mirror of
https://github.com/amd/blis.git
synced 2026-04-21 08:08:51 +00:00
907226c0af
- Add a testsuite for gathering performance (in GFLOPs) and measuring correctness for the POWER10 GEMM reduced precision/integer kernels. - Reworked GENERIC_GEMM template to hardcode the cache parameters. - Remove kernel wrapper that checked that only allowed matrices that weren't transposed or conjugated. However, the kernels still assume the matrices are not transposed. This wrapper was removed for performance reasons. - Renamed and restructured files and functions for clarity. - Editted the POWER10 document to reflect new changes.
89 lines
3.0 KiB
C
89 lines
3.0 KiB
C
/*
|
|
|
|
BLIS
|
|
An object-based framework for developing high-performance BLAS-like
|
|
libraries.
|
|
|
|
Copyright (C) 2014, The University of Texas at Austin
|
|
|
|
Redistribution and use in source and binary forms, with or without
|
|
modification, are permitted provided that the following conditions are
|
|
met:
|
|
- Redistributions of source code must retain the above copyright
|
|
notice, this list of conditions and the following disclaimer.
|
|
- Redistributions in binary form must reproduce the above copyright
|
|
notice, this list of conditions and the following disclaimer in the
|
|
documentation and/or other materials provided with the distribution.
|
|
- Neither the name(s) of the copyright holder(s) nor the names of its
|
|
contributors may be used to endorse or promote products derived
|
|
from this software without specific prior written permission.
|
|
|
|
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
/*
|
|
|
|
Details on bit16_dt vector data structure
|
|
|
|
Vector X = [ X[0,0] X[0,1] X[1,0] X[1,1] X[2,0] X[2,1] X[3,0] X[3,1] ]
|
|
Vector Y = [ Y[0,0] Y[0,1] Y[1,0] Y[1,1] Y[2,0] Y[2,1] Y[3,0] Y[3,1] ]
|
|
|
|
These bit16_dt vectors represent a 4x2 matrix. Hence, in matrix form it
|
|
looks like the following:
|
|
|
|
X = [ X[0,0] X[0,1]
|
|
X[1,0] X[1,1]
|
|
X[2,0] X[2,1]
|
|
X[3,0] X[3,1] ]
|
|
|
|
The outer product instruction: xvbf16ger2 (bfloat16 outer product)
|
|
|
|
Syntax:
|
|
|
|
xvbf16ger2 ACCUMULATOR A, VECTOR X, VECTOR Y
|
|
|
|
Semantics:
|
|
|
|
A = X * Y^T
|
|
|
|
The generic packing routine would load 8 elements from the same column.
|
|
This causes an issue since the instruction expects the vector to be a
|
|
4x2 matrix where the data is packed in contiguous order. Thus, we must make
|
|
a packing routine that will interleave the matrix data. Making it so
|
|
that when we load the 8 contiguous elements from A, it will represent
|
|
a 4x2 section of the matrix.
|
|
|
|
*/
|
|
|
|
#include "pack_a_templates.h"
|
|
#include "pack_b_templates.h"
|
|
#include "bli_sandbox.h"
|
|
|
|
// 16 bit routines
|
|
BIT16_PACK_A(sb, bfloat16);
|
|
BIT16_PACK_B(sb, bfloat16);
|
|
BIT16_PACK_A(sh, float16);
|
|
BIT16_PACK_B(sh, float16);
|
|
BIT16_PACK_A(i16, int16_t);
|
|
BIT16_PACK_B(i16, int16_t);
|
|
|
|
// 8 bit
|
|
BIT8_PACK_A(i8, int8_t);
|
|
BIT8_PACK_B(i8, int8_t);
|
|
|
|
// 4 bit
|
|
BIT4_PACK_A(i4, nibbles);
|
|
BIT4_PACK_B(i4, nibbles);
|
|
|