Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

src: cpu: aarch64: injectors: eltwise_injector - improve gelu performance for block size 16 #2072

Merged
merged 1 commit into from
Oct 3, 2024
Merged

src: cpu: aarch64: injectors: eltwise_injector - improve gelu performance for block size 16 #2072

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
291 changes: 289 additions & 2 deletions src/cpu/aarch64/injectors/jit_uni_eltwise_injector.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
/*******************************************************************************
* Copyright 2019-2023 Intel Corporation
* Copyright 2021-2023 FUJITSU LIMITED
* Copyright 2021-2024 FUJITSU LIMITED
* Copyright 2022 Arm Ltd. and affiliates
*
* Licensed under the Apache License, Version 2.0 (the "License");
Expand Down Expand Up @@ -918,10 +918,87 @@ void jit_uni_eltwise_injector_f32<isa>::log_compute_vector_fwd(
}
h->L(exitL);
}
template <cpu_isa_t isa>
void jit_uni_eltwise_injector_f32<
isa>::gelu_erf_minimax_approx_compute_vector_fwd(const TRegS &vmm_src) {
if (isa != sve_512) { // TODO: change this condition based on cpu id.
return;
}

// register mapping
TRegS vmm_pol = vmm_aux0;
TRegS vmm_src_pos = vmm_aux1;
TRegS vmm_indices = vmm_aux2;
TRegS vmm_tmp = vmm_aux3; // this is for immediate read after write

auto gather_coefficient
= [&](TRegS vmm_coeff, int coeff_idx, TRegS vmm_pol_idx) {
// we actually have 25 polynomials but pad to avoid unaligned accesses/
int gelu_erf_n_polynomials = 32;
h->add_imm(h->X_TMP_1, x_table,
table_off(gelu_erf_minimax_pol,
coeff_idx * gelu_erf_n_polynomials),
h->X_TMP_0);
h->ld1w(ZRegS(IDX(vmm_coeff)), p_all / T_z,
ptr(h->X_TMP_1, ZRegS(IDX(vmm_pol_idx)), SXTW));
};

// we use the erf function symmetry erf(-x) = -erf(x)
// So we make x positive, we will reapply the sign after erf evaluation
h->fabs(vmm_src_pos, p_all / T_z, vmm_src);

// Compute indices for table lookup
h->add(vmm_indices, vmm_src_pos,
ZRegS(IDX(table_val(gelu_erf_idx_bias, z_tmp, 0))));

// An arithmetic shift is needed to properly map denormals to
// their polynomial. we shift by 21 as we use 2 bits of mantissa
// for indexing.
h->asr(ZRegS(IDX(vmm_indices)), ZRegS(IDX(vmm_indices)), 21);

// Apply special rules
h->smax(vmm_indices, p_all / T_z,
ZRegS(IDX(table_val(gelu_erf_one, z_tmp))));
h->smin(vmm_indices, p_all / T_z,
ZRegS(IDX(table_val(gelu_erf_twenty_four, z_tmp))));

// We have to check
// index = x_pos > rbound ? 23 : index;
// for erf to return -1/1 when we should.
h->fcmlt(p_mask.s, p_all / T_z, vmm_src_pos,
ZRegS(IDX(table_val(gelu_erf_rbound, z_tmp))));
h->sel(vmm_indices, p_mask, vmm_indices,
ZRegS(IDX(table_val(gelu_erf_twenty_three, z_tmp))));

// Adjusting indices
h->mul(ZRegS(IDX(vmm_indices)), sizeof(float));

// Evaluate the polynomial
gather_coefficient(vmm_pol, 5, vmm_indices);
for (int deg = 4; deg >= 0; --deg) {
gather_coefficient(vmm_tmp, deg, vmm_indices);
h->fmad(vmm_pol, p_all / T_z, vmm_src_pos, vmm_tmp);
}

// Set the sign of vmm_pol properly
h->mov(ZRegD(IDX(vmm_tmp)), ZRegD(IDX(vmm_src)));
h->and_(ZRegD(IDX(vmm_tmp)), ZRegD(IDX(vmm_tmp)),
ZRegD(IDX(table_val(sign_mask, z_tmp))));
h->eor(ZRegD(IDX(vmm_pol)), p_all / T_z, ZRegD(IDX(vmm_tmp)));

