Commit 3938e557 authored by Leonardo Solis's avatar Leonardo Solis

LS replicated 9 times + removed fences

parent 95912999
......@@ -309,7 +309,11 @@ void Krnl_GA(__global float* restrict GlobPopulationCurrent,
#endif
write_channel_altera(chan_IC2Conf_genotype, LocalPopCurr[pop_cnt][pipe_cnt & MASK_GENOTYPE]);
}
/*
// Maybe this is not needed?
mem_fence(CLK_CHANNEL_MEM_FENCE);
*/
#if defined (DEBUG_KRNL_IC)
printf("\nIC - tx pop: %u", pop_cnt);
#endif
......@@ -543,7 +547,11 @@ void Krnl_GA(__global float* restrict GlobPopulationCurrent,
LocalPopNext [new_pop_cnt][gene_cnt & MASK_GENOTYPE] = tmp_offspring;
write_channel_altera(chan_GG2Conf_genotype, tmp_offspring);
}
/*
// Maybe this is not needed?
mem_fence(CLK_CHANNEL_MEM_FENCE);
*/
#if defined (DEBUG_KRNL_GG)
printf("GG - tx pop: %u", new_pop_cnt);
......@@ -682,8 +690,10 @@ void Krnl_GA(__global float* restrict GlobPopulationCurrent,
}
}
/*
// Maybe this is not needed?
mem_fence(CLK_CHANNEL_MEM_FENCE);
*/
#if defined (DEBUG_KRNL_LS)
printf("LS - got all eval & energies back\n");
......
......@@ -544,6 +544,24 @@ Speedup vs i5 cpu core: 3ptb: 59/25 = 2.36x, 1stp: 84/45 = 1.86x
>>> commit "LS replicated 9 times + added fences"
168. `Krnl_GA`: removed unnecessary mem_fences added in step 167
Only non-instrumented passes placement phase:
freq=187.5MHz
Speedup vs i5 cpu core: 3ptb: 59/23.19 = 2.54x, 1stp: 84/40.91 = 2.05x
>>> commit "LS replicated 9 times + removed fences"
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment