Fixed most of the cache issues, mat_add left
This commit is contained in:
@@ -71,7 +71,7 @@ module VX_cache_miss_resrv
|
||||
output wire miss_resrv_valid_st0,
|
||||
output wire[31:0] miss_resrv_addr_st0,
|
||||
output wire[`WORD_SIZE_RNG] miss_resrv_data_st0,
|
||||
output wire[`vx_clog2(NUMBER_REQUESTS)-1:0] miss_resrv_tid_st0,
|
||||
output wire[`vx_clog2(NUMBER_REQUESTS)-1:0] miss_resrv_tid_st0,
|
||||
output wire[4:0] miss_resrv_rd_st0,
|
||||
output wire[1:0] miss_resrv_wb_st0,
|
||||
output wire[`NW_M1:0] miss_resrv_warp_num_st0,
|
||||
@@ -90,8 +90,11 @@ module VX_cache_miss_resrv
|
||||
reg[`vx_clog2(MRVQ_SIZE)-1:0] head_ptr;
|
||||
reg[`vx_clog2(MRVQ_SIZE)-1:0] tail_ptr;
|
||||
|
||||
reg[31:0] size;
|
||||
|
||||
assign miss_resrv_full = (MRVQ_SIZE != 2) && (tail_ptr+1) == head_ptr;
|
||||
|
||||
// assign miss_resrv_full = (MRVQ_SIZE != 2) && (tail_ptr+1) == head_ptr;
|
||||
assign miss_resrv_full = (MRVQ_SIZE != 2) && (size == MRVQ_SIZE);
|
||||
|
||||
|
||||
wire enqueue_possible = !miss_resrv_full;
|
||||
@@ -108,7 +111,7 @@ module VX_cache_miss_resrv
|
||||
|
||||
|
||||
wire dequeue_possible = valid_table[head_ptr] && ready_table[head_ptr];
|
||||
wire[`vx_clog2(MRVQ_SIZE)-1:0] dequeue_index = head_ptr;
|
||||
wire[`vx_clog2(MRVQ_SIZE)-1:0] dequeue_index = head_ptr;
|
||||
|
||||
assign miss_resrv_valid_st0 = (MRVQ_SIZE != 2) && dequeue_possible;
|
||||
assign miss_resrv_pc_st0 = pc_table[dequeue_index];
|
||||
@@ -126,6 +129,7 @@ module VX_cache_miss_resrv
|
||||
pc_table <= 0;
|
||||
end else begin
|
||||
if (miss_add && enqueue_possible && (MRVQ_SIZE != 2)) begin
|
||||
size <= size + 1;
|
||||
valid_table[enqueue_index] <= 1;
|
||||
ready_table[enqueue_index] <= 0;
|
||||
pc_table[enqueue_index] <= miss_add_pc;
|
||||
@@ -139,6 +143,7 @@ module VX_cache_miss_resrv
|
||||
end
|
||||
|
||||
if (miss_resrv_pop && dequeue_possible) begin
|
||||
size <= size - 1;
|
||||
valid_table[dequeue_index] <= 0;
|
||||
ready_table[dequeue_index] <= 0;
|
||||
addr_table[dequeue_index] <= 0;
|
||||
|
||||
@@ -94,7 +94,7 @@ module VX_tag_data_access
|
||||
wire[`DBANK_LINE_SIZE_RNG][31:0] use_write_data;
|
||||
|
||||
|
||||
wire real_writefill = writefill_st1e && miss_st1e;
|
||||
wire real_writefill = writefill_st1e && ((valid_req_st1e && !use_read_valid_st1e) || (valid_req_st1e && use_read_valid_st1e && (writeaddr_st1e[`TAG_SELECT_ADDR_RNG] != use_read_tag_st1e)));
|
||||
|
||||
|
||||
wire fill_sent;
|
||||
@@ -238,14 +238,14 @@ module VX_tag_data_access
|
||||
wire[3:0] sh_mask = (b0 ? 4'b0011 : 4'b1100);
|
||||
|
||||
wire should_write = (sw || sb || sh) && valid_req_st1e && use_read_valid_st1e && !miss_st1e;
|
||||
