Description:
1. Added group quantization and zero-point (zp) in
aocl_gemm_bf16s4f32o<bf16|f32> API.
2. Group quantization is technique to improve accuracy
where scale factors to dequantize weights varies at group
level instead of per channel and per tensor level.
3. Added zp and scaling in woq packb kernels so that for
large M values zp and scaling are performed at pack-b
stage and bf16 kernels are called
4. Adding zp support and scaling to default path in WoQ kernels
created some performance overhead when M value is very small.
5. Added string group_size to lpgemm bench to read
group size from bench_input.txt and tested for
various combinations of matrix dimensions.
6. The scalefactors could be of type float or bf16
and the zeropoint values are expected to be
in int8 format.
AMD-Internal: [SWLCSG-3168, SWLCSG-3172]
Change-Id: Iff07b54d76edc7408eb2ea0b29ce8b4a04a38f57
Description:
1. The bias type was supported only based on output data type.
2. The option is added in the pre-ops structure to select the bias data
type irrespective of the storage data type in bf16 and WoQ API's
AMD-Internal: SWLCSG-3171
Change-Id: Iac10b946c2d4a5c405b2dc857362be0058615abf
Description:
Implemented sigmoid, tanh as fused post-ops in
aocl_gemm_bf16bf16f32o<f32|bf16) API's
Sigmoid(x) = 1/1+e^(-x)
Tanh(x) = (1-e^(-2x))/(1+e^(2x))
Updated bench_lpgemm to recognize sigmod, tanh
as options for post-ops from bench_input and verified.
AMD-Internal: [SWLCSG-3178]
Change-Id: I78a3ba4a67ab63f9d671fbe315f977b016a0d969
Details:
- In WOQ, if m = 4, special case kernels are added where
s4->bf16 conversion happens inside the compute kernel and
packing is avoided. For all other cases, B matrix is
dequantized and packed at KC loop level and native bf16
kernels are re-used at compute level.
- Fixes in bench to avoid accuracy failures when datatype of
output is bf16.
Change-Id: Ie8db42da536891693d5e82a5336b66514a50ccb2
-Matrix MUL op support added in main as well as fringe bfloat16 element
wise operations kernels.
-Benchmarking/testing framework for the same is added.
-Fixed issues in setting up post-ops node index.
AMD Internal: [SWLCSG-2947, SWLCSG-2953]
Change-Id: Iba7561a6a60df41211efbf06fab1b4900207bcf8
This post-operation computes C = (beta*C + alpha*A*B) * D, where D
is a matrix with dimensions and data type the same as that of C matrix.
AMD-Internal: [SWLCSG-2953]
Change-Id: Id4df2ca76a8f696cb16edbd02c25f621f9a828fd
- Fixed framework of bf16s4f32of32 API to correct
pointer updations.
- Modified pre_op structure to exclude pre-op-offset.
Now offset is passed as a separate parameter to the
scale-pack functions.
- Fixed work-distribution among threads in MT scenario.
- Added Blocksizes and kernel-pointers and verified
functionality for the new API.
AMD-Internal: [SWLCSG-2943]
Change-Id: I58fece240d62c798c880a2b2b7fa64e560cc753d
Description:
1. Added a new API aocl_gemm_bf16s4f32of32 to support
for WoQ (Weight-only-Quantization) in LLM's
2. The API supports only reordered B matrix of data
size signed 4 bits (S4).
3. Substracting zero point and multiplying with scale
on B matrix is performed in packing B.
4. zero point and scale data should be passed by user
through pre-ops data structure.
5. The API is still in experimental state and NOT tested.
AMD-Internal: SWLCSG-2943
Change-Id: I10b159b64c2e2aaf39da5462685618ba8cc800ee
Details:
- To enable Weight-only-Quantization(WOQ) workflow,
new LPGEMM APIs are added where datatypes are A: bf16,
B: int4, C: f32/bf16. To support this, B matrix will
be reordered with type still being int4. New pack
kernels that packs the reordered B matrix after
converting the data from int4 to bf16 and applying
zero-point and scale are added.
AMD-Internal: [SWLCSG-2943]
Change-Id: Iabe23dab607913c0114b97cb2b91248babeaac03
-Quantization of f32 to bf16 (bf16 = (f32 * scale_factor) + zero_point)
instead of just type conversion in aocl_gemm_bf16bf16f32obf16.
-Support for multiple scale/sum/matrix_add/bias post-ops in a single
LPGEMM api call.
-Post-ops mask related fixes in lpgemv kernels .
-Additional scale post-ops sanity checks.
AMD-Internal: [SWLCSG-2945]
Change-Id: I3b35cc413c176bb50bfdbd6acd4839a5ba7e94bb
- Implemented optimized lpgemv for both m == 1 and n == 1 cases.
- Fixed few bugs in LPGEMV for bf16 and f32 datatypes.
- Fixed few bugs in JIT-based implementation of LPGEMM for BF16
datatype.
AMD-Internal: [SWLCSG-2354]
Change-Id: I245fd97c8f160b148656f782d241f86097a0cf38
SWISH post-op computes swish(x) = x / (1 + exp(-1 * alpha * x)).
