Con Kolivas
|
1b5c676de7
|
Use 256 output slots for kernels to allow 1 for each worksize.
|
13 years ago |
Philip Kaufmann
|
f479be0700
|
add goffset support for diakgcn with -v 1 and update kernel version
|
13 years ago |
Philip Kaufmann
|
e2b7c934f6
|
removed 8-component vector support from kernel, as this is not supported in CGMINER anyway
|
13 years ago |
Philip Kaufmann
|
782fc63c50
|
forgot to update kernel modification date, fixed ;)
|
13 years ago |
Philip Kaufmann
|
c326158710
|
reordered an addition in the kernel, which results in less instructions used in the GPU ISA code for GCN
|
13 years ago |
Philip Kaufmann
|
aa52e7dfd7
|
optimized nonce-check and output code for -v 2 and -v 4
|
13 years ago |
Philip Kaufmann
|
77f177b683
|
only small code formating changes
|
13 years ago |
Philip Kaufmann
|
d9d4831b47
|
removed vec_step() as this could lead to errors on older SDKs
|
13 years ago |
Philip Kaufmann
|
d7d0797969
|
unified code for generating nonce in kernel and moved addition of base to the end -> faster
|
13 years ago |
Con Kolivas
|
d32cd583ac
|
Reinstate the old output mechanism setting output[FOUND] per vector.
|
13 years ago |
Con Kolivas
|
fb077c6d59
|
Pass vectors * worksize to kernel to avoid one op.
|
13 years ago |
Con Kolivas
|
70e8ade54f
|
Revert behaviour to old nonce init code.
|
13 years ago |
Philip Kaufmann
|
a731de6e45
|
revert to legacy nonce creation in the kernel without vector offset, but keep GOFFSET code removed
|
13 years ago |
Philip Kaufmann
|
a46ee18118
|
revert to former method and do not use any()-function in check for valid nonces, as this is slower
|
13 years ago |
Philip Kaufmann
|
47747dc8a2
|
revert to legacy nonce creation in the kernel without vector offset, but keep GOFFSET code removed
|
13 years ago |
Philip Kaufmann
|
837f9ebb76
|
revert to former method and do not use any()-function in check for valid nonces, as this is slower
|
13 years ago |
Philip Kaufmann
|
71ab39dbbe
|
Explicitly type the constants in diakgcn kernel as uint, to be in line with poclbm kernel.
|
13 years ago |
Philip Kaufmann
|
74d1c4d89c
|
further fix for new vector offset method
|
13 years ago |
ckolivas
|
01c197d479
|
Use one less op before the branch code in diakgcn.
|
13 years ago |
Philip Kaufmann
|
bb51b628d3
|
tried to fix vector offset with diakgcn
|
13 years ago |
Con Kolivas
|
0bde957912
|
Update all kernel version names.
|
13 years ago |
Con Kolivas
|
8f08a775ad
|
Use any() in kernel output code and revert breakage of diakgcn kernel.
|
13 years ago |
Con Kolivas
|
145f3c0b1d
|
Put the nonce for each vector offset in advance, avoiding one extra addition in the kernel.
|
13 years ago |
Philip Kaufmann
|
52d4581642
|
small changes to speedup no vec for AMD 898.1 OCL runtime
|
13 years ago |
Philip Kaufmann
|
8a008ff505
|
Fixed no vectors, shortened code a bit (only cosmetic)
|
13 years ago |
Philip Kaufmann
|
99d058a1f1
|
re-forked cgminer - initial commit to update DiaKGCN kernel
|
13 years ago |
Con Kolivas
|
83dde50f9d
|
Updates to diakgcn kernel courtesy of Philip Kaufmann <phil.kaufmann@t-online.de>
|
13 years ago |
ckolivas
|
6693898710
|
Fix diakgcn kernel working by passing it D1A variable and renaming it in the cl code accordingly.
Renumber version consistent with change in code since it now works.
|
13 years ago |
ckolivas
|
604b2f7645
|
Set cl_amd_media_ops with the BITALIGN flag and allow non-bitselect devices to build.
|
13 years ago |
Philip Kaufmann
|
ec2dfeacc1
|
instead of uint *, int * was used as type for the output buffer (self-made bug from a former commmit)
|
13 years ago |
Philip Kaufmann
|
c43dd734cb
|
renamed all rot-functions to rotr to be clear that it is a rotate right
|
13 years ago |
Philip Kaufmann
|
ab848057da
|
changed base-nonce generation for non GOFFSET paths to use get_local_size(0) and left shift
|
13 years ago |
Philip Kaufmann
|
1398a805a7
|
leave out kernel parameter "const u base", if GOFFSET (not yet implemented) is defined
|
13 years ago |
Philip Kaufmann
|
dcd286182e
|
changed format of output buffer from ulong * to uint * to be cgminer compatible
|
13 years ago |
Philip Kaufmann
|
15cc3582d8
|
First commit - only changed date in diakgcn120208.cl
|
13 years ago |
ckolivas
|
cce08b6e6d
|
Reinstate original code as much as possible for comparison.
|
13 years ago |
ckolivas
|
9e9db012fb
|
Convert to Unix EOL.
|
13 years ago |
ckolivas
|
49cff57098
|
Use correct variable on output in diakgcn kernel.
|
13 years ago |
ckolivas
|
033913ca8e
|
First working port of the diakgcn kernel.
|
13 years ago |
ckolivas
|
3af3cbe96a
|
Import diakgcn kernel.
|
13 years ago |
ckolivas
|
fa5f4a649b
|
Set cl_amd_media_ops with the BITALIGN flag and allow non-bitselect devices to build.
|
13 years ago |
Philip Kaufmann
|
0fe8b60783
|
instead of uint *, int * was used as type for the output buffer (self-made bug from a former commmit)
|
13 years ago |
Philip Kaufmann
|
6d997d4567
|
renamed all rot-functions to rotr to be clear that it is a rotate right
|
13 years ago |
Philip Kaufmann
|
05c74b05e0
|
changed base-nonce generation for non GOFFSET paths to use get_local_size(0) and left shift
|
13 years ago |
Philip Kaufmann
|
ecd2b0db6b
|
leave out kernel parameter "const u base", if GOFFSET (not yet implemented) is defined
|
13 years ago |
Philip Kaufmann
|
0834250842
|
changed format of output buffer from ulong * to uint * to be cgminer compatible
|
13 years ago |
Philip Kaufmann
|
f25075c889
|
First commit - only changed date in diakgcn120208.cl
|
13 years ago |
ckolivas
|
5904976580
|
Reinstate original code as much as possible for comparison.
|
13 years ago |
ckolivas
|
c5142b7f12
|
Convert to Unix EOL.
|
13 years ago |
ckolivas
|
5eab0e76aa
|
Use correct variable on output in diakgcn kernel.
|
13 years ago |