// Compute the final output
h->fadd(vmm_pol, vmm_pol, ZRegS(IDX(table_val(one, z_tmp))));
h->fmul(vmm_src, p_all / T_z, vmm_pol);
h->fmul(vmm_src, vmm_src, ZRegS(IDX(table_val(half, z_tmp))));
}
template <cpu_isa_t isa>
void jit_uni_eltwise_injector_f32<isa>::gelu_erf_compute_vector_fwd(
const TRegS &vmm_src) {

if (isa == sve_512) { // TODO: consider performance improvement for lower ISA
gelu_erf_minimax_approx_compute_vector_fwd(vmm_src);
return;
}
// Here we approximate erf(x) using the expression by
// Abramowitz and Stegun from ``Handbook of Mathematical
// Functions''
Expand Down Expand Up @@ -1703,6 +1780,215 @@ void jit_uni_eltwise_injector_f32<isa>::register_table_entries() {
{gelu_erf_pol, {0xbfba00e3, true}}, // p4 = -1.453152027f
{gelu_erf_pol, {0x3f87dc22, true}}, // p5 = 1.061405429f
};
// gelu_erf(x) constants for direct erf approximation (formula defined)
static const table_t gelu_erf_minimax_consts {
{gelu_erf_idx_bias, {0xc21fffff, true}},
{gelu_erf_rbound, {0x40b15cee, true}},
{gelu_erf_one, {0x00000001, true}},
{gelu_erf_twenty_three, {0x00000017, true}},
{gelu_erf_twenty_four, {0x00000018, true}},
};
// gelu_erf(x) minimax polynomials for piecewise approximaxtion
static const table_t gelu_erf_minimax_polynomial {
// coefficients of degree 0
{gelu_erf_minimax_pol, {0xa6f2cb94, false}}, // -0x1.e59728p-50
{gelu_erf_minimax_pol, {0x32827792, false}}, // 0x1.04ef24p-26
{gelu_erf_minimax_pol, {0x3381cc0c, false}}, // 0x1.039818p-24
{gelu_erf_minimax_pol, {0x34523d4a, false}}, // 0x1.a47a94p-23
{gelu_erf_minimax_pol, {0x351ac44d, false}}, // 0x1.35889ap-21
{gelu_erf_minimax_pol, {0x35f36d88, false}}, // 0x1.e6db1p-20
{gelu_erf_minimax_pol, {0x36ee8229, false}}, // 0x1.dd0452p-18
{gelu_erf_minimax_pol, {0x37b8a3bb, false}}, // 0x1.714776p-16
{gelu_erf_minimax_pol, {0x3867a213, false}}, // 0x1.cf4426p-15
{gelu_erf_minimax_pol, {0x3940033b, false}}, // 0x1.800676p-13
{gelu_erf_minimax_pol, {0x3a2a5a1d, false}}, // 0x1.54b43ap-11
{gelu_erf_minimax_pol, {0x3ae35863, false}}, // 0x1.c6b0c6p-10
{gelu_erf_minimax_pol, {0x3b7828f2, false}}, // 0x1.f051e4p-9
{gelu_erf_minimax_pol, {0x3c08b14b, false}}, // 0x1.116296p-7
{gelu_erf_minimax_pol, {0x3c515ed3, false}}, // 0x1.a2bda6p-7
{gelu_erf_minimax_pol, {0xbb503236, false}}, // -0x1.a0646cp-9
{gelu_erf_minimax_pol, {0xbd8d8e5e, false}}, // -0x1.1b1cbcp-4
{gelu_erf_minimax_pol, {0xbe8abcd9, false}}, // -0x1.1579b2p-2
{gelu_erf_minimax_pol, {0xbf0c19a2, false}}, // -0x1.183344p-1
{gelu_erf_minimax_pol, {0xbeccb328, false}}, // -0x1.99665p-2
{gelu_erf_minimax_pol, {0x3e176ced, false}}, // 0x1.2ed9dap-3
{gelu_erf_minimax_pol, {0x3f470d99, false}}, // 0x1.8e1b32p-1
{gelu_erf_minimax_pol, {0x3f7abb28, false}}, // 0x1.f5765p-1
{gelu_erf_minimax_pol, {0x3f800000, false}}, // 0x1p0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
