wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

mutiple fixes: parallel printf, fixed cycle in cache, opencl refactored vecadd and sgemm, regen opencl kernels with hard-float, fixed vortex io bus interface, fixed dpi floats APi to support multicore mode, make vlsim multicore default, make rtlsim multi-core default, removed POCL binaries from repository, updated Makefiles to use external POCL

Browse Source
This commit is contained in:
Blaise Tine
2020-09-19 14:45:42 -04:00
parent 80f929eb61
commit f6f95e0c46
146 changed files with 116779 additions and 194258 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
hw/rtl/fp_cores/VX_fp_addmul.v
View File

@@ -161,10 +161,16 @@ module VX_fp_addmul #(
defparam mac_fp_mul.adder_input_clock = "none";
defparam mac_fp_mul.accum_adder_clock = "none";
`else
integer fadd_h, fsub_h, fmul_h;
initial begin
fadd_h = dpi_register();
fsub_h = dpi_register();
fmul_h = dpi_register();
end
always @(posedge clk) begin
dpi_fadd(0*LANES+i, enable, dataa[i], datab[i], result_add);
dpi_fsub(1*LANES+i, enable, dataa[i], datab[i], result_sub);
dpi_fmul(2*LANES+i, enable, dataa[i], datab[i], result_mul);
dpi_fadd(fadd_h, enable, dataa[i], datab[i], result_add);
dpi_fsub(fsub_h, enable, dataa[i], datab[i], result_sub);
dpi_fmul(fmul_h, enable, dataa[i], datab[i], result_mul);
end
`endif

6
hw/rtl/fp_cores/VX_fp_div.v
View File

@@ -39,8 +39,12 @@ module VX_fp_div #(
.q (result[i])
);
`else
integer fdiv_h;
initial begin
fdiv_h = dpi_register();
end
always @(posedge clk) begin
dpi_fdiv(8*LANES+i, enable, dataa[i], datab[i], result[i]);
dpi_fdiv(fdiv_h, enable, dataa[i], datab[i], result[i]);
end
`endif
end

9
hw/rtl/fp_cores/VX_fp_ftoi.v
View File

@@ -53,9 +53,14 @@ module VX_fp_ftoi #(
.q (result_u)
);
`else
integer ftoi_h, ftou_h;
initial begin
ftoi_h = dpi_register();
ftou_h = dpi_register();
end
always @(posedge clk) begin
dpi_ftoi(10*LANES+i, enable, dataa[i], result_s);
dpi_ftou(11*LANES+i, enable, dataa[i], result_u);
dpi_ftoi(ftoi_h, enable, dataa[i], result_s);
dpi_ftou(ftou_h, enable, dataa[i], result_u);
end
`endif

9
hw/rtl/fp_cores/VX_fp_itof.v
View File

@@ -53,9 +53,14 @@ module VX_fp_itof #(
.q (result_u)
);
`else
integer itof_h, utof_h;
initial begin
itof_h = dpi_register();
utof_h = dpi_register();
end
always @(posedge clk) begin
dpi_itof(12*LANES+i, enable, dataa[i], result_s);
dpi_utof(13*LANES+i, enable, dataa[i], result_u);
dpi_itof(itof_h, enable, dataa[i], result_s);
dpi_utof(utof_h, enable, dataa[i], result_u);
end
`endif

9
hw/rtl/fp_cores/VX_fp_madd.v
View File

@@ -121,9 +121,14 @@ module VX_fp_madd #(
defparam mac_fp_msub.adder_input_clock = "0";
defparam mac_fp_msub.accum_adder_clock = "none";
`else
integer fmadd_h, fmsub_h;
initial begin
fmadd_h = dpi_register();
fmsub_h = dpi_register();
end
always @(posedge clk) begin
dpi_fmadd(3*LANES+i, enable, dataa[i], datab[i], datac[i], result_madd);
dpi_fmsub(4*LANES+i, enable, dataa[i], datab[i], datac[i], result_msub);
dpi_fmadd(fmadd_h, enable, dataa[i], datab[i], datac[i], result_madd);
dpi_fmsub(fmsub_h, enable, dataa[i], datab[i], datac[i], result_msub);
end
`endif

6
hw/rtl/fp_cores/VX_fp_sqrt.v
View File

@@ -37,8 +37,12 @@ module VX_fp_sqrt #(
.q (result[i])
);
`else
integer fsqrt_h;
initial begin
fsqrt_h = dpi_register();
end
always @(posedge clk) begin
dpi_fsqrt(9*LANES+i, enable, dataa[i], result[i]);
dpi_fsqrt(fsqrt_h, enable, dataa[i], result[i]);
end
`endif
end

19
hw/rtl/fp_cores/svdpi/float_dpi.cpp
View File

@@ -3,11 +3,13 @@
#include <unordered_map>
#include <vector>
#include <mutex>
#include <iostream>
#include "svdpi.h"
#include "verilated_vpi.h"
#include "VX_config.h"
extern "C" {
int dpi_register();
void dpi_fadd(int inst, bool enable, int a, int b, int* result);
void dpi_fsub(int inst, bool enable, int a, int b, int* result);
void dpi_fmul(int inst, bool enable, int a, int b, int* result);
@@ -66,19 +68,28 @@ union Float_t {
class Instances {
public:
ShiftRegister& get(int inst) {
mutex_.lock();
ShiftRegister& sr = instances_[inst];
return instances_.at(inst);
}
int allocate() {
mutex_.lock();
int inst = instances_.size();
instances_.resize(inst + 1);
mutex_.unlock();
return sr;
return inst;
}
private:
std::unordered_map<int, ShiftRegister> instances_;
std::vector<ShiftRegister> instances_;
std::mutex mutex_;
};
Instances instances;
int dpi_register() {
return instances.allocate();
}
void dpi_fadd(int inst, bool enable, int a, int b, int* result) {
ShiftRegister& sr = instances.get(inst);

2
hw/rtl/fp_cores/svdpi/float_dpi.vh
View File

@@ -1,6 +1,8 @@
`ifndef FLOAT_DPI
`define FLOAT_DPI
import "DPI-C" context function int dpi_register();
import "DPI-C" context function void dpi_fadd(int inst, input logic enable, input int a, input int b, output int result);
import "DPI-C" context function void dpi_fsub(int inst, input logic enable, input int a, input int b, output int result);
import "DPI-C" context function void dpi_fmul(int inst, input logic enable, input int a, input int b, output int result);