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

cache refactoring (fixed redundant fill requests, merged fill and writeback queues), optimized priority encoder, fixed crs cycles count

Browse Source
This commit is contained in:
Blaise Tine
2020-11-02 01:50:12 -08:00
parent 3fe31fc337
commit 5be1d85648
39 changed files with 1145 additions and 1322 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
driver/opae/vlsim/Makefile
View File

@@ -9,6 +9,7 @@ DBG_PRINT_FLAGS += -DDBG_PRINT_CORE_DCACHE
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_BANK
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_SNP
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_MSRQ
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_DATA
DBG_PRINT_FLAGS += -DDBG_PRINT_DRAM
DBG_PRINT_FLAGS += -DDBG_PRINT_PIPELINE
DBG_PRINT_FLAGS += -DDBG_PRINT_OPAE
@@ -58,7 +59,7 @@ VL_FLAGS += verilator.vlt
# Debugigng
ifdef DEBUG
VL_FLAGS += -DVCD_OUTPUT --assert --trace-fst --trace-threads 1 $(DBG_FLAGS)
VL_FLAGS += -DVCD_OUTPUT --assert --trace --trace-structs --trace-threads 1 $(DBG_FLAGS)
CFLAGS += -DVCD_OUTPUT $(DBG_FLAGS)
else
VL_FLAGS += -DNDEBUG

4
driver/opae/vlsim/opae_sim.cpp
View File

@@ -31,9 +31,9 @@ opae_sim::opae_sim() {
#ifdef VCD_OUTPUT
Verilated::traceEverOn(true);
trace_ = new VerilatedFstC();
trace_ = new VerilatedVcdC();
vortex_afu_->trace(trace_, 99);
trace_->open("trace.fst");
trace_->open("trace.vcd");
#endif
this->reset();

4
driver/opae/vlsim/opae_sim.h
View File

@@ -5,7 +5,7 @@
#include "verilated.h"
#ifdef VCD_OUTPUT
#include <verilated_fst_c.h>
#include <verilated_vcd_c.h>
#endif
#include <VX_config.h>
@@ -88,6 +88,6 @@ private:
RAM ram_;
Vvortex_afu_shim *vortex_afu_;
#ifdef VCD_OUTPUT
VerilatedFstC *trace_;
VerilatedVcdC *trace_;
#endif
};

3
driver/rtlsim/Makefile
View File

@@ -9,6 +9,7 @@ DBG_PRINT_FLAGS += -DDBG_PRINT_CORE_DCACHE
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_BANK
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_SNP
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_MSRQ
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_DATA
DBG_PRINT_FLAGS += -DDBG_PRINT_DRAM
DBG_PRINT_FLAGS += -DDBG_PRINT_PIPELINE
DBG_PRINT_FLAGS += -DDBG_PRINT_OPAE
@@ -55,7 +56,7 @@ VL_FLAGS += verilator.vlt
# Debugigng
ifdef DEBUG
VL_FLAGS += -DVCD_OUTPUT --assert --trace-fst --trace-threads 1 $(DBG_FLAGS)
VL_FLAGS += -DVCD_OUTPUT --assert --trace --trace-structs --trace-threads 1 $(DBG_FLAGS)
CFLAGS += -DVCD_OUTPUT $(DBG_FLAGS)
else
VL_FLAGS += -DNDEBUG

2
driver/tests/dogfood/dogfood.cpp
View File

@@ -177,7 +177,7 @@ int main(int argc, char *argv[]) {
size_t buf_size = num_points * sizeof(uint32_t);
std::cout << "number of points: " << num_points << std::endl;
std::cout << "buffer size: " << std::hex << buf_size << std::dec << " bytes" << std::endl;
std::cout << "buffer size: " << buf_size << " bytes" << std::endl;
// upload program
std::cout << "upload kernel" << std::endl;

5
hw/opae/README
View File

@@ -83,6 +83,8 @@ tar -zcvf vortex.vcd.tar.gz build_ase_1c/work/vortex.vcd
tar -zcvf run.log.tar.gz build_ase_1c/work/run.log
tar -zcvf vx_scope.vcd.tar.gz vx_scope.vcd
tar -cvjf vx_scope.vcd.tar.bz2 vx_scope.vcd
tar -cvjf trace.fst.tar.bz2 trace.fst run.log
tar -cvjf trace.vcd.tar.bz2 trace.vcd run.log
# decompress VCD trace
tar -zxvf /mnt/c/Users/Blaise/Downloads/vortex.vcd.tar.gz
@@ -114,3 +116,6 @@ make -C ../../rtlsim clean && reset && make -C ../../rtlsim
# split tar into multiple parts
split -b 50M home.tar.bz2 "home.tar.bz2.part"
cat run.log | grep -c "cache[0-9]*:[0-9]* dram_req"

1
hw/opae/sources_1c.txt
View File

@@ -10,6 +10,7 @@
#+define+DBG_PRINT_CACHE_BANK
#+define+DBG_PRINT_CACHE_SNP
#+define+DBG_PRINT_CACHE_MSRQ
#+define+DBG_PRINT_CACHE_DATA
#+define+DBG_PRINT_DRAM
#+define+DBG_PRINT_PIPELINE
#+define+DBG_PRINT_OPAE

2
hw/opae/vortex_afu.sv
View File

@@ -1105,7 +1105,7 @@ wire scope_changed = `SCOPE_TRIGGER;
VX_scope #(
.DATAW ($bits({`SCOPE_DATA_LIST,`SCOPE_UPDATE_LIST})),
.BUSW (64),
.SIZE (4096),
.SIZE (`SCOPE_SIZE),
.UPDW ($bits({`SCOPE_UPDATE_LIST}))
) scope (
.clk (clk),

