liangfu commented on a change in pull request #4392: [VTA] Enable streamlined
GEMM execution
URL: https://github.com/apache/incubator-tvm/pull/4392#discussion_r350550653
##########
File path: vta/hardware/chisel/src/main/scala/core/TensorGemm.scala
##########
@@ -236,11 +254,14 @@ class TensorGemm(debug: Boolean = false)(implicit p:
Parameters)
when(state === sIdle) {
inflight := 0.U
}.elsewhen(!dec.reset) {
- when(state === sReadTensor) { // issue a tensor
- inflight := inflight + 1.U
- }.elsewhen(mvc.io.acc_o.data.valid) { // commit a tensor
- inflight := inflight - 1.U
- }
+ when((state === sReadTensor) && mvc.io.acc_o.data.valid) { // issue &
commit
Review comment:
Yes, if `latency=3` and we don't handle this case here, the hardware would
wait infinitely long for `inflight` going to zero. But it never does, because
sReadTensor state and `mvc.io.acc_o.data.valid` might simultaneously fire when
latency equals to 3, and in that case, the increase amount in the `inflight`
counter should be zero.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
