The expected two cycle passing between meta write and idle states is
independent of the number of streams. Specifically addresses an
off-by-one timestamp error with the bladeRF 1.0.
The conditional that moves the meta state to META_DOWNCOUNT contains a
max value check using the `val+1=0` idiom. We can remove the carry
propagation delay by comparing against a max value constant
assigned (others => '1'). The fixes a setup time violation.
FX3 GPIF buffers set to 2048 samples and number of buffers reduced to
11 per dma channel. This substantially increases the theoretical USB
SuperSpeed throughput from roughly 228MB/s to 356MB/s.
$ make test_libbladeRF
[ OK ] libbladeRF.scheduled_retune (19755 ms)
[----------] 29 tests from libbladeRF (308802 ms total)
[----------] Global test environment tear-down
[==========] 29 tests from 1 test suite ran. (308802 ms total)
[ PASSED ] 29 tests.
$ make test_hdl
[----------] Global test environment tear-down
[==========] 63 tests from 2 test suites ran. (381209 ms total)
[ PASSED ] 63 tests.
[100%] Built target test_hdl
As we add changes to the FX3, it will be much easier to have a single
source of truth for the GPIF DMA sizes. This commit should not introduce
any changes to simulation or synthesis results.
We were limiting ourselves to half of the fifo depth and subtracting
4 gpif buffers from that, which is 1/4 of the current fifo depth. No
sign of a full fifo during siso, mimo, meta, and non-meta streaming
tests at low sample rates. The fifo writes are appropriately held off.
The channel shift would be left at 1 after the state moved to
COMPUTE_ENABLED_CHANNELS. Samples left would be reassigned to 1 in
COMPUTE_OFFSETS. Then channel offset would be assigned 2, an invalid
value.
This change only assigns samples left and channel shift if at the 0th
sample index, which allows for sample continuity.
This copies the gpif 8b testbench, but with 7 blocks per TX iteration
and 2 subsequent iteration repeats. Run for 4ms to find channel shift
reset failure.
The expected two cycle passing between meta write and idle states is
independent of the number of streams. Specifically addresses an
off-by-one timestamp error with the bladeRF 1.0.
Resolves Nios and FX3 lockup when SI53304 clock select assignments are
made. The FX3 pclk and Nios system PLLs lose lock when the SI53304
transitions between clock sources and this prevents invalid cycles from
making it through.
This fixes the first sample being dropped in both meta and non-meta modes.
fifo_write is now asserted through procedure and executes on HOLDOFF exit.
Regression test results:
$ make test_libbladeRF
[ OK ] libbladeRF.scheduled_retune (19450 ms)
[----------] 29 tests from libbladeRF (297804 ms total)
[==========] 29 tests from 1 test suite ran. (297804 ms total)
[ PASSED ] 29 tests.
$ make test_hdl
[----------] 31 tests from hdl (1354828 ms total)
[----------] 32 tests from hdl/verify (2319711 ms total)
[==========] 63 tests from 2 test suites ran. (3674539 ms total)
[ PASSED ] 63 tests.
Accommodates the following warning issued by CMake v3.27.1.
CMake Deprecation Warning:
Compatibility with CMake < 3.5 will be removed from a
future version of CMake.