154622311&iu=/4140/ostg.clktrk
_______________________________________________
pocl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pocl-devel