389
hw/rtl/VX_cluster.v
View File

@@ -211,34 +211,34 @@ module VX_cluster #(
.reset (reset),
// input requests
.in_io_req_valid (per_core_io_req_valid),
.in_io_req_rw (per_core_io_req_rw),
.in_io_req_byteen (per_core_io_req_byteen),
.in_io_req_addr (per_core_io_req_addr),
.in_io_req_data (per_core_io_req_data),
.in_io_req_tag (per_core_io_req_tag),
.in_io_req_ready (per_core_io_req_ready),
.io_req_valid_in (per_core_io_req_valid),
.io_req_rw_in (per_core_io_req_rw),
.io_req_byteen_in (per_core_io_req_byteen),
.io_req_addr_in (per_core_io_req_addr),
.io_req_data_in (per_core_io_req_data),
.io_req_tag_in (per_core_io_req_tag),
.io_req_ready_in (per_core_io_req_ready),
// input responses
.in_io_rsp_valid (per_core_io_rsp_valid),
.in_io_rsp_data (per_core_io_rsp_data),
.in_io_rsp_tag (per_core_io_rsp_tag),
.in_io_rsp_ready (per_core_io_rsp_ready),
.io_rsp_valid_in (per_core_io_rsp_valid),
.io_rsp_data_in (per_core_io_rsp_data),
.io_rsp_tag_in (per_core_io_rsp_tag),
.io_rsp_ready_in (per_core_io_rsp_ready),
// output request
.out_io_req_valid (io_req_valid),
.out_io_req_rw (io_req_rw),
.out_io_req_byteen (io_req_byteen),
.out_io_req_addr (io_req_addr),
.out_io_req_data (io_req_data),
.out_io_req_tag (io_req_tag),
.out_io_req_ready (io_req_ready),
.io_req_valid_out (io_req_valid),
.io_req_rw_out (io_req_rw),
.io_req_byteen_out (io_req_byteen),
.io_req_addr_out (io_req_addr),
.io_req_data_out (io_req_data),
.io_req_tag_out (io_req_tag),
.io_req_ready_out (io_req_ready),
// output response
.out_io_rsp_valid (io_rsp_valid),
.out_io_rsp_tag (io_rsp_tag),
.out_io_rsp_data (io_rsp_data),
.out_io_rsp_ready (io_rsp_ready)
.io_rsp_valid_out (io_rsp_valid),
.io_rsp_tag_out (io_rsp_tag),
.io_rsp_data_out (io_rsp_data),
.io_rsp_ready_out (io_rsp_ready)
);
VX_csr_io_arb #(
@@ -250,28 +250,28 @@ module VX_cluster #(
.request_id (csr_io_req_coreid),
// input requests
.in_csr_io_req_valid (csr_io_req_valid),
.in_csr_io_req_addr (csr_io_req_addr),
.in_csr_io_req_rw (csr_io_req_rw),
.in_csr_io_req_data (csr_io_req_data),
.in_csr_io_req_ready (csr_io_req_ready),
.csr_io_req_valid_in (csr_io_req_valid),
.csr_io_req_addr_in (csr_io_req_addr),
.csr_io_req_rw_in (csr_io_req_rw),
.csr_io_req_data_in (csr_io_req_data),
.csr_io_req_ready_in (csr_io_req_ready),
// input responses
.in_csr_io_rsp_valid (per_core_csr_io_rsp_valid),
.in_csr_io_rsp_data (per_core_csr_io_rsp_data),
.in_csr_io_rsp_ready (per_core_csr_io_rsp_ready),
.csr_io_rsp_valid_in (per_core_csr_io_rsp_valid),
.csr_io_rsp_data_in (per_core_csr_io_rsp_data),
.csr_io_rsp_ready_in (per_core_csr_io_rsp_ready),
// output request
.out_csr_io_req_valid (per_core_csr_io_req_valid),
.out_csr_io_req_addr (per_core_csr_io_req_addr),
.out_csr_io_req_rw (per_core_csr_io_req_rw),
.out_csr_io_req_data (per_core_csr_io_req_data),
.out_csr_io_req_ready (per_core_csr_io_req_ready),
.csr_io_req_valid_out (per_core_csr_io_req_valid),
.csr_io_req_addr_out (per_core_csr_io_req_addr),
.csr_io_req_rw_out (per_core_csr_io_req_rw),
.csr_io_req_data_out (per_core_csr_io_req_data),
.csr_io_req_ready_out (per_core_csr_io_req_ready),
// output response
.out_csr_io_rsp_valid (csr_io_rsp_valid),
.out_csr_io_rsp_data (csr_io_rsp_data),
.out_csr_io_rsp_ready (csr_io_rsp_ready)
.csr_io_rsp_valid_out (csr_io_rsp_valid),
.csr_io_rsp_data_out (csr_io_rsp_data),
.csr_io_rsp_ready_out (csr_io_rsp_ready)
);
assign busy = (| per_core_busy);
@@ -281,72 +281,72 @@ module VX_cluster #(
// L2 Cache ///////////////////////////////////////////////////////////
wire[`L2NUM_REQUESTS-1:0] l2_core_req_valid;
wire[`L2NUM_REQUESTS-1:0] l2_core_req_rw;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_BYTEEN_WIDTH-1:0] l2_core_req_byteen;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_ADDR_WIDTH-1:0] l2_core_req_addr;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] l2_core_req_tag;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] l2_core_req_data;
wire l2_core_req_ready;
wire[`L2NUM_REQUESTS-1:0] core_dram_req_valid;
wire[`L2NUM_REQUESTS-1:0] core_dram_req_rw;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_BYTEEN_WIDTH-1:0] core_dram_req_byteen;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_ADDR_WIDTH-1:0] core_dram_req_addr;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] core_dram_req_tag;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] core_dram_req_data;
wire core_dram_req_ready;
wire[`L2NUM_REQUESTS-1:0] l2_core_rsp_valid;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] l2_core_rsp_data;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] l2_core_rsp_tag;
wire l2_core_rsp_ready;
wire[`L2NUM_REQUESTS-1:0] core_dram_rsp_valid;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] core_dram_rsp_data;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] core_dram_rsp_tag;
wire core_dram_rsp_ready;
wire[`NUM_CORES-1:0] l2_snp_fwdout_valid;
wire[`NUM_CORES-1:0][`DDRAM_ADDR_WIDTH-1:0] l2_snp_fwdout_addr;
wire[`NUM_CORES-1:0] l2_snp_fwdout_invalidate;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] l2_snp_fwdout_tag;
wire[`NUM_CORES-1:0] l2_snp_fwdout_ready;
wire[`NUM_CORES-1:0] core_snp_fwdout_valid;
wire[`NUM_CORES-1:0][`DDRAM_ADDR_WIDTH-1:0] core_snp_fwdout_addr;
wire[`NUM_CORES-1:0] core_snp_fwdout_invalidate;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] core_snp_fwdout_tag;
wire[`NUM_CORES-1:0] core_snp_fwdout_ready;
wire[`NUM_CORES-1:0] l2_snp_fwdin_valid;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] l2_snp_fwdin_tag;
wire[`NUM_CORES-1:0] l2_snp_fwdin_ready;
wire[`NUM_CORES-1:0] core_snp_fwdin_valid;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] core_snp_fwdin_tag;
wire[`NUM_CORES-1:0] core_snp_fwdin_ready;
for (genvar i = 0; i < `L2NUM_REQUESTS; i = i + 2) begin
assign l2_core_req_valid [i] = per_core_D_dram_req_valid[(i/2)];
assign l2_core_req_valid [i+1] = per_core_I_dram_req_valid[(i/2)];
assign core_dram_req_valid [i] = per_core_D_dram_req_valid[(i/2)];
assign core_dram_req_valid [i+1] = per_core_I_dram_req_valid[(i/2)];
assign l2_core_req_rw [i] = per_core_D_dram_req_rw[(i/2)];
assign l2_core_req_rw [i+1] = per_core_I_dram_req_rw[(i/2)];
assign core_dram_req_rw [i] = per_core_D_dram_req_rw[(i/2)];
assign core_dram_req_rw [i+1] = per_core_I_dram_req_rw[(i/2)];
assign l2_core_req_byteen [i] = per_core_D_dram_req_byteen[(i/2)];
assign l2_core_req_byteen [i+1] = per_core_I_dram_req_byteen[(i/2)];
assign core_dram_req_byteen [i] = per_core_D_dram_req_byteen[(i/2)];
assign core_dram_req_byteen [i+1] = per_core_I_dram_req_byteen[(i/2)];
assign l2_core_req_addr [i] = per_core_D_dram_req_addr[(i/2)];
assign l2_core_req_addr [i+1] = per_core_I_dram_req_addr[(i/2)];
assign core_dram_req_addr [i] = per_core_D_dram_req_addr[(i/2)];
assign core_dram_req_addr [i+1] = per_core_I_dram_req_addr[(i/2)];
assign l2_core_req_data [i] = per_core_D_dram_req_data[(i/2)];
assign l2_core_req_data [i+1] = per_core_I_dram_req_data[(i/2)];
assign core_dram_req_data [i] = per_core_D_dram_req_data[(i/2)];
assign core_dram_req_data [i+1] = per_core_I_dram_req_data[(i/2)];
assign l2_core_req_tag [i] = per_core_D_dram_req_tag[(i/2)];
assign l2_core_req_tag [i+1] = per_core_I_dram_req_tag[(i/2)];
assign core_dram_req_tag [i] = per_core_D_dram_req_tag[(i/2)];
assign core_dram_req_tag [i+1] = per_core_I_dram_req_tag[(i/2)];
assign per_core_D_dram_req_ready [(i/2)] = l2_core_req_ready;
assign per_core_I_dram_req_ready [(i/2)] = l2_core_req_ready;
assign per_core_D_dram_req_ready [(i/2)] = core_dram_req_ready;
assign per_core_I_dram_req_ready [(i/2)] = core_dram_req_ready;
assign per_core_D_dram_rsp_valid [(i/2)] = l2_core_rsp_valid[i] && l2_core_rsp_ready;
assign per_core_I_dram_rsp_valid [(i/2)] = l2_core_rsp_valid[i+1] && l2_core_rsp_ready;
assign per_core_D_dram_rsp_valid [(i/2)] = core_dram_rsp_valid[i] && core_dram_rsp_ready;
assign per_core_I_dram_rsp_valid [(i/2)] = core_dram_rsp_valid[i+1] && core_dram_rsp_ready;
assign per_core_D_dram_rsp_data [(i/2)] = l2_core_rsp_data[i];
assign per_core_I_dram_rsp_data [(i/2)] = l2_core_rsp_data[i+1];
assign per_core_D_dram_rsp_data [(i/2)] = core_dram_rsp_data[i];
assign per_core_I_dram_rsp_data [(i/2)] = core_dram_rsp_data[i+1];
assign per_core_D_dram_rsp_tag [(i/2)] = l2_core_rsp_tag[i];
assign per_core_I_dram_rsp_tag [(i/2)] = l2_core_rsp_tag[i+1];
assign per_core_D_dram_rsp_tag [(i/2)] = core_dram_rsp_tag[i];
assign per_core_I_dram_rsp_tag [(i/2)] = core_dram_rsp_tag[i+1];
assign per_core_snp_req_valid [(i/2)] = l2_snp_fwdout_valid [(i/2)];
assign per_core_snp_req_addr [(i/2)] = l2_snp_fwdout_addr [(i/2)];
assign per_core_snp_req_invalidate [(i/2)] = l2_snp_fwdout_invalidate [(i/2)];
assign per_core_snp_req_tag [(i/2)] = l2_snp_fwdout_tag [(i/2)];
assign l2_snp_fwdout_ready [(i/2)] = per_core_snp_req_ready[(i/2)];
assign per_core_snp_req_valid [(i/2)] = core_snp_fwdout_valid [(i/2)];
assign per_core_snp_req_addr [(i/2)] = core_snp_fwdout_addr [(i/2)];
assign per_core_snp_req_invalidate [(i/2)] = core_snp_fwdout_invalidate [(i/2)];
assign per_core_snp_req_tag [(i/2)] = core_snp_fwdout_tag [(i/2)];
assign core_snp_fwdout_ready [(i/2)] = per_core_snp_req_ready[(i/2)];
assign l2_snp_fwdin_valid [(i/2)] = per_core_snp_rsp_valid [(i/2)];
assign l2_snp_fwdin_tag [(i/2)] = per_core_snp_rsp_tag [(i/2)];
assign per_core_snp_rsp_ready [(i/2)] = l2_snp_fwdin_ready [(i/2)];
assign core_snp_fwdin_valid [(i/2)] = per_core_snp_rsp_valid [(i/2)];
assign core_snp_fwdin_tag [(i/2)] = per_core_snp_rsp_tag [(i/2)];
assign per_core_snp_rsp_ready [(i/2)] = core_snp_fwdin_ready [(i/2)];
end
assign l2_core_rsp_ready = (& per_core_D_dram_rsp_ready) && (& per_core_I_dram_rsp_ready);
assign core_dram_rsp_ready = (& per_core_D_dram_rsp_ready) && (& per_core_I_dram_rsp_ready);
VX_cache #(
.CACHE_ID (`L2CACHE_ID),
@@ -357,11 +357,10 @@ module VX_cluster #(
.NUM_REQUESTS (`L2NUM_REQUESTS),
.CREQ_SIZE (`L2CREQ_SIZE),
.MRVQ_SIZE (`L2MRVQ_SIZE),
.DFPQ_SIZE (`L2DFPQ_SIZE),
.DRPQ_SIZE (`L2DRPQ_SIZE),
.SNRQ_SIZE (`L2SNRQ_SIZE),
.CWBQ_SIZE (`L2CWBQ_SIZE),
.DWBQ_SIZE (`L2DWBQ_SIZE),
.DFQQ_SIZE (`L2DFQQ_SIZE),
.DREQ_SIZE (`L2DREQ_SIZE),
.DRAM_ENABLE (1),
.WRITE_ENABLE (1),
.SNOOP_FORWARDING (1),
@@ -378,19 +377,19 @@ module VX_cluster #(
.reset (reset),
// Core request
.core_req_valid (l2_core_req_valid),
.core_req_rw (l2_core_req_rw),
.core_req_byteen (l2_core_req_byteen),
.core_req_addr (l2_core_req_addr),
.core_req_data (l2_core_req_data),
.core_req_tag (l2_core_req_tag),
.core_req_ready (l2_core_req_ready),
.core_req_valid (core_dram_req_valid),
.core_req_rw (core_dram_req_rw),
.core_req_byteen (core_dram_req_byteen),
.core_req_addr (core_dram_req_addr),
.core_req_data (core_dram_req_data),
.core_req_tag (core_dram_req_tag),
.core_req_ready (core_dram_req_ready),
// Core response
.core_rsp_valid (l2_core_rsp_valid),
.core_rsp_data (l2_core_rsp_data),
.core_rsp_tag (l2_core_rsp_tag),
.core_rsp_ready (l2_core_rsp_ready),
.core_rsp_valid (core_dram_rsp_valid),
.core_rsp_data (core_dram_rsp_data),
.core_rsp_tag (core_dram_rsp_tag),
.core_rsp_ready (core_dram_rsp_ready),
// DRAM request
.dram_req_valid (dram_req_valid),
@@ -420,86 +419,86 @@ module VX_cluster #(
.snp_rsp_ready (snp_rsp_ready),
// Snoop forwarding out
.snp_fwdout_valid (l2_snp_fwdout_valid),
.snp_fwdout_addr (l2_snp_fwdout_addr),
.snp_fwdout_invalidate(l2_snp_fwdout_invalidate),
.snp_fwdout_tag (l2_snp_fwdout_tag),
.snp_fwdout_ready (l2_snp_fwdout_ready),
.snp_fwdout_valid (core_snp_fwdout_valid),
.snp_fwdout_addr (core_snp_fwdout_addr),
.snp_fwdout_invalidate(core_snp_fwdout_invalidate),
.snp_fwdout_tag (core_snp_fwdout_tag),
.snp_fwdout_ready (core_snp_fwdout_ready),
// Snoop forwarding in
.snp_fwdin_valid (l2_snp_fwdin_valid),
.snp_fwdin_tag (l2_snp_fwdin_tag),
.snp_fwdin_ready (l2_snp_fwdin_ready)
.snp_fwdin_valid (core_snp_fwdin_valid),
.snp_fwdin_tag (core_snp_fwdin_tag),
.snp_fwdin_ready (core_snp_fwdin_ready)
);
end else begin
wire[`L2NUM_REQUESTS-1:0] arb_dram_req_valid;
wire[`L2NUM_REQUESTS-1:0] arb_dram_req_rw;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_BYTEEN_WIDTH-1:0] arb_dram_req_byteen;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_ADDR_WIDTH-1:0] arb_dram_req_addr;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] arb_dram_req_tag;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] arb_dram_req_data;
wire[`L2NUM_REQUESTS-1:0] arb_dram_req_ready;
wire[`L2NUM_REQUESTS-1:0] core_dram_req_valid;
wire[`L2NUM_REQUESTS-1:0] core_dram_req_rw;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_BYTEEN_WIDTH-1:0] core_dram_req_byteen;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_ADDR_WIDTH-1:0] core_dram_req_addr;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] core_dram_req_tag;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] core_dram_req_data;
wire[`L2NUM_REQUESTS-1:0] core_dram_req_ready;
wire[`L2NUM_REQUESTS-1:0] arb_dram_rsp_valid;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] arb_dram_rsp_data;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] arb_dram_rsp_tag;
wire[`L2NUM_REQUESTS-1:0] arb_dram_rsp_ready;
wire[`L2NUM_REQUESTS-1:0] core_dram_rsp_valid;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_LINE_WIDTH-1:0] core_dram_rsp_data;
wire[`L2NUM_REQUESTS-1:0][`DDRAM_TAG_WIDTH-1:0] core_dram_rsp_tag;
wire[`L2NUM_REQUESTS-1:0] core_dram_rsp_ready;
wire[`NUM_CORES-1:0] arb_snp_fwdout_valid;
wire[`NUM_CORES-1:0][`DDRAM_ADDR_WIDTH-1:0] arb_snp_fwdout_addr;
wire[`NUM_CORES-1:0] arb_snp_fwdout_invalidate;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] arb_snp_fwdout_tag;
wire[`NUM_CORES-1:0] arb_snp_fwdout_ready;
wire[`NUM_CORES-1:0] core_snp_fwdout_valid;
wire[`NUM_CORES-1:0][`DDRAM_ADDR_WIDTH-1:0] core_snp_fwdout_addr;
wire[`NUM_CORES-1:0] core_snp_fwdout_invalidate;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] core_snp_fwdout_tag;
wire[`NUM_CORES-1:0] core_snp_fwdout_ready;
wire[`NUM_CORES-1:0] arb_snp_fwdin_valid;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] arb_snp_fwdin_tag;
wire[`NUM_CORES-1:0] arb_snp_fwdin_ready;
wire[`NUM_CORES-1:0] core_snp_fwdin_valid;
wire[`NUM_CORES-1:0][`DSNP_TAG_WIDTH-1:0] core_snp_fwdin_tag;
wire[`NUM_CORES-1:0] core_snp_fwdin_ready;
for (genvar i = 0; i < `L2NUM_REQUESTS; i = i + 2) begin
assign arb_dram_req_valid [i] = per_core_D_dram_req_valid[(i/2)];
assign arb_dram_req_valid [i+1] = per_core_I_dram_req_valid[(i/2)];