SiLU = SWISH with alpha = 1. Adding the support for swish in JIT
based BF16 kernels.
AMD-Internal: [SWLCSG-2387]
Change-Id: I9eea0c801f5f067a5cfbd2941bc991708b86e45e
- Modified all structs that are passed to JIT-generated code to use
integer of type uint64_t rather than dim_t so that functionality
is not affected when size of BLIS-internal integer is modified
during configure time.
Change-Id: Ib81c088072badf13da4ca73be2d4af4551b713d8
Details:
- Added new folder named JIT/ under addon/aocl_gemm/. This folder
will contain all the JIT related code.
- Modified lpgemm_cntx_init code to generate main and fringe kernels
for 6x64 bf16 microkernel and store function pointers to all the
generated kernels in a global function pointer array. This happens
only when gcc version is < 11.2
- When gcc version < 11.2, microkernel uses JIT-generated kernels.
otherwise, microkernel uses the intrinsics based implementation.
AMD-Internal: [SWLCSG-2622]
Change-Id: I16256c797b2546a8cd2049680001947346260461
-This post-operation computes C = (beta*C + alpha*A*B) + D, where D is
a matrix with dimensions and data type the same as that of C matrix.
-For clang compilers (including aocc), -march=znver1 is not enabled for
zen kernels. Have updated CKVECFLAGS to capture the same.
AMD-Internal: [SWLCSG-2424]
Change-Id: Ie369f7ea5c80ab69eea3f3e03a8d9546e14f5c09
-As it stands, in LPGEMM, users are expected to pass an array of values
with length the same as N dimension as inputs for zero point or scale
factor. However at times, a single scalar value is used as zero point
or scale factor for the entire downscaling operation. The mandate to
pass an array requires the user to allocate extra memory and fill it
with the scalar value so as to be used in downscaling. This limitation
is lifted as part of this commit, and now scalar values can be passed
as zero point or scale factor.
-LPGEMM bench enhancements along with new input format to improve
readability as well as flexibility.
AMD-Internal: [SWLCSG-2581]
Change-Id: Ibd0d89f03e1acadd099382dffcabfec324ceb50f
Downscaling is used when GEMM output is accumulated at a higher
precision and needs to be converted to a lower precision afterwards.
Currently the u8s8s16 flavor of api only supports downscaling to s8
(int8_t) via aocl_gemm_u8s8s16os8 after results are accumulated at
int16_t.
LPGEMM is modified to support downscaling to different data types,
like u8, s16, apart from s8. The framework (5 loop) passes the
downscale data type to the micro-kernels. Within the micro-kernel,
based on the downscale type, appropriate beta scaling and output
buffer store logic is executed. This support is only enabled for
u8s8s16 flavor of api's.
The LPGEMM bench is also modified to support passing downscale data
type for performance and accuracy testing.
AMD-Internal: [SWLCSG-2313]
Change-Id: I723d0802baf8649e5e41236b239880a6043bfd30
1. New LPGEMM type - s8s8s16os16 and s8s8s16os8 are added.
2. New interface, frame and kernel files are added.
3. Frame and kernel level files added and modified for s8s8s16
4. s8s8s16 type involves design changes of 2 operations -
Pack B and Mat Mul
5. Pack B kernel routines to pack B matrix for s16 FMA and compute the
sum of every column of B matrix to implement the s8s8s16 operation
using the s16 FMA instructions.
5. Mat Mul Kernel files to compute the GEMM output using s16 FMA.
Here the A matrix elements are converted from int8 to uint8 (s16 FMA
works with A matrix type uint8 only) by adding extra 128 to
every A matrix element
6. Post GEMM computation, additional operations are performed on the
accumulated outputs to get the correct results.
Final C = C - ( (sum of column of B matrix) * 128 )
This is done to compensate for the addition of extra 128 to every
A matrix elements
7. With this change, two new LPGEMM APIs are introduced in LPGEMM -
s8s8s16os16 and s8s8s16os8.
8. All previously added post-ops are supported on s8s8os16/os8 also.
AMD-Internal: [CPUPL-3234]
Change-Id: I3cc23e3dcf27f215151dda7c8db29b3a7505f05c
1. Custom Clip is an element-wise post-op which is used to
clip the accumulated GEMM output within a certain range.
2. The Clip Post-Op is used in downscaled and non-downscaled
LPGEMM APIs and SGEMM.
3. Changes are done at frame and microkernel level to implement
this post-op.
4. Different versions are implemented - AVX-512, AVX-2, SSE-2
to enable custom clipping for various LPGEMM types and SGEMM
AMD-Internal: [CPUPL-3207]
Change-Id: I71c60be69e5a0dc47ca9336d58181c097b9aa0c6
1. New LPGEMM type - S8S8S32/S8 is added.
2. New interface, frame and kernel files are added.
3. Frame and kernel files added/modified for S8S8S32/S8 have
2 operations - Pack B and Mat Mul
4. Pack B kernel routines to pack B matrix for VNNI and compute the sum
of every column of B matrix to implement the S8S8S32 operation using
the VNNI instructions.