// coefficients of degree 1
{gelu_erf_minimax_pol, {0x3f4c422a, false}}, // 0x1.988454p-1
{gelu_erf_minimax_pol, {0x3f4c421f, false}}, // 0x1.98843ep-1
{gelu_erf_minimax_pol, {0x3f4c4207, false}}, // 0x1.98840ep-1
{gelu_erf_minimax_pol, {0x3f4c41cb, false}}, // 0x1.988396p-1
{gelu_erf_minimax_pol, {0x3f4c413b, false}}, // 0x1.988276p-1
{gelu_erf_minimax_pol, {0x3f4c3fad, false}}, // 0x1.987f5ap-1
{gelu_erf_minimax_pol, {0x3f4c3a2f, false}}, // 0x1.98745ep-1
{gelu_erf_minimax_pol, {0x3f4c2d40, false}}, // 0x1.985a8p-1
{gelu_erf_minimax_pol, {0x3f4c146a, false}}, // 0x1.9828d4p-1
{gelu_erf_minimax_pol, {0x3f4bc341, false}}, // 0x1.978682p-1
{gelu_erf_minimax_pol, {0x3f4ad08c, false}}, // 0x1.95a118p-1
{gelu_erf_minimax_pol, {0x3f48f8cf, false}}, // 0x1.91f19ep-1
{gelu_erf_minimax_pol, {0x3f45fac7, false}}, // 0x1.8bf58ep-1
{gelu_erf_minimax_pol, {0x3f404e07, false}}, // 0x1.809c0ep-1
{gelu_erf_minimax_pol, {0x3f3b980f, false}}, // 0x1.77301ep-1
{gelu_erf_minimax_pol, {0x3f48dff3, false}}, // 0x1.91bfe6p-1
{gelu_erf_minimax_pol, {0x3f78b21b, false}}, // 0x1.f16436p-1
{gelu_erf_minimax_pol, {0x3fbb0704, false}}, // 0x1.760e08p0
{gelu_erf_minimax_pol, {0x40019c32, false}}, // 0x1.033864p1
{gelu_erf_minimax_pol, {0x3fe536d6, false}}, // 0x1.ca6dacp0
{gelu_erf_minimax_pol, {0x3f81331e, false}}, // 0x1.02663cp0
{gelu_erf_minimax_pol, {0x3e6c8684, false}}, // 0x1.d90d08p-3
{gelu_erf_minimax_pol, {0x3c98f936, false}}, // 0x1.31f26cp-6
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x3f800000, false}}, // 0x1p0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
// coefficients of degree 2
{gelu_erf_minimax_pol, {0xb62173f4, false}}, // -0x1.42e7e8p-19
{gelu_erf_minimax_pol, {0x3735e4cf, false}}, // 0x1.6bc99ep-17
{gelu_erf_minimax_pol, {0x37f2ff89, false}}, // 0x1.e5ff12p-16
{gelu_erf_minimax_pol, {0x388c23be, false}}, // 0x1.18477cp-14
{gelu_erf_minimax_pol, {0x3917535c, false}}, // 0x1.2ea6b8p-13
{gelu_erf_minimax_pol, {0x39ab2ab0, false}}, // 0x1.56556p-12
{gelu_erf_minimax_pol, {0x3a60fadb, false}}, // 0x1.c1f5b6p-11
{gelu_erf_minimax_pol, {0x3af9b960, false}}, // 0x1.f372cp-10
{gelu_erf_minimax_pol, {0x3b6e5491, false}}, // 0x1.dca922p-9
{gelu_erf_minimax_pol, {0x3c0a4ec5, false}}, // 0x1.149d8ap-7
{gelu_erf_minimax_pol, {0x3ca5aa8c, false}}, // 0x1.4b5518p-6
{gelu_erf_minimax_pol, {0x3d2138d9, false}}, // 0x1.4271b2p-5
{gelu_erf_minimax_pol, {0x3d8737d4, false}}, // 0x1.0e6fa8p-4
{gelu_erf_minimax_pol, {0x3ddfb660, false}}, // 0x1.bf6ccp-4
{gelu_erf_minimax_pol, {0x3e0f27ab, false}}, // 0x1.1e4f56p-3
{gelu_erf_minimax_pol, {0x3d94004b, false}}, // 0x1.280096p-4
{gelu_erf_minimax_pol, {0xbe0efdeb, false}}, // -0x1.1dfbd6p-3
{gelu_erf_minimax_pol, {0xbf1d96c3, false}}, // -0x1.3b2d86p-1
{gelu_erf_minimax_pol, {0xbf89db58, false}}, // -0x1.13b6bp0