assign core_dram_req_valid [i] = per_core_D_dram_req_valid[(i/2)];
assign core_dram_req_valid [i+1] = per_core_I_dram_req_valid[(i/2)];
assign arb_dram_req_rw [i] = per_core_D_dram_req_rw[(i/2)];
assign arb_dram_req_rw [i+1] = per_core_I_dram_req_rw[(i/2)];
assign core_dram_req_rw [i] = per_core_D_dram_req_rw[(i/2)];
assign core_dram_req_rw [i+1] = per_core_I_dram_req_rw[(i/2)];
assign arb_dram_req_byteen[i] = per_core_D_dram_req_byteen[(i/2)];
assign arb_dram_req_byteen[i+1] = per_core_I_dram_req_byteen[(i/2)];
assign core_dram_req_byteen[i] = per_core_D_dram_req_byteen[(i/2)];
assign core_dram_req_byteen[i+1] = per_core_I_dram_req_byteen[(i/2)];
assign arb_dram_req_addr [i] = per_core_D_dram_req_addr[(i/2)];
assign arb_dram_req_addr [i+1] = per_core_I_dram_req_addr[(i/2)];
assign core_dram_req_addr [i] = per_core_D_dram_req_addr[(i/2)];
assign core_dram_req_addr [i+1] = per_core_I_dram_req_addr[(i/2)];
assign arb_dram_req_data [i] = per_core_D_dram_req_data[(i/2)];
assign arb_dram_req_data [i+1] = per_core_I_dram_req_data[(i/2)];
assign core_dram_req_data [i] = per_core_D_dram_req_data[(i/2)];
assign core_dram_req_data [i+1] = per_core_I_dram_req_data[(i/2)];
assign arb_dram_req_tag [i] = per_core_D_dram_req_tag[(i/2)];
assign arb_dram_req_tag [i+1] = per_core_I_dram_req_tag[(i/2)];
assign core_dram_req_tag [i] = per_core_D_dram_req_tag[(i/2)];
assign core_dram_req_tag [i+1] = per_core_I_dram_req_tag[(i/2)];
assign per_core_D_dram_req_ready [(i/2)] = arb_dram_req_ready[i];
assign per_core_I_dram_req_ready [(i/2)] = arb_dram_req_ready[i+1];
assign per_core_D_dram_req_ready [(i/2)] = core_dram_req_ready[i];
assign per_core_I_dram_req_ready [(i/2)] = core_dram_req_ready[i+1];
assign per_core_D_dram_rsp_valid [(i/2)] = arb_dram_rsp_valid[i];
assign per_core_I_dram_rsp_valid [(i/2)] = arb_dram_rsp_valid[i+1];
assign per_core_D_dram_rsp_valid [(i/2)] = core_dram_rsp_valid[i];
assign per_core_I_dram_rsp_valid [(i/2)] = core_dram_rsp_valid[i+1];
assign per_core_D_dram_rsp_data [(i/2)] = arb_dram_rsp_data[i];
assign per_core_I_dram_rsp_data [(i/2)] = arb_dram_rsp_data[i+1];
assign per_core_D_dram_rsp_data [(i/2)] = core_dram_rsp_data[i];
assign per_core_I_dram_rsp_data [(i/2)] = core_dram_rsp_data[i+1];
assign per_core_D_dram_rsp_tag [(i/2)] = arb_dram_rsp_tag[i];
assign per_core_I_dram_rsp_tag [(i/2)] = arb_dram_rsp_tag[i+1];
assign per_core_D_dram_rsp_tag [(i/2)] = core_dram_rsp_tag[i];
assign per_core_I_dram_rsp_tag [(i/2)] = core_dram_rsp_tag[i+1];
assign arb_dram_rsp_ready [i] = per_core_D_dram_rsp_ready[(i/2)];
assign arb_dram_rsp_ready [i+1] = per_core_I_dram_rsp_ready[(i/2)];
assign core_dram_rsp_ready [i] = per_core_D_dram_rsp_ready[(i/2)];
assign core_dram_rsp_ready [i+1] = per_core_I_dram_rsp_ready[(i/2)];
assign per_core_snp_req_valid [(i/2)] = arb_snp_fwdout_valid [(i/2)];
assign per_core_snp_req_addr [(i/2)] = arb_snp_fwdout_addr [(i/2)];
assign per_core_snp_req_invalidate [(i/2)] = arb_snp_fwdout_invalidate [(i/2)];
assign per_core_snp_req_tag [(i/2)] = arb_snp_fwdout_tag [(i/2)];
assign arb_snp_fwdout_ready [(i/2)] = per_core_snp_req_ready[(i/2)];
assign per_core_snp_req_valid [(i/2)] = core_snp_fwdout_valid [(i/2)];
assign per_core_snp_req_addr [(i/2)] = core_snp_fwdout_addr [(i/2)];
assign per_core_snp_req_invalidate [(i/2)] = core_snp_fwdout_invalidate [(i/2)];
assign per_core_snp_req_tag [(i/2)] = core_snp_fwdout_tag [(i/2)];
assign core_snp_fwdout_ready [(i/2)] = per_core_snp_req_ready[(i/2)];
assign arb_snp_fwdin_valid [(i/2)] = per_core_snp_rsp_valid [(i/2)];
assign arb_snp_fwdin_tag [(i/2)] = per_core_snp_rsp_tag [(i/2)];
assign per_core_snp_rsp_ready [(i/2)] = arb_snp_fwdin_ready [(i/2)];
assign core_snp_fwdin_valid [(i/2)] = per_core_snp_rsp_valid [(i/2)];
assign core_snp_fwdin_tag [(i/2)] = per_core_snp_rsp_tag [(i/2)];
assign per_core_snp_rsp_ready [(i/2)] = core_snp_fwdin_ready [(i/2)];
end
if (`NUM_CORES > 1) begin
@@ -525,26 +524,26 @@ module VX_cluster #(
.snp_rsp_tag (snp_rsp_tag),
.snp_rsp_ready (snp_rsp_ready),
.snp_fwdout_valid (arb_snp_fwdout_valid),
.snp_fwdout_addr (arb_snp_fwdout_addr),
.snp_fwdout_invalidate(arb_snp_fwdout_invalidate),
.snp_fwdout_tag (arb_snp_fwdout_tag),
.snp_fwdout_ready (arb_snp_fwdout_ready),
.snp_fwdout_valid (core_snp_fwdout_valid),
.snp_fwdout_addr (core_snp_fwdout_addr),
.snp_fwdout_invalidate(core_snp_fwdout_invalidate),
.snp_fwdout_tag (core_snp_fwdout_tag),
.snp_fwdout_ready (core_snp_fwdout_ready),
.snp_fwdin_valid (arb_snp_fwdin_valid),
.snp_fwdin_tag (arb_snp_fwdin_tag),
.snp_fwdin_ready (arb_snp_fwdin_ready)
.snp_fwdin_valid (core_snp_fwdin_valid),
.snp_fwdin_tag (core_snp_fwdin_tag),
.snp_fwdin_ready (core_snp_fwdin_ready)
);
end else begin
assign arb_snp_fwdout_valid = snp_req_valid;
assign arb_snp_fwdout_addr = snp_req_addr;
assign arb_snp_fwdout_invalidate = snp_req_invalidate;
assign arb_snp_fwdout_tag = snp_req_tag;
assign snp_req_ready = arb_snp_fwdout_ready;
assign core_snp_fwdout_valid = snp_req_valid;
assign core_snp_fwdout_addr = snp_req_addr;
assign core_snp_fwdout_invalidate = snp_req_invalidate;
assign core_snp_fwdout_tag = snp_req_tag;
assign snp_req_ready = core_snp_fwdout_ready;
assign snp_rsp_valid = arb_snp_fwdin_valid;
assign snp_rsp_tag = arb_snp_fwdin_tag;
assign arb_snp_fwdin_ready = snp_rsp_ready;
assign snp_rsp_valid = core_snp_fwdin_valid;
assign snp_rsp_tag = core_snp_fwdin_tag;
assign core_snp_fwdin_ready = snp_rsp_ready;
end
VX_mem_arb #(
@@ -557,34 +556,34 @@ module VX_cluster #(
.reset (reset),
// Core request
.in_mem_req_valid (arb_dram_req_valid),
.in_mem_req_rw (arb_dram_req_rw),
.in_mem_req_byteen (arb_dram_req_byteen),
.in_mem_req_addr (arb_dram_req_addr),
.in_mem_req_data (arb_dram_req_data),
.in_mem_req_tag (arb_dram_req_tag),
.in_mem_req_ready (arb_dram_req_ready),
.mem_req_valid_in (core_dram_req_valid),
.mem_req_rw_in (core_dram_req_rw),
.mem_req_byteen_in (core_dram_req_byteen),
.mem_req_addr_in (core_dram_req_addr),
.mem_req_data_in (core_dram_req_data),
.mem_req_tag_in (core_dram_req_tag),
.mem_req_ready_in (core_dram_req_ready),
// Core response
.in_mem_rsp_valid (arb_dram_rsp_valid),
.in_mem_rsp_data (arb_dram_rsp_data),
.in_mem_rsp_tag (arb_dram_rsp_tag),
.in_mem_rsp_ready (arb_dram_rsp_ready),
.mem_rsp_valid_in (core_dram_rsp_valid),
.mem_rsp_data_in (core_dram_rsp_data),
.mem_rsp_tag_in (core_dram_rsp_tag),
.mem_rsp_ready_in (core_dram_rsp_ready),
// DRAM request
.out_mem_req_valid (dram_req_valid),
.out_mem_req_rw (dram_req_rw),
.out_mem_req_byteen (dram_req_byteen),
.out_mem_req_addr (dram_req_addr),
.out_mem_req_data (dram_req_data),
.out_mem_req_tag (dram_req_tag),
.out_mem_req_ready (dram_req_ready),
.mem_req_valid_out (dram_req_valid),
.mem_req_rw_out (dram_req_rw),
.mem_req_byteen_out (dram_req_byteen),
.mem_req_addr_out (dram_req_addr),
.mem_req_data_out (dram_req_data),
.mem_req_tag_out (dram_req_tag),
.mem_req_ready_out (dram_req_ready),
// DRAM response
.out_mem_rsp_valid (dram_rsp_valid),
.out_mem_rsp_tag (dram_rsp_tag),
.out_mem_rsp_data (dram_rsp_data),
.out_mem_rsp_ready (dram_rsp_ready)
.mem_rsp_valid_out (dram_rsp_valid),
.mem_rsp_tag_out (dram_rsp_tag),
.mem_rsp_data_out (dram_rsp_data),
.mem_rsp_ready_out (dram_rsp_ready)
);
end

