gpu-compute: support in-order data delivery in GM pipe

this patch adds an ordered response buffer to the GM pipeline
to ensure in-order data delivery. the buffer is implemented as
a stl ordered map, which sorts the request in program order by
using their sequence ID. when requests return to the GM pipeline
they are marked as done. only the oldest request may be serviced
from the ordered buffer, and only if is marked as done.

the FIFO response buffers are kept and used in OoO delivery mode
This commit is contained in:
Tony Gutierrez
2016-10-26 22:48:28 -04:00
parent b63eb1302b
commit de72e36619
8 changed files with 173 additions and 46 deletions

View File

@@ -1082,7 +1082,7 @@ namespace HsailISA
gpuDynInst->useContinuation = false;
GlobalMemPipeline* gmp = &(w->computeUnit->globalMemoryPipe);
gmp->getGMReqFIFO().push(gpuDynInst);
gmp->issueRequest(gpuDynInst);
w->wrGmReqsInPipe--;
w->rdGmReqsInPipe--;

View File

@@ -263,7 +263,7 @@ namespace HsailISA
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsRdGm++;
w->rdGmReqsInPipe--;
break;
@@ -288,7 +288,7 @@ namespace HsailISA
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsRdGm++;
w->rdGmReqsInPipe--;
break;
@@ -312,7 +312,7 @@ namespace HsailISA
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsRdGm++;
w->rdGmReqsInPipe--;
break;
@@ -330,7 +330,7 @@ namespace HsailISA
}
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsRdGm++;
w->rdGmReqsInPipe--;
break;
@@ -440,7 +440,7 @@ namespace HsailISA
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsWrGm++;
w->wrGmReqsInPipe--;
break;
@@ -460,7 +460,7 @@ namespace HsailISA
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsWrGm++;
w->wrGmReqsInPipe--;
break;
@@ -486,7 +486,7 @@ namespace HsailISA
}
}
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsWrGm++;
w->wrGmReqsInPipe--;
break;
@@ -591,7 +591,7 @@ namespace HsailISA
m->latency.set(w->computeUnit->shader->ticks(64));
m->pipeId = GLBMEM_PIPE;
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsWrGm++;
w->wrGmReqsInPipe--;
w->outstandingReqsRdGm++;

View File

@@ -648,7 +648,7 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(64));
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsWrGm++;
w->wrGmReqsInPipe--;
w->outstandingReqsRdGm++;
@@ -688,7 +688,7 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(64));
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsWrGm++;
w->wrGmReqsInPipe--;
w->outstandingReqsRdGm++;
@@ -727,7 +727,7 @@ namespace HsailISA
m->pipeId = GLBMEM_PIPE;
m->latency.set(w->computeUnit->shader->ticks(1));
w->computeUnit->globalMemoryPipe.getGMReqFIFO().push(m);
w->computeUnit->globalMemoryPipe.issueRequest(m);
w->outstandingReqsRdGm++;
w->rdGmReqsInPipe--;
w->outstandingReqs++;