Fix testbench xt build fail #8289
Conversation
lgirdwood
requested review from
ranj063,
jxstelter,
plbossart,
mmaka1,
lbetlej,
dbaluta and
kv2019i
as code owners
| @@ -121,7 +121,7 @@ void mfcc_fill_prev_samples(struct mfcc_buffer *buf, int16_t *prev_data, | |||
|
|
|||
| buf->s_avail -= prev_data_length; | |||
| buf->s_free += prev_data_length; | |||
| buf->r_ptr = in; | |||
| buf->r_ptr = (void *)in; /* int16_t pointer but direct cast is not possible */ | |||
There was a problem hiding this comment.
is in the wrong ptr type, do we need to look at other changes here ?
There was a problem hiding this comment.
It's up to @andrula-song to confirm but in the patch she used ae_int32 type to load two ae_int16 types, that's in comments a few lines above this typecast. I was wondering too if this can be correct but it appears so.
There was a problem hiding this comment.
ok, intrinsic types are non portable. Lets revisist as we do need to support AVX here (and the API would need to become more generic).
There was a problem hiding this comment.
@singalsu @lgirdwood , I used to use AE_LA16X4_XC to load a 16bit * 4 data, but I found the result is not right, I think it is a bug of xtensa. So I use AE_L32_XC to load a 32bit data(16bit * 2) to replace twice AE_L16_XC, and force the datatype to ae_int16 to copy the last 16 bit if the input data size is odd.
| @@ -1578,10 +1578,12 @@ int module_adapter_reset(struct comp_dev *dev) | |||
| } | |||
| #endif /* CONFIG_ZEPHYR_DP_SCHEDULER */ | |||
| if (IS_PROCESSING_MODE_SINK_SOURCE(mod)) { | |||
| int i; | |||
There was a problem hiding this comment.
I wonder if we could build testbench in C99 mode. When we build SOF currently (at least Zephyr builds), we are using C99 mode:
-std=c99 -std=gnu99
... so basicly C99'ism can pass through our CI checks and testbench will always be in catch-up mode. Zephyr is C99+ -> https://docs.zephyrproject.org/latest/develop/languages/c/index.html
That's a harder question where to set this, but would be a longer term solution.
There was a problem hiding this comment.
Need to check how to add this, this commit could be possible to drop with that change.
There was a problem hiding this comment.
Using "std=c99 -std=gnu99 -fgnu89-inline" seems to work, thanks for suggestion!
This patch adds "-std=c99 -std=gnu99 -fgnu89-inline options"
to host architecture build to prevent error with
"scripts/rebuild-testbench -p tgl".
src/audio/module_adapter/module_adapter.c:
In function ‘module_adapter_reset’:
src/audio/module_adapter/module_adapter.c:1582:
error: ‘for’ loop initial declaration used outside C99 mode
src/audio/module_adapter/module_adapter.c:1584: error:
redefinition of ‘i’
Signed-off-by: Seppo Ingalsuo <[email protected]>
Build with "scripts/rebuild-testbench -p tgl" fails to error: src/audio/mfcc/mfcc_hifi3.c: In function ‘mfcc_source_copy_s16’: src/audio/mfcc/mfcc_hifi3.c:91: warning: assignment from incompatible pointer type src/audio/mfcc/mfcc_hifi3.c: In function ‘mfcc_fill_prev_samples’: src/audio/mfcc/mfcc_hifi3.c:124: warning: assignment from incompatible pointer type src/audio/mfcc/mfcc_hifi3.c: In function ‘mfcc_fill_fft_buffer’: src/audio/mfcc/mfcc_hifi3.c:159: warning: assignment from incompatible pointer type fixes: d0cb478 ("Audio: MFCC: Add HiFi3 implementation of MFCC") Signed-off-by: Seppo Ingalsuo <[email protected]>
singalsu
force-pushed
the
testbench_xt_fixes
branch
from
4cef327 to
2da8ab7
Compare
| @@ -121,7 +121,7 @@ void mfcc_fill_prev_samples(struct mfcc_buffer *buf, int16_t *prev_data, | |||
|
|
|||
| buf->s_avail -= prev_data_length; | |||
| buf->s_free += prev_data_length; | |||
| buf->r_ptr = in; | |||
| buf->r_ptr = (void *)in; /* int16_t pointer but direct cast is not possible */ | |||
There was a problem hiding this comment.
ok, intrinsic types are non portable. Lets revisist as we do need to support AVX here (and the API would need to become more generic).
No description provided.