102
hw/rtl/VX_config.vh
View File

@@ -213,29 +213,24 @@
`define DMRVQ_SIZE `MAX(`NUM_WARPS*`NUM_THREADS, 8)
`endif
// Dram Fill Rsp Queue Size
`ifndef DDFPQ_SIZE
`define DDFPQ_SIZE 8
`endif
// Snoop Req Queue Size
`ifndef DSNRQ_SIZE
`define DSNRQ_SIZE 8
`endif
// Core Writeback Queue Size
`ifndef DCWBQ_SIZE
`define DCWBQ_SIZE `DCREQ_SIZE
`endif
// Dram Writeback Queue Size
`ifndef DDWBQ_SIZE
`define DDWBQ_SIZE 4
// DRAM Request Queue Size
`ifndef DDREQ_SIZE
`define DDREQ_SIZE 8
`endif
// Dram Fill Req Queue Size
`ifndef DDFQQ_SIZE
`define DDFQQ_SIZE `DCREQ_SIZE
// DRAM Response Queue Size
`ifndef DDRPQ_SIZE
`define DDRPQ_SIZE 8
`endif
// Snoop Req Queue Size
`ifndef DSNRQ_SIZE
`define DSNRQ_SIZE 8
`endif
// Icache Configurable Knobs ==================================================
@@ -270,24 +265,19 @@
`define IMRVQ_SIZE `MAX(`ICREQ_SIZE, 8)
`endif
// Dram Fill Rsp Queue Size
`ifndef IDFPQ_SIZE
`define IDFPQ_SIZE 8
`endif
// Core Writeback Queue Size
`ifndef ICWBQ_SIZE
`define ICWBQ_SIZE `ICREQ_SIZE
`endif
// Dram Writeback Queue Size
`ifndef IDWBQ_SIZE
`define IDWBQ_SIZE 8
// DRAM Request Queue Size
`ifndef IDREQ_SIZE
`define IDREQ_SIZE 8
`endif
// Dram Fill Req Queue Size
`ifndef IDFQQ_SIZE
`define IDFQQ_SIZE `ICREQ_SIZE
// DRAM Response Queue Size
`ifndef IDRPQ_SIZE
`define IDRPQ_SIZE 8
`endif
// SM Configurable Knobs ======================================================
@@ -354,29 +344,24 @@
`define L2MRVQ_SIZE `MAX(`L2CREQ_SIZE, 8)
`endif
// Dram Fill Rsp Queue Size
`ifndef L2DFPQ_SIZE
`define L2DFPQ_SIZE 8
`endif
// Snoop Req Queue Size
`ifndef L2SNRQ_SIZE
`define L2SNRQ_SIZE 8
`endif
// Core Writeback Queue Size
`ifndef L2CWBQ_SIZE
`define L2CWBQ_SIZE `L2CREQ_SIZE
`endif
// Dram Writeback Queue Size
`ifndef L2DWBQ_SIZE
`define L2DWBQ_SIZE 8
// DRAM Request Queue Size
`ifndef L2DREQ_SIZE
`define L2DREQ_SIZE 8
`endif
// Dram Fill Req Queue Size
`ifndef L2DFQQ_SIZE
`define L2DFQQ_SIZE `L2CREQ_SIZE
// DRAM Response Queue Size
`ifndef L2DRPQ_SIZE
`define L2DRPQ_SIZE 8
`endif
// Snoop Req Queue Size
`ifndef L2SNRQ_SIZE
`define L2SNRQ_SIZE 8
`endif
// L3cache Configurable Knobs =================================================
@@ -411,9 +396,19 @@
`define L3MRVQ_SIZE `MAX(`L3CREQ_SIZE, 8)
`endif
// Dram Fill Rsp Queue Size
`ifndef L3DFPQ_SIZE
`define L3DFPQ_SIZE 8
// Core Writeback Queue Size
`ifndef L3CWBQ_SIZE
`define L3CWBQ_SIZE `L3CREQ_SIZE
`endif
// DRAM Request Queue Size
`ifndef L3DREQ_SIZE
`define L3DREQ_SIZE 8
`endif
// DRAM Response Queue Size
`ifndef L3DRPQ_SIZE
`define L3DRPQ_SIZE 8
`endif
// Snoop Req Queue Size
@@ -421,19 +416,4 @@
`define L3SNRQ_SIZE 8
`endif
// Core Writeback Queue Size
`ifndef L3CWBQ_SIZE
`define L3CWBQ_SIZE `L3CREQ_SIZE
`endif
// Dram Writeback Queue Size
`ifndef L3DWBQ_SIZE
`define L3DWBQ_SIZE 8
`endif
// Dram Fill Req Queue Size
`ifndef L3DFQQ_SIZE
`define L3DFQQ_SIZE `L3CREQ_SIZE
`endif
`endif

4
hw/rtl/VX_core.v
View File

@@ -260,7 +260,7 @@ module VX_core #(
.core_dcache_req_if (arb_dcache_req_if),
.core_dcache_rsp_if (arb_dcache_rsp_if),
// Dram <-> Dcache
// DRAM <-> Dcache
.dcache_dram_req_if (dcache_dram_req_if),
.dcache_dram_rsp_if (dcache_dram_rsp_if),
.dcache_snp_req_if (dcache_snp_req_if),
@@ -270,7 +270,7 @@ module VX_core #(
.core_icache_req_if (core_icache_req_if),
.core_icache_rsp_if (core_icache_rsp_if),
// Dram <-> Icache
// DRAM <-> Icache
.icache_dram_req_if (icache_dram_req_if),
.icache_dram_rsp_if (icache_dram_rsp_if)
);

6
hw/rtl/VX_csr_data.v
View File

@@ -17,7 +17,9 @@ module VX_csr_data #(
input wire write_enable,
input wire[`CSR_ADDR_BITS-1:0] write_addr,
input wire[`NW_BITS-1:0] write_wid,
input wire[`CSR_WIDTH-1:0] write_data
input wire[`CSR_WIDTH-1:0] write_data,
input wire busy
);
reg [`CSR_WIDTH-1:0] csr_satp;
reg [`CSR_WIDTH-1:0] csr_mstatus;
@@ -86,7 +88,9 @@ module VX_csr_data #(
csr_cycle <= 0;
csr_instret <= 0;
end else begin
if (busy) begin
csr_cycle <= csr_cycle + 1;
end
if (cmt_to_csr_if.valid) begin
csr_instret <= csr_instret + 64'(cmt_to_csr_if.num_commits);
end

