Skip to content

Commit

Permalink
[software] Add MMSE fp8
Browse files Browse the repository at this point in the history
  • Loading branch information
mbertuletti committed Oct 16, 2024
1 parent d05ce74 commit b8720eb
Show file tree
Hide file tree
Showing 11 changed files with 794 additions and 379 deletions.
84 changes: 32 additions & 52 deletions software/apps/baremetal/mimo_mmse_f16/main.c
Original file line number Diff line number Diff line change
Expand Up @@ -18,6 +18,7 @@
#include "baremetal/mempool_mimo_mmse_f16s.h"

#include "data_mimo_mmse_f16.h"
#define ZF (0) // When asserted use zero-forcing
#define NUM_BANKS (BANKING_FACTOR * NUM_CORES)
//#define DOUBLE_BUFFERING

Expand All @@ -35,14 +36,12 @@
#ifndef DOUBLE_BUFFERING

__fp16 l1_H[2 * N_TX * N_RX * N_ITR]
__attribute__((aligned(NUM_BANKS * sizeof(int32_t)), section(".l1_prio")));
__attribute__((aligned(sizeof(int32_t)), section(".l1_prio")));
__fp16 l1_G[2 * N_TX * N_TX * N_ITR]
__attribute__((aligned(NUM_BANKS * sizeof(int32_t)), section(".l1_prio")));
__attribute__((aligned(sizeof(int32_t)), section(".l1_prio")));
__fp16 l1_L[2 * N_TX * N_TX * N_ITR]
__attribute__((aligned(NUM_BANKS * sizeof(int32_t)), section(".l1_prio")));
__attribute__((aligned(sizeof(int32_t)), section(".l1_prio")));

uint32_t l1_beamgroups[N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));
__fp16 l1_S[2 * N_TX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));
__fp16 l1_y[2 * N_RX * N_ITR]
Expand All @@ -58,8 +57,10 @@ __fp16 l1_x[2 * N_TX * N_ITR]
int main() {

uint32_t core_id = mempool_get_core_id();
#ifndef BANSHEE
uint32_t num_cores = mempool_get_core_count();
mempool_barrier_init(core_id); // Initialize barrier and synchronize
#endif

#ifdef BANSHEE
/* Initialize matrices */
Expand All @@ -71,41 +72,38 @@ int main() {
(*(uint32_t *)&l1_y[2 * i]) = *(uint32_t *)&l2_y[2 * i];
}
for (uint32_t i = 0; i < N_TX * N_ITR; i++) {
(*(uint32_t *)&l1_S[2 * i]) = *(uint32_t *)&l2_Sigma[2 * i];
(*(uint32_t *)&l1_S[2 * i]) = *(uint32_t *)&l2_S[2 * i];
}
}
mempool_barrier(num_cores);
#else
/* Initialize matrices */
if (core_id == 0) {
dma_memcpy_blocking(l1_beamgroups, l2_beamgroups, N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1_H, l2_H, N_TX * N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1_y, l2_y, N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1_S, l2_Sigma, N_TX * N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1_H, l2_H, 2 * N_TX * N_RX * N_ITR * sizeof(int16_t));
dma_memcpy_blocking(l1_y, l2_y, 2 * N_RX * N_ITR * sizeof(int16_t));
dma_memcpy_blocking(l1_S, l2_S, 2 * N_TX * N_ITR * sizeof(int16_t));
}
mempool_barrier(num_cores);
#endif

#ifdef BANSHEE
#ifdef SINGLE
/* Benchmark */
if (core_id == 0) {
mempool_start_benchmark();
for (uint32_t itr = 0; itr < N_ITR; itr++) {
__fp16 *PtrH = l1_H + itr * (2 * N_TX * N_RX);
__fp16 *Ptry = l1_y + itr * (2 * N_RX);
__fp16 *PtrSigma = l1_S + itr * N_TX;
__fp16 *PtrS = l1_S + itr * (2 * N_TX);
__fp16 *PtrG = l1_G + itr * (2 * N_TX * N_TX);
__fp16 *PtrL = l1_L + itr * (2 * N_TX * N_TX);
__fp16 *Ptry2 = y2 + itr * (2 * N_TX);
__fp16 *Ptry3 = y3 + itr * (2 * N_TX);
__fp16 *Ptrx = l1_x + itr * (2 * N_TX);

