remotego commented on a change in pull request #9:
URL: https://github.com/apache/incubator-tvm-vta/pull/9#discussion_r479844168
##########
File path: src/sim/sim_driver.cc
##########
@@ -181,6 +181,51 @@ class SRAM {
}
memset(sram_ptr, 0, kElemBytes * xtotal * op->y_pad_1);
}
+
+ // This is for load 8bits to ACC only
+ void Load_int8(const VTAMemInsn* op,
+ DRAM* dram,
+ uint64_t* load_counter,
+ bool skip_exec) {
+ CHECK_EQ(kBits, VTA_ACC_WIDTH);
+
+ // TODO(zhanghao): extend to other width
+ CHECK_EQ(VTA_ACC_WIDTH, 32);
+ CHECK_EQ(VTA_INP_WIDTH, 8);
Review comment:
This code section is for the newly added load_int8 instructions in fsim.
The instruction will load and cast 8-bit inputs from input memory and copy onto
the 32-bit acc memory. As of now, this instruction only support 8-bit source
and 32-bit acc mem as destination, although we may expend the instruction
further.
As this instruction does not operate on wgt_mem, no requirement on wgt width
is specified here.
On the other hand, for our current intelfocl/vitis implementation, we do
have a restriction on the width of wgt mem, and we have already added an
assertion there.
----------------------------------------------------------------
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]