66
hw/rtl/VX_csr_io_arb.v
View File

@@ -10,28 +10,28 @@ module VX_csr_io_arb #(
input wire [REQS_BITS-1:0] request_id,
// input requests
input wire in_csr_io_req_valid,
input wire [11:0] in_csr_io_req_addr,
input wire in_csr_io_req_rw,
input wire [31:0] in_csr_io_req_data,
output wire in_csr_io_req_ready,
input wire csr_io_req_valid_in,
input wire [11:0] csr_io_req_addr_in,
input wire csr_io_req_rw_in,
input wire [31:0] csr_io_req_data_in,
output wire csr_io_req_ready_in,
// input response
input wire [NUM_REQUESTS-1:0] in_csr_io_rsp_valid,
input wire [NUM_REQUESTS-1:0][31:0] in_csr_io_rsp_data,
output wire [NUM_REQUESTS-1:0] in_csr_io_rsp_ready,
input wire [NUM_REQUESTS-1:0] csr_io_rsp_valid_in,
input wire [NUM_REQUESTS-1:0][31:0] csr_io_rsp_data_in,
output wire [NUM_REQUESTS-1:0] csr_io_rsp_ready_in,
// output request
output wire [NUM_REQUESTS-1:0] out_csr_io_req_valid,
output wire [NUM_REQUESTS-1:0][11:0] out_csr_io_req_addr,
output wire [NUM_REQUESTS-1:0] out_csr_io_req_rw,
output wire [NUM_REQUESTS-1:0][31:0] out_csr_io_req_data,
input wire [NUM_REQUESTS-1:0] out_csr_io_req_ready,
output wire [NUM_REQUESTS-1:0] csr_io_req_valid_out,
output wire [NUM_REQUESTS-1:0][11:0] csr_io_req_addr_out,
output wire [NUM_REQUESTS-1:0] csr_io_req_rw_out,
output wire [NUM_REQUESTS-1:0][31:0] csr_io_req_data_out,
input wire [NUM_REQUESTS-1:0] csr_io_req_ready_out,
// output response
output wire out_csr_io_rsp_valid,
output wire [31:0] out_csr_io_rsp_data,
input wire out_csr_io_rsp_ready
output wire csr_io_rsp_valid_out,
output wire [31:0] csr_io_rsp_data_out,
input wire csr_io_rsp_ready_out
);
if (NUM_REQUESTS == 1) begin
@@ -39,26 +39,26 @@ module VX_csr_io_arb #(
`UNUSED_VAR (reset)
`UNUSED_VAR (request_id)
assign out_csr_io_req_valid = in_csr_io_req_valid;
assign out_csr_io_req_rw = in_csr_io_req_rw;
assign out_csr_io_req_addr = in_csr_io_req_addr;
assign out_csr_io_req_data = in_csr_io_req_data;
assign in_csr_io_req_ready = out_csr_io_req_ready;
assign csr_io_req_valid_out = csr_io_req_valid_in;
assign csr_io_req_rw_out = csr_io_req_rw_in;
assign csr_io_req_addr_out = csr_io_req_addr_in;
assign csr_io_req_data_out = csr_io_req_data_in;
assign csr_io_req_ready_in = csr_io_req_ready_out;
assign out_csr_io_rsp_valid = in_csr_io_rsp_valid;
assign out_csr_io_rsp_data = in_csr_io_rsp_data;
assign in_csr_io_rsp_ready = out_csr_io_rsp_ready;
assign csr_io_rsp_valid_out = csr_io_rsp_valid_in;
assign csr_io_rsp_data_out = csr_io_rsp_data_in;
assign csr_io_rsp_ready_in = csr_io_rsp_ready_out;
end else begin
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign out_csr_io_req_valid[i] = in_csr_io_req_valid && (request_id == `REQS_BITS'(i));
assign out_csr_io_req_rw[i] = in_csr_io_req_rw;
assign out_csr_io_req_addr[i] = in_csr_io_req_addr;
assign out_csr_io_req_data[i] = in_csr_io_req_data;
assign csr_io_req_valid_out[i] = csr_io_req_valid_in && (request_id == `REQS_BITS'(i));
assign csr_io_req_rw_out[i] = csr_io_req_rw_in;
assign csr_io_req_addr_out[i] = csr_io_req_addr_in;
assign csr_io_req_data_out[i] = csr_io_req_data_in;
end
assign in_csr_io_req_ready = out_csr_io_req_ready[request_id];
assign csr_io_req_ready_in = csr_io_req_ready_out[request_id];
reg [REQS_BITS-1:0] bus_rsp_sel;
@@ -67,17 +67,17 @@ module VX_csr_io_arb #(
) arbiter (
.clk (clk),
.reset (reset),
.requests (in_csr_io_rsp_valid),
.requests (csr_io_rsp_valid_in),
.grant_index (bus_rsp_sel),
`UNUSED_PIN (grant_valid),
`UNUSED_PIN (grant_onehot)
);
assign out_csr_io_rsp_valid = in_csr_io_rsp_valid [bus_rsp_sel];
assign out_csr_io_rsp_data = in_csr_io_rsp_data [bus_rsp_sel];
assign csr_io_rsp_valid_out = csr_io_rsp_valid_in [bus_rsp_sel];
assign csr_io_rsp_data_out = csr_io_rsp_data_in [bus_rsp_sel];
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign in_csr_io_rsp_ready[i] = out_csr_io_rsp_ready && (bus_rsp_sel == `REQS_BITS'(i));
assign csr_io_rsp_ready_in[i] = csr_io_rsp_ready_out && (bus_rsp_sel == `REQS_BITS'(i));
end
end

7
hw/rtl/VX_csr_unit.v
View File

@@ -13,7 +13,9 @@ module VX_csr_unit #(
VX_csr_io_rsp_if csr_io_rsp_if,
VX_csr_req_if csr_req_if,
VX_exu_to_cmt_if csr_commit_if
VX_exu_to_cmt_if csr_commit_if,
input wire busy
);
VX_csr_req_if csr_pipe_req_if();
VX_exu_to_cmt_if csr_pipe_rsp_if();
@@ -53,7 +55,8 @@ module VX_csr_unit #(
.write_enable (csr_we_s1),
.write_addr (csr_addr_s1),
.write_wid (csr_pipe_rsp_if.wid),
.write_data (csr_updated_data_s1[`CSR_WIDTH-1:0])
.write_data (csr_updated_data_s1[`CSR_WIDTH-1:0]),
.busy (busy)
);
wire csr_hazard = (csr_addr_s1 == csr_pipe_req_if.csr_addr)

4
hw/rtl/VX_execute.v
View File

@@ -38,6 +38,7 @@ module VX_execute #(
VX_fpu_to_cmt_if fpu_commit_if,
VX_exu_to_cmt_if gpu_commit_if,
input wire busy,
output wire ebreak
);
@@ -73,7 +74,8 @@ module VX_execute #(
.csr_io_req_if (csr_io_req_if),
.csr_io_rsp_if (csr_io_rsp_if),
.csr_req_if (csr_req_if),
.csr_commit_if (csr_commit_if)
.csr_commit_if (csr_commit_if),
.busy (busy)
);
`ifdef EXT_M_ENABLE

28
hw/rtl/VX_gpu_unit.v
View File

@@ -73,22 +73,24 @@ module VX_gpu_unit #(
// output
assign warp_ctl_if.valid = gpu_req_if.valid && gpu_commit_if.ready;
assign warp_ctl_if.wid = gpu_commit_if.wid;
assign warp_ctl_if.tmc = tmc;
assign warp_ctl_if.wspawn = wspawn;
assign warp_ctl_if.split = split;
assign warp_ctl_if.barrier = barrier;
wire stall = ~gpu_commit_if.ready && gpu_commit_if.valid;
assign gpu_commit_if.valid = gpu_req_if.valid;
assign gpu_commit_if.wid = gpu_req_if.wid;
assign gpu_commit_if.tmask = gpu_req_if.tmask;
assign gpu_commit_if.PC = gpu_req_if.PC;
assign gpu_commit_if.rd = gpu_req_if.rd;
assign gpu_commit_if.wb = gpu_req_if.wb;
VX_generic_register #(
.N(1 + `NW_BITS + `NUM_THREADS + 32 + `NR_BITS + 1 + `GPU_TMC_SIZE + `GPU_WSPAWN_SIZE + `GPU_SPLIT_SIZE + `GPU_BARRIER_SIZE)
) csr_reg (
.clk (clk),
.reset (reset),
.stall (stall),
.flush (1'b0),
.in ({gpu_req_if.valid, gpu_req_if.wid, gpu_req_if.tmask, gpu_req_if.PC, gpu_req_if.rd, gpu_req_if.wb, tmc, wspawn, split, barrier}),
.out ({gpu_commit_if.valid, gpu_commit_if.wid, gpu_commit_if.tmask, gpu_commit_if.PC, gpu_commit_if.rd, gpu_commit_if.wb, warp_ctl_if.tmc, warp_ctl_if.wspawn, warp_ctl_if.split, warp_ctl_if.barrier})
);
assign warp_ctl_if.valid = gpu_commit_if.valid && gpu_commit_if.ready;
assign warp_ctl_if.wid = gpu_commit_if.wid;
// can accept new request?
assign gpu_req_if.ready = gpu_commit_if.ready;
assign gpu_req_if.ready = ~stall;
`SCOPE_ASSIGN (gpu_req_fire, gpu_req_if.valid && gpu_req_if.ready);
`SCOPE_ASSIGN (gpu_req_wid, gpu_req_if.wid);