#ifdef VEC
mempool_hermitian_f16vecs(PtrH, PtrG, PtrSigma, N_RX, N_TX);
mempool_hermitian_f16vecs(PtrH, PtrG, PtrS, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f16vecs(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16vecs(PtrG, PtrL, N_TX);
#else
mempool_hermitian_f16s(PtrH, PtrG, PtrSigma, N_RX, N_TX, 0, 0);
mempool_hermitian_f16s(PtrH, PtrG, PtrS, N_RX, N_TX, 0, ZF);
mempool_MVP_conjtransp_f16s(PtrH, Ptry, Ptry2, N_RX, N_TX, 0);
mempool_cholesky_f16s(PtrG, PtrL, N_TX);
#endif
Expand All @@ -114,20 +112,6 @@ int main() {
}
mempool_stop_benchmark();
}
mempool_barrier(num_cores);
#endif

#ifdef SINGLE
if (core_id == 0) {
mempool_start_benchmark();
mempool_hermitian_f16s(l1_H, l1_G, l1_S, N_RX, N_TX, 0, 0);
mempool_MVP_conjtransp_f16s(l1_H, l1_y, y2, N_RX, N_TX, 0);
mempool_cholesky_f16vecs(l1_G, l1_L, N_TX);
mempool_Ltrisol_f16s(l1_L, y2, y3, N_TX);
mempool_Lttrisol_f16s(l1_L, y3, l1_x, N_TX);
mempool_stop_benchmark();
}
mempool_barrier(num_cores);
#endif

#ifdef PARALLEL
Expand All @@ -136,27 +120,24 @@ int main() {
for (uint32_t itr = core_id; itr < N_ITR; itr += num_cores) {
__fp16 *PtrH = l1_H + itr * (2 * N_TX * N_RX);
__fp16 *Ptry = l1_y + itr * (2 * N_RX);
__fp16 *PtrSigma = l1_S + itr * (2 * N_TX);
__fp16 *PtrS = l1_S + itr * (2 * N_TX);
// Auxiliary vectors
__fp16 *PtrG = l1_G + itr * (2 * N_TX * N_TX);
__fp16 *PtrL = l1_L + itr * (2 * N_TX * N_TX);
__fp16 *Ptry2 = y2 + itr * (2 * N_TX);
__fp16 *Ptry3 = y3 + itr * (2 * N_TX);
__fp16 *Ptrx = l1_x + itr * (2 * N_TX);
// Serial beamgroup Beamgroups loop
uint32_t N_bg = l1_beamgroups[itr];
uint32_t N_TX_bg = N_TX / N_bg;
for (uint32_t itr_bg = 0; itr_bg < N_bg; itr_bg++) {
mempool_hermitian_f16vecs(PtrH, PtrG, PtrSigma, N_RX, N_TX_bg);
mempool_MVP_conjtransp_f16vecs(PtrH, Ptry, Ptry2, N_RX, N_TX_bg);
mempool_cholesky_f16vecs(PtrG, PtrL, N_TX_bg);
mempool_Ltrisol_f16s(PtrL, Ptry2, Ptry3, N_TX_bg);
mempool_Lttrisol_f16s(PtrL, Ptry3, Ptrx, N_TX_bg);
// Shift over the subsequent beamgroup
PtrH += 2 * itr_bg * N_TX_bg * N_RX;
PtrSigma += 2 * itr_bg * N_TX_bg;
Ptrx += 2 * itr_bg * N_TX_bg;
}
#ifdef VEC
mempool_hermitian_f16vecs(PtrH, PtrG, PtrS, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f16vecs(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16vecs(PtrG, PtrL, N_TX);
#else
mempool_hermitian_f16s(PtrH, PtrG, PtrS, N_RX, N_TX, 0, ZF);
mempool_MVP_conjtransp_f16s(PtrH, Ptry, Ptry2, N_RX, N_TX, 0);
mempool_cholesky_f16s(PtrG, PtrL, N_TX);
#endif
mempool_Ltrisol_f16s(PtrL, Ptry2, Ptry3, N_TX);
mempool_Lttrisol_f16s(PtrL, Ptry3, Ptrx, N_TX);
}
mempool_barrier(num_cores);
mempool_stop_benchmark();
Expand All @@ -170,11 +151,10 @@ int main() {
printf("RES=%04x\n", x);
}
}
mempool_barrier(num_cores);
#else
mempool_check_f16(l1_x, l2_x, 2 * N_TX, 0.01f, 0);
mempool_barrier(num_cores);
#endif