{gelu_erf_minimax_pol, {0xbf6d9897, false}}, // -0x1.db312ep-1
{gelu_erf_minimax_pol, {0xbef69fb8, false}}, // -0x1.ed3f7p-2
{gelu_erf_minimax_pol, {0xbdc4f8a8, false}}, // -0x1.89f15p-4
{gelu_erf_minimax_pol, {0xbbde6422, false}}, // -0x1.bcc844p-8
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
// coefficients of degree 3
{gelu_erf_minimax_pol, {0xbe081a19, false}}, // -0x1.103432p-3
{gelu_erf_minimax_pol, {0xbe084570, false}}, // -0x1.108aep-3
{gelu_erf_minimax_pol, {0xbe08639b, false}}, // -0x1.10c736p-3
{gelu_erf_minimax_pol, {0xbe089837, false}}, // -0x1.11306ep-3
{gelu_erf_minimax_pol, {0xbe08f409, false}}, // -0x1.11e812p-3
{gelu_erf_minimax_pol, {0xbe09ab95, false}}, // -0x1.13572ap-3
{gelu_erf_minimax_pol, {0xbe0b66d0, false}}, // -0x1.16cdap-3
{gelu_erf_minimax_pol, {0xbe0e400a, false}}, // -0x1.1c8014p-3
{gelu_erf_minimax_pol, {0xbe124df8, false}}, // -0x1.249bfp-3
{gelu_erf_minimax_pol, {0xbe1bde02, false}}, // -0x1.37bc04p-3
{gelu_erf_minimax_pol, {0xbe2f19c9, false}}, // -0x1.5e3392p-3
{gelu_erf_minimax_pol, {0xbe4931bf, false}}, // -0x1.92637ep-3
{gelu_erf_minimax_pol, {0xbe685fbc, false}}, // -0x1.d0bf78p-3
{gelu_erf_minimax_pol, {0xbe89c95f, false}}, // -0x1.1392bep-2
{gelu_erf_minimax_pol, {0xbe96cbca, false}}, // -0x1.2d9794p-2
{gelu_erf_minimax_pol, {0xbe8044aa, false}}, // -0x1.008954p-2
{gelu_erf_minimax_pol, {0xbe0550f2, false}}, // -0x1.0aa1e4p-3
{gelu_erf_minimax_pol, {0x3dcfd6a1, false}}, // 0x1.9fad42p-4
{gelu_erf_minimax_pol, {0x3e94c826, false}}, // 0x1.29904cp-2
{gelu_erf_minimax_pol, {0x3e79345f, false}}, // 0x1.f268bep-3
{gelu_erf_minimax_pol, {0x3decec91, false}}, // 0x1.d9d922p-4
{gelu_erf_minimax_pol, {0x3ca46568, false}}, // 0x1.48cadp-6
{gelu_erf_minimax_pol, {0x3aa1e00a, false}}, // 0x1.43c014p-10
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
// coefficients of degree 4
{gelu_erf_minimax_pol, {0xba3d61db, false}}, // -0x1.7ac3b6p-11
{gelu_erf_minimax_pol, {0x39f097a3, false}}, // 0x1.e12f46p-12
{gelu_erf_minimax_pol, {0x3a5845dc, false}}, // 0x1.b08bb8p-11
{gelu_erf_minimax_pol, {0x3ab1fa35, false}}, // 0x1.63f46ap-10
{gelu_erf_minimax_pol, {0x3b0cefb8, false}}, // 0x1.19df7p-9
{gelu_erf_minimax_pol, {0x3b653ab6, false}}, // 0x1.ca756cp-9
{gelu_erf_minimax_pol, {0x3bcae527, false}}, // 0x1.95ca4ep-8
{gelu_erf_minimax_pol, {0x3c221712, false}}, // 0x1.442e24p-7
{gelu_erf_minimax_pol, {0x3c6c5840, false}}, // 0x1.d8b08p-7
{gelu_erf_minimax_pol, {0x3cc0a703, false}}, // 0x1.814e06p-6
{gelu_erf_minimax_pol, {0x3d1dcc19, false}}, // 0x1.3b9832p-5
{gelu_erf_minimax_pol, {0x3d63656d, false}}, // 0x1.c6cadap-5
{gelu_erf_minimax_pol, {0x3d955907, false}}, // 0x1.2ab20ep-4
{gelu_erf_minimax_pol, {0x3dbf9910, false}}, // 0x1.7f322p-4
{gelu_erf_minimax_pol, {0x3dd53f69, false}}, // 0x1.aa7ed2p-4
{gelu_erf_minimax_pol, {0x3db7dcef, false}}, // 0x1.6fb9dep-4