wire force_write = writefill_st1e && valid_req_st1e && miss_st1e && (!use_read_valid_st1e || (use_read_valid_st1e && !miss_st1e));
|
||||
wire force_write = real_writefill && valid_req_st1e && miss_st1e && (!use_read_valid_st1e || (use_read_valid_st1e && !miss_st1e));
|
||||
|
||||
wire[`DBANK_LINE_SIZE_RNG][3:0] we;
|
||||
wire[`DBANK_LINE_SIZE_RNG][31:0] data_write;
|
||||
genvar g;
|
||||
generate
|
||||
for (g = 0; g < `DBANK_LINE_SIZE_WORDS; g = g + 1) begin : write_enables
|
||||
wire normal_write = (block_offset == g[`WORD_SELECT_SIZE_RNG]) && should_write && !writefill_st1e;
|
||||
wire normal_write = (block_offset == g[`WORD_SELECT_SIZE_RNG]) && should_write && !real_writefill;
|
||||
|
||||
assign we[g] = (force_write) ? 4'b1111 :
|
||||
(normal_write && (FUNC_ID == `LLFUNC_ID)) ? 4'b1111 :
|
||||
|
||||
@@ -12,7 +12,7 @@ int main(int argc, char **argv)
|
||||
Verilated::traceEverOn(true);
|
||||
|
||||
|
||||
#define ALL_TESTS
|
||||
// #define ALL_TESTS
|
||||
#ifdef ALL_TESTS
|
||||
bool passed = true;
|
||||
|
||||
@@ -88,9 +88,9 @@ int main(int argc, char **argv)
|
||||
|
||||
#else
|
||||
|
||||
// char testing[] = "../../runtime/mains/simple/vx_simple_main.hex";
|
||||
char testing[] = "../../runtime/mains/simple/vx_simple_main.hex";
|
||||
// char testing[] = "../../emulator/riscv_tests/rv32ui-p-lw.hex";
|
||||
char testing[] = "../../emulator/riscv_tests/rv32ui-p-sw.hex";
|
||||
// char testing[] = "../../emulator/riscv_tests/rv32ui-p-sw.hex";
|
||||
Vortex v;
|
||||
// const char *testing;
|
||||
|
||||
|
||||
@@ -1,5 +1,5 @@
|
||||
|
||||
COMP = /opt/riscv-new/drops/bin/riscv32-unknown-elf-gcc
|
||||
COMP = /opt/riscv-new/drops/bin/riscv32-unknown-elf-g++
|
||||
CC_FLAGS = -march=rv32im -mabi=ilp32 -O0 -Wl,-Bstatic,-T,../vortex_link.ld -ffreestanding -nostdlib
|
||||
|
||||
DMP = /opt/riscv-new/drops/bin/riscv32-unknown-elf-objdump
|
||||
|
||||
@@ -52,7 +52,7 @@ int main()
|
||||
// Main is called with all threads active of warp 0
|
||||
vx_tmc(1);
|
||||
|
||||
vx_print_str("Let's start...\n");
|
||||
vx_print_str("Let's start... (This might take a while)\n");
|
||||
unsigned what[36];
|
||||
for (int i = 0; i < 36; i++)
|
||||
{
|
||||
@@ -98,31 +98,31 @@ int main()
|
||||
|
||||
}
|
||||
|
||||
vx_print_str("vx_spawnWarps mat_add_kernel\n");
|
||||
// vx_print_str("vx_spawnWarps mat_add_kernel\n");
|
||||
|
||||
mat_add_args_t arguments;
|
||||
arguments.x = x;
|
||||
arguments.y = y;
|
||||
arguments.z = z;
|
||||
arguments.numColums = 4;
|
||||
arguments.numRows = 4;
|
||||
// mat_add_args_t arguments;
|
||||
// arguments.x = x;
|
||||
// arguments.y = y;
|
||||
// arguments.z = z;
|
||||
// arguments.numColums = 4;
|
||||
