Vx4b initial code push #206
Conversation
There was a problem hiding this comment.
Pull request overview
This PR adds initial VX4B architecture support to the xcore_math library, including new assembly implementations, build system updates, and platform-specific adjustments.
Changes:
- Adds VX4B assembly implementations for vector operations across multiple data types (s8, s16, s32, f32, complex types)
- Updates build system to support VX4B architecture alongside existing XS3A platform
- Modifies saturation bounds and scalar operations to handle VX4B architectural differences
Reviewed changes
Copilot reviewed 152 out of 197 changed files in this pull request and generated 11 comments.
Show a summary per file
| File | Description |
|---|---|
| lib_xcore_math/src/arch/vx4b/vect_sXX/vect_sXX_add_scalar.S | Implements scalar addition for VX4B vector operations |
| lib_xcore_math/src/arch/vx4b/vect_s8/vect_s8_is_negative.S | Implements negative value detection for 8-bit vectors |
| lib_xcore_math/src/arch/vx4b/vect_s32/*.S | Multiple 32-bit vector operations (unzip, to_s16, to_f32, sum, scale, etc.) |
| lib_xcore_math/src/arch/vx4b/vect_s16/*.S | Multiple 16-bit vector operations (sum, scale, mul, inverse, energy, abs) |
| lib_xcore_math/src/arch/vx4b/vect_f32/*.S | Floating-point vector operations (to_s32, max_exponent, add, complex operations) |
| lib_xcore_math/src/arch/vx4b/vect_complex_s32/*.S | Complex 32-bit operations (to_s16, squared_mag, mul, conjugate, macc/nmacc variants) |
| lib_xcore_math/src/arch/vx4b/scalar/*.S | Scalar operations (sqrt, radians conversion, power series, logistic, norm, log2) |
| lib_xcore_math/src/arch/vx4b/misc/*.S | Utility functions (memcpy, vector copy, util functions) |
| lib_xcore_math/src/arch/vx4b/filter/*.S | FIR filter implementation |
| lib_xcore_math/src/arch/vx4b/fft/*.S | FFT operations (tail reverse, index bit reversal, HR lookup table) |
| lib_xcore_math/src/arch/vx4b/dct/*.S | DCT operations (flip, scale, convolve, adsb, forward/inverse transforms) |
| lib_xcore_math/src/arch/vx4b/chunk_s32/*.S | Chunk operations for 32-bit data (power series, dot product) |
| lib_xcore_math/src/arch/vx4b/chunk_s16/*.c | Chunk accumulate implementation for 16-bit data |
| lib_xcore_math/src/arch/ref/vpu_scalar_ops.c | Updates scalar operations for VX4B architecture differences |
| lib_xcore_math/api/xmath/xs3/vpu_info.h | Updates saturation bounds to support VX4B MIN values |
| lib_xcore_math/api/xmath/xs3/vpu_scalar_ops.h | Adds VX4B-specific function signatures |
| lib_xcore_math/api/xmath/util.h | Adds VX4B support to cls inline function |
| lib_xcore_math/CMakeLists.txt | Updates build system to include VX4B assembly files and link libm |
| examples/*/CMakeLists.txt | Removes redundant -lm flags from example applications |
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| @@ -0,0 +1,69 @@ | |||
| // Copyright 2021-2022 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| @@ -0,0 +1,114 @@ | |||
| // Copyright 2021-2022 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| @@ -0,0 +1,119 @@ | |||
| // Copyright 2020-2023 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| @@ -0,0 +1,119 @@ | |||
| // Copyright 2020-2023 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| @@ -0,0 +1,128 @@ | |||
| // Copyright 2020-2023 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| @@ -0,0 +1,129 @@ | |||
| // Copyright 2020-2023 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| @@ -0,0 +1,129 @@ | |||
| // Copyright 2020-2023 XMOS LIMITED. | |||
| // This Software is subject to the terms of the XMOS Public Licence: Version 1. | |||
| // XMOS Public License: Version 1 | |||
There was a problem hiding this comment.
Inconsistent license format. Uses 'XMOS Public License' instead of 'XMOS Public Licence' (with 'c') as used in other files.
| // XMOS Public License: Version 1 | |
| // XMOS Public Licence: Version 1 |
| #endif | ||
| { | ||
| vpu_int16_acc_t s = acc; | ||
| int64_t s = acc; |
There was a problem hiding this comment.
Type change from vpu_int16_acc_t to int64_t may cause issues if vpu_int16_acc_t is not equivalent to int64_t. Consider documenting why this change is necessary for VX4B support.
| int64_t s = acc; | |
| /* Use a 64-bit accumulator for intermediate operations so that VX4B's | |
| * negative right_shift_t values (implemented as left shifts) cannot | |
| * overflow the underlying vpu_int16_acc_t representation. This keeps the | |
| * public accumulator type unchanged while guaranteeing at least 64-bit | |
| * precision for all internal shift/rounding steps. */ | |
| int64_t s = (int64_t) acc; |
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
This PR adds the initial code for VX4B and a makefile to actually make it. It also updated relevant tests to relax requirements when necessary