{gelu_erf_minimax_pol, {0x3d639ebe, false}}, // 0x1.c73d7cp-5
{gelu_erf_minimax_pol, {0xba6ede48, false}}, // -0x1.ddbc9p-11
{gelu_erf_minimax_pol, {0xbd22be69, false}}, // -0x1.457cd2p-5
{gelu_erf_minimax_pol, {0xbd041cf1, false}}, // -0x1.0839e2p-5
{gelu_erf_minimax_pol, {0xbc64f5ab, false}}, // -0x1.c9eb56p-7
{gelu_erf_minimax_pol, {0xbb097a32, false}}, // -0x1.12f464p-9
{gelu_erf_minimax_pol, {0xb8ebf380, false}}, // -0x1.d7e7p-14
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
// coefficients of degree 5
{gelu_erf_minimax_pol, {0x3cb7d80c, false}}, // 0x1.6fb018p-6
{gelu_erf_minimax_pol, {0x3c9b6050, false}}, // 0x1.36c0ap-6
{gelu_erf_minimax_pol, {0x3c978d11, false}}, // 0x1.2f1a22p-6
{gelu_erf_minimax_pol, {0x3c92e850, false}}, // 0x1.25d0ap-6
{gelu_erf_minimax_pol, {0x3c8d058b, false}}, // 0x1.1a0b16p-6
{gelu_erf_minimax_pol, {0x3c848454, false}}, // 0x1.0908a8p-6
{gelu_erf_minimax_pol, {0x3c6cd623, false}}, // 0x1.d9ac46p-7
{gelu_erf_minimax_pol, {0x3c4c824b, false}}, // 0x1.990496p-7
{gelu_erf_minimax_pol, {0x3c2a7935, false}}, // 0x1.54f26ap-7
{gelu_erf_minimax_pol, {0x3be0b390, false}}, // 0x1.c1672p-8
{gelu_erf_minimax_pol, {0x3b0651ac, false}}, // 0x1.0ca358p-9
{gelu_erf_minimax_pol, {0xbb232f53, false}}, // -0x1.465ea6p-9
{gelu_erf_minimax_pol, {0xbbd42fa0, false}}, // -0x1.a85f4p-8
{gelu_erf_minimax_pol, {0xbc2c5366, false}}, // -0x1.58a6ccp-7
{gelu_erf_minimax_pol, {0xbc492c9e, false}}, // -0x1.92593cp-7
{gelu_erf_minimax_pol, {0xbc2a7aa6, false}}, // -0x1.54f54cp-7
{gelu_erf_minimax_pol, {0xbbd55d04, false}}, // -0x1.aaba08p-8
{gelu_erf_minimax_pol, {0xba823a76, false}}, // -0x1.0474ecp-10
{gelu_erf_minimax_pol, {0x3b102aa8, false}}, // 0x1.20555p-9
{gelu_erf_minimax_pol, {0x3ae25a7e, false}}, // 0x1.c4b4fcp-10
{gelu_erf_minimax_pol, {0x3a31f792, false}}, // 0x1.63ef24p-11
{gelu_erf_minimax_pol, {0x38b84375, false}}, // 0x1.7086eap-14
{gelu_erf_minimax_pol, {0x3689bb5a, false}}, // 0x1.1376b4p-18
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
{gelu_erf_minimax_pol, {0x00000000, false}}, // 0 padd
};

// This object takes care about which constants and polynomials to include.
struct need_t {
Expand Down Expand Up @@ -1780,7 +2066,8 @@ void jit_uni_eltwise_injector_f32<isa>::register_table_entries() {
if (need.gelu_tanh()) push_entries_of(gelu_tanh_consts);
if (need.gelu_erf()) push_entries_of(gelu_erf_consts);
if (need.gelu_erf()) push_entries_of(gelu_erf_polynomial);

if (need.gelu_erf()) push_entries_of(gelu_erf_minimax_consts);
if (need.gelu_erf()) push_entries_of(gelu_erf_minimax_polynomial);
// Now that we registered the entries, we set the offsets. No
// entries should be registered after this point. This allows to
// expect the same order when injecting the table entries in
Expand Down
Loading
Loading