diff options
author | Xavier Hernandez <xhernandez@datalab.es> | 2015-12-01 11:50:54 +0100 |
---|---|---|
committer | Pranith Kumar Karampuri <pkarampu@redhat.com> | 2016-09-08 10:08:25 -0700 |
commit | 593b7a83f7408e59ab7b3ef7dfc4fe4096d6e3cd (patch) | |
tree | 1a60ff784e7759c5dddb2352c5526f16e8696013 /xlators/cluster/ec/src/ec-code-avx.c | |
parent | 8a6236e4cfc7bf86e881e5c770a19040a6060ad0 (diff) |
cluster/ec: Add support for hardware acceleration
This patch implements functionalities for fast encoding/decoding
using hardware support. Currently optimized x86_64, SSE and AVX is
added.
Additionally this patch implements a caching mecanism for inverse
matrices to reduce computation time, as well as a new method for
computing the inverse that takes quadratic time instead of cubic.
Finally some unnecessary memory copies have been eliminated to
further increase performance.
Change-Id: I26c75f26fb4201bd22b51335448ea4357235065a
BUG: 1289922
Signed-off-by: Xavier Hernandez <xhernandez@datalab.es>
Reviewed-on: http://review.gluster.org/12837
Tested-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Smoke: Gluster Build System <jenkins@build.gluster.org>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
Reviewed-by: Pranith Kumar Karampuri <pkarampu@redhat.com>
Diffstat (limited to 'xlators/cluster/ec/src/ec-code-avx.c')
-rw-r--r-- | xlators/cluster/ec/src/ec-code-avx.c | 116 |
1 files changed, 116 insertions, 0 deletions
diff --git a/xlators/cluster/ec/src/ec-code-avx.c b/xlators/cluster/ec/src/ec-code-avx.c new file mode 100644 index 00000000000..92bd3e83c5e --- /dev/null +++ b/xlators/cluster/ec/src/ec-code-avx.c @@ -0,0 +1,116 @@ +/* + Copyright (c) 2015 DataLab, s.l. <http://www.datalab.es> + This file is part of GlusterFS. + + This file is licensed to you under your choice of the GNU Lesser + General Public License, version 3 or any later version (LGPLv3 or + later), or the GNU General Public License, version 2 (GPLv2), in all + cases as published by the Free Software Foundation. +*/ + +#include <errno.h> + +#include "ec-code-intel.h" + +static void +ec_code_avx_prolog(ec_code_builder_t *builder) +{ + builder->loop = builder->address; +} + +static void +ec_code_avx_epilog(ec_code_builder_t *builder) +{ + ec_code_intel_op_add_i2r(builder, 32, REG_DX); + ec_code_intel_op_add_i2r(builder, 32, REG_DI); + ec_code_intel_op_test_i2r(builder, builder->width - 1, REG_DX); + ec_code_intel_op_jne(builder, builder->loop); + + ec_code_intel_op_ret(builder, 0); +} + +static void +ec_code_avx_load(ec_code_builder_t *builder, uint32_t dst, uint32_t idx, + uint32_t bit) +{ + if (builder->linear) { + ec_code_intel_op_mov_m2avx(builder, REG_SI, REG_DX, 1, + idx * builder->width * builder->bits + + bit * builder->width, + dst); + } else { + if (builder->base != idx) { + ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8, + REG_AX); + builder->base = idx; + } + ec_code_intel_op_mov_m2avx(builder, REG_AX, REG_DX, 1, + bit * builder->width, dst); + } +} + +static void +ec_code_avx_store(ec_code_builder_t *builder, uint32_t src, uint32_t bit) +{ + ec_code_intel_op_mov_avx2m(builder, src, REG_DI, REG_NULL, 0, + bit * builder->width); +} + +static void +ec_code_avx_copy(ec_code_builder_t *builder, uint32_t dst, uint32_t src) +{ + ec_code_intel_op_mov_avx2avx(builder, src, dst); +} + +static void +ec_code_avx_xor2(ec_code_builder_t *builder, uint32_t dst, uint32_t src) +{ + ec_code_intel_op_xor_avx2avx(builder, src, dst); +} + +static void +ec_code_avx_xor3(ec_code_builder_t *builder, uint32_t dst, uint32_t src1, + uint32_t src2) +{ + ec_code_intel_op_mov_avx2avx(builder, src1, dst); + ec_code_intel_op_xor_avx2avx(builder, src2, dst); +} + +static void +ec_code_avx_xorm(ec_code_builder_t *builder, uint32_t dst, uint32_t idx, + uint32_t bit) +{ + if (builder->linear) { + ec_code_intel_op_xor_m2avx(builder, REG_SI, REG_DX, 1, + idx * builder->width * builder->bits + + bit * builder->width, + dst); + } else { + if (builder->base != idx) { + ec_code_intel_op_mov_m2r(builder, REG_SI, REG_NULL, 0, idx * 8, + REG_AX); + builder->base = idx; + } + ec_code_intel_op_xor_m2avx(builder, REG_AX, REG_DX, 1, + bit * builder->width, dst); + } +} + +static char *ec_code_avx_needed_flags[] = { + "avx2", + NULL +}; + +ec_code_gen_t ec_code_gen_avx = { + .name = "avx", + .flags = ec_code_avx_needed_flags, + .width = 32, + .prolog = ec_code_avx_prolog, + .epilog = ec_code_avx_epilog, + .load = ec_code_avx_load, + .store = ec_code_avx_store, + .copy = ec_code_avx_copy, + .xor2 = ec_code_avx_xor2, + .xor3 = ec_code_avx_xor3, + .xorm = ec_code_avx_xorm +}; |