return 0;
}

Expand Down Expand Up @@ -233,7 +213,7 @@ int main() {
if (core_id == 0) {
dma_memcpy_blocking(l1A_H, l2_H, N_TX * N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1A_y, l2_y, N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1A_S, l2_Sigma, N_TX * N_ITR * sizeof(int32_t));
dma_memcpy_blocking(l1A_S, l2_S, N_TX * N_ITR * sizeof(int32_t));
}
mempool_barrier(num_cores);

Expand Down Expand Up @@ -269,7 +249,7 @@ int main() {
dma_memcpy_nonblocking(trsf_H, l2_H,
N_TX * N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_nonblocking(trsf_y, l2_y, N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_nonblocking(trsf_S, l2_Sigma, N_TX * N_ITR * sizeof(int32_t));
dma_memcpy_nonblocking(trsf_S, l2_S, N_TX * N_ITR * sizeof(int32_t));
if (round >= 1) // Transfer to L2 is done only if not the first round
dma_memcpy_nonblocking(l2_x, trsf_x, (N_TX * N_ITR) * sizeof(int32_t));
}
Expand All @@ -284,7 +264,7 @@ int main() {
__fp16 *PtrL = L + itr * (2 * N_TX * N_TX);
__fp16 *Ptry2 = y2 + itr * (2 * N_TX);
__fp16 *Ptry3 = y3 + itr * (2 * N_TX);
mempool_hermitian_f16vecs(PtrH, PtrG, PtrS, N_RX, N_TX);
mempool_hermitian_f16vecs(PtrH, PtrG, PtrS, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f16vecs(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16vecs(PtrG, PtrL, N_TX);
mempool_Ltrisol_f16s(PtrL, Ptry2, Ptry3, N_TX);
Expand All @@ -311,7 +291,7 @@ int main() {
__fp16 *PtrS = cmpt_S + itr * (2 * N_TX);
__fp16 *PtrG = G + itr * (2 * N_TX * N_TX);
__fp16 *Ptry2 = y2 + itr * (2 * N_TX);
mempool_hermitian_f16vecs(PtrH, PtrG, PtrS, N_RX, N_TX);
mempool_hermitian_f16vecs(PtrH, PtrG, PtrS, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f16vecs(PtrH, Ptry, Ptry2, N_RX, N_TX);
}
mempool_log_barrier(2, core_id);
Expand All @@ -321,7 +301,7 @@ int main() {
dma_memcpy_nonblocking(trsf_H, l2_H,
N_TX * N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_nonblocking(trsf_y, l2_y, N_RX * N_ITR * sizeof(int32_t));
dma_memcpy_nonblocking(trsf_S, l2_Sigma, N_TX * N_ITR * sizeof(int32_t));
dma_memcpy_nonblocking(trsf_S, l2_S, N_TX * N_ITR * sizeof(int32_t));
if (round >= 1) // Transfer to L2 is done only if not the
dma_memcpy_nonblocking(l2_x, trsf_x, (N_TX * N_ITR) * sizeof(int32_t));
}
Expand Down
152 changes: 152 additions & 0 deletions software/apps/baremetal/mimo_mmse_f8/main.c
Original file line number Diff line number Diff line change
@@ -0,0 +1,152 @@
// Copyright 2021 ETH Zurich and University of Bologna.
// Licensed under the Apache License, Version 2.0, see LICENSE for details.
// SPDX-License-Identifier: Apache-2.0