92
hw/rtl/VX_io_arb.v
View File

@@ -14,52 +14,52 @@ module VX_io_arb #(
input wire reset,
// input requests
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0] in_io_req_valid,
input wire [NUM_REQUESTS-1:0] in_io_req_rw,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0][WORD_SIZE-1:0] in_io_req_byteen,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0][ADDR_WIDTH-1:0] in_io_req_addr,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0][WORD_WIDTH-1:0] in_io_req_data,
input wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] in_io_req_tag,
output wire [NUM_REQUESTS-1:0] in_io_req_ready,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0] io_req_valid_in,
input wire [NUM_REQUESTS-1:0] io_req_rw_in,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0][WORD_SIZE-1:0] io_req_byteen_in,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0][ADDR_WIDTH-1:0] io_req_addr_in,
input wire [NUM_REQUESTS-1:0][`NUM_THREADS-1:0][WORD_WIDTH-1:0] io_req_data_in,
input wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] io_req_tag_in,
output wire [NUM_REQUESTS-1:0] io_req_ready_in,
// input response
output wire [NUM_REQUESTS-1:0] in_io_rsp_valid,
output wire [NUM_REQUESTS-1:0][WORD_WIDTH-1:0] in_io_rsp_data,
output wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] in_io_rsp_tag,
input wire [NUM_REQUESTS-1:0] in_io_rsp_ready,
output wire [NUM_REQUESTS-1:0] io_rsp_valid_in,
output wire [NUM_REQUESTS-1:0][WORD_WIDTH-1:0] io_rsp_data_in,
output wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] io_rsp_tag_in,
input wire [NUM_REQUESTS-1:0] io_rsp_ready_in,
// output request
output wire [`NUM_THREADS-1:0] out_io_req_valid,
output wire out_io_req_rw,
output wire [`NUM_THREADS-1:0][WORD_SIZE-1:0] out_io_req_byteen,
output wire [`NUM_THREADS-1:0][ADDR_WIDTH-1:0] out_io_req_addr,
output wire [`NUM_THREADS-1:0][WORD_WIDTH-1:0] out_io_req_data,
output wire [TAG_OUT_WIDTH-1:0] out_io_req_tag,
input wire out_io_req_ready,
output wire [`NUM_THREADS-1:0] io_req_valid_out,
output wire io_req_rw_out,
output wire [`NUM_THREADS-1:0][WORD_SIZE-1:0] io_req_byteen_out,
output wire [`NUM_THREADS-1:0][ADDR_WIDTH-1:0] io_req_addr_out,
output wire [`NUM_THREADS-1:0][WORD_WIDTH-1:0] io_req_data_out,
output wire [TAG_OUT_WIDTH-1:0] io_req_tag_out,
input wire io_req_ready_out,
// output response
input wire out_io_rsp_valid,
input wire [WORD_WIDTH-1:0] out_io_rsp_data,
input wire [TAG_OUT_WIDTH-1:0] out_io_rsp_tag,
output wire out_io_rsp_ready
input wire io_rsp_valid_out,
input wire [WORD_WIDTH-1:0] io_rsp_data_out,
input wire [TAG_OUT_WIDTH-1:0] io_rsp_tag_out,
output wire io_rsp_ready_out
);
if (NUM_REQUESTS == 1) begin
`UNUSED_VAR (clk)
`UNUSED_VAR (reset)
assign out_io_req_valid = in_io_req_valid;
assign out_io_req_rw = in_io_req_rw;
assign out_io_req_byteen = in_io_req_byteen;
assign out_io_req_addr = in_io_req_addr;
assign out_io_req_data = in_io_req_data;
assign out_io_req_tag = in_io_req_tag;
assign in_io_req_ready = out_io_req_ready;
assign io_req_valid_out = io_req_valid_in;
assign io_req_rw_out = io_req_rw_in;
assign io_req_byteen_out = io_req_byteen_in;
assign io_req_addr_out = io_req_addr_in;
assign io_req_data_out = io_req_data_in;
assign io_req_tag_out = io_req_tag_in;
assign io_req_ready_in = io_req_ready_out;
assign in_io_rsp_valid = out_io_rsp_valid;
assign in_io_rsp_data = out_io_rsp_data;
assign in_io_rsp_tag = out_io_rsp_tag;
assign out_io_rsp_ready = in_io_rsp_ready;
assign io_rsp_valid_in = io_rsp_valid_out;
assign io_rsp_data_in = io_rsp_data_out;
assign io_rsp_tag_in = io_rsp_tag_out;
assign io_rsp_ready_out = io_rsp_ready_in;
end else begin
@@ -68,7 +68,7 @@ module VX_io_arb #(
wire [NUM_REQUESTS-1:0] valid_requests;
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign valid_requests[i] = (| in_io_req_valid[i]);
assign valid_requests[i] = (| io_req_valid_in[i]);
end
VX_rr_arbiter #(
@@ -82,25 +82,25 @@ module VX_io_arb #(
`UNUSED_PIN (grant_onehot)
);
assign out_io_req_valid = in_io_req_valid [bus_req_sel];
assign out_io_req_rw = in_io_req_rw [bus_req_sel];
assign out_io_req_byteen = in_io_req_byteen [bus_req_sel];
assign out_io_req_addr = in_io_req_addr [bus_req_sel];
assign out_io_req_data = in_io_req_data [bus_req_sel];
assign out_io_req_tag = {in_io_req_tag [bus_req_sel], REQS_BITS'(bus_req_sel)};
assign io_req_valid_out = io_req_valid_in [bus_req_sel];
assign io_req_rw_out = io_req_rw_in [bus_req_sel];
assign io_req_byteen_out = io_req_byteen_in [bus_req_sel];
assign io_req_addr_out = io_req_addr_in [bus_req_sel];
assign io_req_data_out = io_req_data_in [bus_req_sel];
assign io_req_tag_out = {io_req_tag_in [bus_req_sel], REQS_BITS'(bus_req_sel)};
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign in_io_req_ready[i] = out_io_req_ready && (bus_req_sel == REQS_BITS'(i));
assign io_req_ready_in[i] = io_req_ready_out && (bus_req_sel == REQS_BITS'(i));
end
wire [REQS_BITS-1:0] bus_rsp_sel = out_io_rsp_tag[REQS_BITS-1:0];
wire [REQS_BITS-1:0] bus_rsp_sel = io_rsp_tag_out[REQS_BITS-1:0];
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign in_io_rsp_valid[i] = out_io_rsp_valid && (bus_rsp_sel == REQS_BITS'(i));
assign in_io_rsp_data[i] = out_io_rsp_data;
assign in_io_rsp_tag[i] = out_io_rsp_tag[REQS_BITS +: TAG_IN_WIDTH];
assign io_rsp_valid_in[i] = io_rsp_valid_out && (bus_rsp_sel == REQS_BITS'(i));
assign io_rsp_data_in[i] = io_rsp_data_out;
assign io_rsp_tag_in[i] = io_rsp_tag_out[REQS_BITS +: TAG_IN_WIDTH];
end
assign out_io_rsp_ready = in_io_rsp_ready[bus_rsp_sel];
assign io_rsp_ready_out = io_rsp_ready_in[bus_rsp_sel];
end

92
hw/rtl/VX_mem_arb.v
View File

@@ -14,52 +14,52 @@ module VX_mem_arb #(
input wire reset,
// input requests
input wire [NUM_REQUESTS-1:0] in_mem_req_valid,
input wire [NUM_REQUESTS-1:0] in_mem_req_rw,
input wire [NUM_REQUESTS-1:0][WORD_SIZE-1:0] in_mem_req_byteen,
input wire [NUM_REQUESTS-1:0][ADDR_WIDTH-1:0] in_mem_req_addr,
input wire [NUM_REQUESTS-1:0][WORD_WIDTH-1:0] in_mem_req_data,
input wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] in_mem_req_tag,
output wire [NUM_REQUESTS-1:0] in_mem_req_ready,
input wire [NUM_REQUESTS-1:0] mem_req_valid_in,
input wire [NUM_REQUESTS-1:0] mem_req_rw_in,
input wire [NUM_REQUESTS-1:0][WORD_SIZE-1:0] mem_req_byteen_in,
input wire [NUM_REQUESTS-1:0][ADDR_WIDTH-1:0] mem_req_addr_in,
input wire [NUM_REQUESTS-1:0][WORD_WIDTH-1:0] mem_req_data_in,
input wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] mem_req_tag_in,
output wire [NUM_REQUESTS-1:0] mem_req_ready_in,
// input response
output wire [NUM_REQUESTS-1:0] in_mem_rsp_valid,
output wire [NUM_REQUESTS-1:0][WORD_WIDTH-1:0] in_mem_rsp_data,
output wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] in_mem_rsp_tag,
input wire [NUM_REQUESTS-1:0] in_mem_rsp_ready,
output wire [NUM_REQUESTS-1:0] mem_rsp_valid_in,
output wire [NUM_REQUESTS-1:0][WORD_WIDTH-1:0] mem_rsp_data_in,
output wire [NUM_REQUESTS-1:0][TAG_IN_WIDTH-1:0] mem_rsp_tag_in,
input wire [NUM_REQUESTS-1:0] mem_rsp_ready_in,
// output request
output wire out_mem_req_valid,
output wire out_mem_req_rw,
output wire [WORD_SIZE-1:0] out_mem_req_byteen,
output wire [ADDR_WIDTH-1:0] out_mem_req_addr,
output wire [WORD_WIDTH-1:0] out_mem_req_data,
output wire [TAG_OUT_WIDTH-1:0] out_mem_req_tag,
input wire out_mem_req_ready,
output wire mem_req_valid_out,
output wire mem_req_rw_out,
output wire [WORD_SIZE-1:0] mem_req_byteen_out,
output wire [ADDR_WIDTH-1:0] mem_req_addr_out,
output wire [WORD_WIDTH-1:0] mem_req_data_out,
output wire [TAG_OUT_WIDTH-1:0] mem_req_tag_out,
input wire mem_req_ready_out,
// output response
input wire out_mem_rsp_valid,
input wire [WORD_WIDTH-1:0] out_mem_rsp_data,
input wire [TAG_OUT_WIDTH-1:0] out_mem_rsp_tag,
output wire out_mem_rsp_ready
input wire mem_rsp_valid_out,
input wire [WORD_WIDTH-1:0] mem_rsp_data_out,
input wire [TAG_OUT_WIDTH-1:0] mem_rsp_tag_out,
output wire mem_rsp_ready_out
);
if (NUM_REQUESTS == 1) begin
`UNUSED_VAR (clk)
`UNUSED_VAR (reset)
assign out_mem_req_valid = in_mem_req_valid;
assign out_mem_req_rw = in_mem_req_rw;
assign out_mem_req_byteen = in_mem_req_byteen;
assign out_mem_req_addr = in_mem_req_addr;
assign out_mem_req_data = in_mem_req_data;
assign out_mem_req_tag = in_mem_req_tag;
assign in_mem_req_ready = out_mem_req_ready;
assign mem_req_valid_out = mem_req_valid_in;
assign mem_req_rw_out = mem_req_rw_in;
assign mem_req_byteen_out = mem_req_byteen_in;
assign mem_req_addr_out = mem_req_addr_in;
assign mem_req_data_out = mem_req_data_in;
assign mem_req_tag_out = mem_req_tag_in;
assign mem_req_ready_in = mem_req_ready_out;
assign in_mem_rsp_valid = out_mem_rsp_valid;
assign in_mem_rsp_data = out_mem_rsp_data;
assign in_mem_rsp_tag = out_mem_rsp_tag;
assign out_mem_rsp_ready = in_mem_rsp_ready;
assign mem_rsp_valid_in = mem_rsp_valid_out;
assign mem_rsp_data_in = mem_rsp_data_out;
assign mem_rsp_tag_in = mem_rsp_tag_out;
assign mem_rsp_ready_out = mem_rsp_ready_in;
end else begin
@@ -70,31 +70,31 @@ module VX_mem_arb #(
) arbiter (
.clk (clk),
.reset (reset),
.requests (in_mem_req_valid),
.requests (mem_req_valid_in),
.grant_index (bus_req_sel),
`UNUSED_PIN (grant_valid),
`UNUSED_PIN (grant_onehot)
);
assign out_mem_req_valid = in_mem_req_valid [bus_req_sel];
assign out_mem_req_rw = in_mem_req_rw [bus_req_sel];
assign out_mem_req_byteen = in_mem_req_byteen [bus_req_sel];
assign out_mem_req_addr = in_mem_req_addr [bus_req_sel];
assign out_mem_req_data = in_mem_req_data [bus_req_sel];
assign out_mem_req_tag = {in_mem_req_tag [bus_req_sel], REQS_BITS'(bus_req_sel)};
assign mem_req_valid_out = mem_req_valid_in [bus_req_sel];
assign mem_req_rw_out = mem_req_rw_in [bus_req_sel];
assign mem_req_byteen_out = mem_req_byteen_in [bus_req_sel];
assign mem_req_addr_out = mem_req_addr_in [bus_req_sel];
assign mem_req_data_out = mem_req_data_in [bus_req_sel];
assign mem_req_tag_out = {mem_req_tag_in [bus_req_sel], REQS_BITS'(bus_req_sel)};
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign in_mem_req_ready[i] = out_mem_req_ready && (bus_req_sel == REQS_BITS'(i));
assign mem_req_ready_in[i] = mem_req_ready_out && (bus_req_sel == REQS_BITS'(i));
end
wire [REQS_BITS-1:0] bus_rsp_sel = out_mem_rsp_tag[REQS_BITS-1:0];
wire [REQS_BITS-1:0] bus_rsp_sel = mem_rsp_tag_out[REQS_BITS-1:0];
for (genvar i = 0; i < NUM_REQUESTS; i++) begin
assign in_mem_rsp_valid[i] = out_mem_rsp_valid && (bus_rsp_sel == REQS_BITS'(i));
assign in_mem_rsp_data[i] = out_mem_rsp_data;
assign in_mem_rsp_tag[i] = out_mem_rsp_tag[REQS_BITS +: TAG_IN_WIDTH];
assign mem_rsp_valid_in[i] = mem_rsp_valid_out && (bus_rsp_sel == REQS_BITS'(i));
assign mem_rsp_data_in[i] = mem_rsp_data_out;
assign mem_rsp_tag_in[i] = mem_rsp_tag_out[REQS_BITS +: TAG_IN_WIDTH];
end
assign out_mem_rsp_ready = in_mem_rsp_ready[bus_rsp_sel];
assign mem_rsp_ready_out = mem_rsp_ready_in[bus_rsp_sel];
end