5. Mat Mul Kernel files to compute the GEMM output using the VNNI.
Here the A matrix elements are converted from int8 to uint8 (VNNI
works with A matrix type uint8 only).
6. Post GEMM computation, additional operations are performed on the
accumulated outputs to get the correct results.
7. With this change, two new LPGEMM APIs are introduced in LPGEMM -
s8s8s32os32 and s8s8s32os8.
8. All previously added post-ops are supported on S8S8S32/S8 also.
AMD-Internal: [CPUPL-3154]
Change-Id: Ib18f82bde557ea4a815a63adc7870c4234bfb9d3
1. Implemented efficient AVX-512, AVX-2 and SSE-2 version of the
error function - ERF
2. Added error function based GeLU activation post-ops for the
S32, S16 and BF16 (LPGEMM) and SGEMM APIs.
3. Changes for this includes frame and micro-kernel level changes in
addition to adding the marco based function definations of the
ERF function in the math-utils and gelu headerfiles.
AMD-Internal: [CPUPL-3036]
Change-Id: Ie50f6dcabf8896b7a6d30bbc16aa44392cc512be
1. Added Tanh approximation based GeLU Post-Op for S16, S32 and BF16
2. Changes are done at frame and micro-kernel level to
implement this post-op.
3. Efficient AVX-512 and AVX-2 vector versions of TANHF and EXPF
functions are implemented for the GeLU post-operation.
4. TANH and EXPF math functions are efficiently implemented in
macro-based fashion to exploit register level fusion of GeLU
with GEMM operations for improved performance
5. LPGEMM bench is changed to pass GeLU post-op as input and
support accuracy check to verify functional correctness
AMD-Internal: [CPUPL-2978]
Change-Id: I472ac35c00a4ea1ab983cc5f6ff6a123c8035f28
-The post operations attributes are moved to a new struct
lpgemm_post_op_attr, and an object of this struct is passed to the
low precision gemm kernels in place of the multiple parameters.
-The u8s8s32s8 api (downscale api) performance is low when the k
value is less (k < KC). Two scenarios are observed here:
a. beta = 0: Currently, for downscale api, a temporary buffer is
used to accumulate intermediate s32 output, so that it can be used
in later iterations of pc loop (k dim). The usage of this buffer
(store) can be avoided if k < KC. Here intermediate accumulation
is not required, since the after the first iteration of the pc loop,
the output can be downscaled and stored.
b. beta != 0: In this case the existing values of the original s8 C
output matrix needs to be converted to s32 and beta scaled. Currently
the s8 values are converted to s32 and stored in temporary buffer in
pc loop (5 loop algorithm) in blocks of mxNC. This temporary buffer
is passed to the micro kernel and beta scaling is applied on this.
However the mxNC block copy is costly and can be avoided if a new
condition is introduced for beta scaling in the micro kernel, whereby
the original s8 data is loaded instead of from the temporary buffer
to a register, converted to s32 and beta scaling applied on it.
AMD-Internal: [CPUPL-2884]
Change-Id: Id9b4650d500e1b553e48c4f1e4c902b3f553211c
-The bf16 gemm framework is modified to swap input column major matrices
and compute gemm for the transposed matrices (now row major) using the
existing row-major kernels. The output is written to C matrix assuming
it is transposed.
-Framework changes to support leading dimensions that are greater than
matrix widths.
-Bench changes to test low precision gemm for column major inputs.
AMD-Internal: [CPUPL-2570]
Change-Id: I22c76f52619fd76d0c0e41531828b437a1935495
- Downscaling is used when GEMM output is accumulated at a higher
precision and needs to be converted to a lower precision afterwards.
This is required in AI workloads where quantization/dequantization
routines are used.
- New GEMM APIs are introduced specifically to support this use case.
Currently downscaling support is added for s32, s16 and bfloat16 GEMM.
AMD-Internal: [CPUPL-2475]
Change-Id: I81c3ee1ba5414f62427a7a0abb6ecef0c5ff71bf
-Parametric ReLU is the generalization of leaky ReLU in which the
leakage coefficient is tunable. The support for the same is added
following the register-level fusion technique.
-Low precision bench enhancement to check accuracy/performance of
low precision gemm with PReLU.
-Bug fixes in low precision gemm kernels.
AMD-Internal: [CPUPL-2442]
Change-Id: I81336405b185a994297d122b2d868b758ae6dad5
- Low precision gemm is often used in ML/DNN workloads and is used
in conjunction with pre and post operations. Performing gemm and ops
together at the micro kernel level results in better overall performance
due to cache/register reuse of output matrix. The provision for defining
the post-operations and invoking the micro-kernel with it from the
framework is added as part of this change. This includes adding new data
structures/functions to define the post-ops to be applied and an
extensible template using which new post-ops can easily be integrated.
As for the post-operations, RELU and Bias Add for u8s8s32 is implemented
in this first cut.
- aocl_gemm bench modifications to test/benchmark RELU and Bias Add.
AMD-Internal: [CPUPL-2316]
Change-Id: Iad5fe9e54965bb52d5381ae459a69800946c7d18