// Author: Marco Bertuletti, ETH Zurich

#include <math.h>

#include "dma.h"
#include "encoding.h"
#include "printf.h"
#include "runtime.h"
#include "synchronization.h"

#include "baremetal/mempool_checks.h"
#include "baremetal/mempool_cholesky_f16s.h"
#include "baremetal/mempool_cholesky_f8s.h"
#include "baremetal/mempool_linearsolver_f16s.h"
#include "baremetal/mempool_linearsolver_f8s.h"
#include "baremetal/mempool_mimo_mmse_f8s.h"

#include "data_mimo_mmse_f8.h"
#define ZF (0) // When asserted use zero-forcing
#define NUM_BANKS (BANKING_FACTOR * NUM_CORES)

__fp8 l1_H[2 * N_TX * N_RX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1_prio")));
__fp8 l1_S[2 * N_TX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));
__fp8 l1_y[2 * N_RX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));

__fp16 l1_G[2 * N_TX * N_TX * N_ITR]
__attribute__((aligned(NUM_BANKS * sizeof(int32_t)), section(".l1_prio")));
__fp16 l1_L[2 * N_TX * N_TX * N_ITR]
__attribute__((aligned(NUM_BANKS * sizeof(int32_t)), section(".l1_prio")));
__fp16 y2[2 * N_TX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));
__fp16 y3[2 * N_TX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));
__fp16 l1_x[2 * N_TX * N_ITR]
__attribute__((aligned(sizeof(int32_t)), section(".l1")));

// Driver program
int main() {

uint32_t core_id = mempool_get_core_id();
#ifndef BANSHEE
uint32_t num_cores = mempool_get_core_count();
mempool_barrier_init(core_id); // Initialize barrier and synchronize
#endif

/* Initialize matrices */
if (core_id == 0) {
#ifdef BANSHEE
for (uint32_t i = 0; i < 2 * N_RX * N_TX * N_ITR; i++) {
l1_H[i] = l2_H[i];
}
for (uint32_t i = 0; i < 2 * N_RX * N_ITR; i++) {
l1_y[i] = l2_y[i];
}
for (uint32_t i = 0; i < 2 * N_TX * N_ITR; i++) {
l1_S[i] = l2_S[i];
}
#else
dma_memcpy_blocking(l1_H, l2_H, 2 * N_TX * N_RX * N_ITR * sizeof(int8_t));
dma_memcpy_blocking(l1_y, l2_y, 2 * N_RX * N_ITR * sizeof(int8_t));
dma_memcpy_blocking(l1_S, l2_S, 2 * N_TX * N_ITR * sizeof(int16_t));
#endif
printf("Data transferred\n");
}
#ifndef BANSHEE
mempool_barrier(num_cores);
#endif

#ifdef SINGLE
/* Benchmark */
if (core_id == 0) {
mempool_start_benchmark();
for (uint32_t itr = 0; itr < N_ITR; itr++) {
// 8b inputs
__fp8 *PtrH = l1_H + itr * (2 * N_TX * N_RX);
__fp8 *Ptry = l1_y + itr * (2 * N_RX);
__fp8 *PtrS = l1_S + itr * (2 * N_TX);
// 16b intermediates and outputs
__fp16 *PtrG = l1_G + itr * (2 * N_TX * N_TX);
__fp16 *PtrL = l1_L + itr * (2 * N_TX * N_TX);
__fp16 *Ptry2 = y2 + itr * (2 * N_TX);
__fp16 *Ptry3 = y3 + itr * (2 * N_TX);
__fp16 *Ptrx = l1_x + itr * (2 * N_TX);
// 8b
#ifdef VEC
mempool_hermitian_f8vecs(PtrH, PtrS, PtrG, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f8vecs(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16vecs(PtrG, PtrL, N_TX);
#else
mempool_hermitian_f8s(PtrH, PtrS, PtrG, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f8s(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16s(PtrG, PtrL, N_TX);
#endif
// 16b
mempool_Ltrisol_f16s(PtrL, Ptry2, Ptry3, N_TX);
mempool_Lttrisol_f16s(PtrL, Ptry3, Ptrx, N_TX);
}
mempool_stop_benchmark();
}
#endif