// arguments.numRows = 4;
|
||||
|
||||
|
||||
int numWarps = 4;
|
||||
int numThreads = 4;
|
||||
// int numWarps = 4;
|
||||
// int numThreads = 4;
|
||||
|
||||
vx_spawnWarps(numWarps, numThreads, mat_add_kernel, &arguments);
|
||||
// vx_spawnWarps(numWarps, numThreads, mat_add_kernel, &arguments);
|
||||
|
||||
for (int i = 0; i < arguments.numRows; i++)
|
||||
{
|
||||
for (int j = 0; j < arguments.numColums; j++)
|
||||
{
|
||||
unsigned index = (i * arguments.numColums) + j;
|
||||
vx_print_hex(z[index]);
|
||||
vx_print_str(" ");
|
||||
}
|
||||
vx_print_str("\n");
|
||||
}
|
||||
// for (int i = 0; i < arguments.numRows; i++)
|
||||
// {
|
||||
// for (int j = 0; j < arguments.numColums; j++)
|
||||
// {
|
||||
// unsigned index = (i * arguments.numColums) + j;
|
||||
// vx_print_hex(z[index]);
|
||||
// vx_print_str(" ");
|
||||
// }
|
||||
// vx_print_str("\n");
|
||||
// }
|
||||
|
||||
return 0;
|
||||
}
|
||||
File diff suppressed because it is too large
Load Diff
Binary file not shown.
File diff suppressed because it is too large
Load Diff
@@ -17,27 +17,34 @@
|
||||
.global _start
|
||||
.type _start, @function
|
||||
_start:
|
||||
# Initialize SP
|
||||
# la sp, __stack_top
|
||||
# li a0, 4
|
||||
# .word 0x0005006b # tmc 4
|
||||
# csrr a2, 0x20 # get tid
|
||||
# slli a2, a2, 2
|
||||
# la a3, 0x80000000
|
||||
# add a3, a3, a2
|
||||
# lw a4, 0(a3)
|
||||
la a1, vx_set_sp
|
||||
li a0, 4
|
||||
.word 0x00b5106b # wspawn a0(numWarps), a1(PC SPAWN)
|
||||
jal vx_set_sp
|
||||
##########################################
|
||||
# li a0, 1
|
||||
# .word 0x0005006b # tmc 1
|
||||
# Initialize global pointerp
|
||||
# call __cxx_global_var_init
|
||||
# Clear the bss segment
|
||||
# la a0, _edata
|
||||
# la a2, _end
|
||||
# sub a2, a2, a0
|
||||
# li a1, 0
|
||||
# call memset
|
||||
# la a0, __libc_fini_array # Register global termination functions
|
||||
# call atexit # to be called upon exit
|
||||
# call __libc_init_array # Run global initialization functions
|
||||
# li a0, 4
|
||||
# .word 0x0005006b # tmc 4
|
||||
# # Initialize global pointerp
|
||||
# call __cxx_global_var_init
|
||||
# # Clear the bss segment
|
||||
# la a0, _edata
|
||||
# la a2, _end
|
||||
# sub a2, a2, a0
|
||||
# li a1, 0
|
||||
# call memset
|
||||
# la a0, __libc_fini_array # Register global termination functions
|
||||
# call atexit # to be called upon exit
|
||||
# call __libc_init_array # Run global initialization functions
|
||||
li a0, 4
|
||||
.word 0x0005006b # tmc 4
|
||||
##############################################
|
||||
call main
|
||||
tail exit
|
||||
.size _start, .-_start
|
||||
|
||||
Reference in New Issue
Block a user