#ifdef PARALLEL
mempool_start_benchmark();
// Parallel subcarrier loop
for (uint32_t itr = core_id; itr < N_ITR; itr += num_cores) {
__fp8 *PtrH = l1_H + itr * (2 * N_TX * N_RX);
__fp8 *Ptry = l1_y + itr * (2 * N_RX);
__fp8 *PtrS = l1_S + itr * (2 * N_TX);
// Auxiliary vectors
__fp16 *PtrG = l1_G + itr * (2 * N_TX * N_TX);
__fp16 *PtrL = l1_L + itr * (2 * N_TX * N_TX);
__fp16 *Ptry2 = y2 + itr * (2 * N_TX);
__fp16 *Ptry3 = y3 + itr * (2 * N_TX);
__fp16 *Ptrx = l1_x + itr * (2 * N_TX);
#ifdef VEC
mempool_hermitian_f8vecs(PtrH, PtrS, PtrG, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f8vecs(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16vecs(PtrG, PtrL, N_TX);
#else
mempool_hermitian_f8s(PtrH, PtrS, PtrG, N_RX, N_TX, ZF);
mempool_MVP_conjtransp_f8s(PtrH, Ptry, Ptry2, N_RX, N_TX);
mempool_cholesky_f16s(PtrG, PtrL, N_TX);
#endif
// 16b
mempool_Ltrisol_f16s(PtrL, Ptry2, Ptry3, N_TX);
mempool_Lttrisol_f16s(PtrL, Ptry3, Ptrx, N_TX);
}
mempool_barrier(num_cores);
mempool_stop_benchmark();
#endif

#ifdef BANSHEE
// Check the result
if (core_id == 0) {
for (uint32_t i = 0; i < 2 * N_TX * N_ITR; i++) {
uint32_t x = (*(uint32_t *)&l1_x[i]) & 0x0000FFFF;
printf("RES=%04x\n", x);
}
}
#else
mempool_barrier(num_cores);
#endif

return 0;
}
4 changes: 1 addition & 3 deletions software/data/data_mimo_mmse_f16.h.tpl
Original file line number Diff line number Diff line change
Expand Up @@ -34,15 +34,13 @@

// Inputs

uint32_t l2_beamgroups[${N_itr}] = ${array_to_str(beamgroups)};

__fp16 l2_H[${2 * N_tx * N_rx * N_itr}] = ${array_to_cstr(H)};

__fp16 __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_G[${2 * N_tx * N_tx * N_itr}] = ${array_to_cstr(G)};

__fp16 __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_y[${2 * N_rx * N_itr}] = ${array_to_cstr(y)};

__fp16 __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_Sigma[${2 * N_tx * N_itr}] = ${array_to_cstr(sigma)};
__fp16 __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_S[${2 * N_tx * N_itr}] = ${array_to_cstr(N)};

// Outputs

Expand Down
2 changes: 1 addition & 1 deletion software/data/data_mimo_mmse_f32.h.tpl
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@ float __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_G[${2 * N_tx

float __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_y[${2 * N_rx * N_itr}] = ${array_to_cstr(y)};

float __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_Sigma[${N_tx * N_itr}] = ${array_to_cstr(sigma)};
float __attribute__((aligned(sizeof(int32_t)), section(".l2"))) l2_Sigma[${N_tx * N_itr}] = ${array_to_cstr(N)};

// Outputs

Expand Down
Loading

0 comments on commit b8720eb

Please sign in to comment.