19
hw/rtl/VX_mem_unit.v
View File

@@ -12,7 +12,7 @@ module VX_mem_unit # (
VX_cache_core_req_if core_dcache_req_if,
VX_cache_core_rsp_if core_dcache_rsp_if,
// Dram <-> Dcache
// DRAM <-> Dcache
VX_cache_dram_req_if dcache_dram_req_if,
VX_cache_dram_rsp_if dcache_dram_rsp_if,
VX_cache_snp_req_if dcache_snp_req_if,
@@ -22,7 +22,7 @@ module VX_mem_unit # (
VX_cache_core_req_if core_icache_req_if,
VX_cache_core_rsp_if core_icache_rsp_if,
// Dram <-> Icache
// DRAM <-> Icache
VX_cache_dram_req_if icache_dram_req_if,
VX_cache_dram_rsp_if icache_dram_rsp_if
);
@@ -65,11 +65,10 @@ module VX_mem_unit # (
.NUM_REQUESTS (`SNUM_REQUESTS),
.CREQ_SIZE (`SCREQ_SIZE),
.MRVQ_SIZE (8),
.DFPQ_SIZE (1),
.DRPQ_SIZE (1),
.SNRQ_SIZE (1),
.CWBQ_SIZE (`SCWBQ_SIZE),
.DWBQ_SIZE (1),
.DFQQ_SIZE (1),
.DREQ_SIZE (1),
.SNOOP_FORWARDING (0),
.DRAM_ENABLE (0),
.WRITE_ENABLE (1),
@@ -146,11 +145,10 @@ module VX_mem_unit # (
.NUM_REQUESTS (`DNUM_REQUESTS),
.CREQ_SIZE (`DCREQ_SIZE),
.MRVQ_SIZE (`DMRVQ_SIZE),
.DFPQ_SIZE (`DDFPQ_SIZE),
.DRPQ_SIZE (`DDRPQ_SIZE),
.SNRQ_SIZE (`DSNRQ_SIZE),
.CWBQ_SIZE (`DCWBQ_SIZE),
.DWBQ_SIZE (`DDWBQ_SIZE),
.DFQQ_SIZE (`DDFQQ_SIZE),
.DREQ_SIZE (`DDREQ_SIZE),
.SNOOP_FORWARDING (0),
.DRAM_ENABLE (1),
.WRITE_ENABLE (1),
@@ -228,11 +226,10 @@ module VX_mem_unit # (
.NUM_REQUESTS (`INUM_REQUESTS),
.CREQ_SIZE (`ICREQ_SIZE),
.MRVQ_SIZE (`IMRVQ_SIZE),
.DFPQ_SIZE (`IDFPQ_SIZE),
.DRPQ_SIZE (`IDRPQ_SIZE),
.SNRQ_SIZE (1),
.CWBQ_SIZE (`ICWBQ_SIZE),
.DWBQ_SIZE (`IDWBQ_SIZE),
.DFQQ_SIZE (`IDFQQ_SIZE),
.DREQ_SIZE (`IDREQ_SIZE),
.SNOOP_FORWARDING (0),
.DRAM_ENABLE (1),
.WRITE_ENABLE (0),

1
hw/rtl/VX_pipeline.v
View File

@@ -200,6 +200,7 @@ module VX_pipeline #(
.fpu_commit_if (fpu_commit_if),
.gpu_commit_if (gpu_commit_if),
.busy (busy),
.ebreak (ebreak)
);

2
hw/rtl/VX_scope.vh
View File

@@ -7,6 +7,8 @@
`define SCOPE_ASSIGN(d,s) assign scope_``d = s
`define SCOPE_SIZE 4096
`else
`define SCOPE_IO_VX_icache_stage

195
hw/rtl/Vortex.v
View File

@@ -135,7 +135,7 @@ module Vortex (
wire [`NUM_CLUSTERS-1:0][`L2DRAM_ADDR_WIDTH-1:0] per_cluster_dram_req_addr;
wire [`NUM_CLUSTERS-1:0][`L2DRAM_LINE_WIDTH-1:0] per_cluster_dram_req_data;
wire [`NUM_CLUSTERS-1:0][`L2DRAM_TAG_WIDTH-1:0] per_cluster_dram_req_tag;
wire l3_core_req_ready;
wire cluster_dram_req_ready;
wire [`NUM_CLUSTERS-1:0] per_cluster_dram_rsp_valid;
wire [`NUM_CLUSTERS-1:0][`L2DRAM_LINE_WIDTH-1:0] per_cluster_dram_rsp_data;
@@ -196,7 +196,7 @@ module Vortex (
.dram_req_addr (per_cluster_dram_req_addr [i]),
.dram_req_data (per_cluster_dram_req_data [i]),
.dram_req_tag (per_cluster_dram_req_tag [i]),
.dram_req_ready (l3_core_req_ready),
.dram_req_ready (cluster_dram_req_ready),
.dram_rsp_valid (per_cluster_dram_rsp_valid [i]),
.dram_rsp_data (per_cluster_dram_rsp_data [i]),
@@ -252,34 +252,34 @@ module Vortex (
.reset (reset),
// input requests
.in_io_req_valid (per_cluster_io_req_valid),
.in_io_req_rw (per_cluster_io_req_rw),
.in_io_req_byteen (per_cluster_io_req_byteen),
.in_io_req_addr (per_cluster_io_req_addr),
.in_io_req_data (per_cluster_io_req_data),
.in_io_req_tag (per_cluster_io_req_tag),
.in_io_req_ready (per_cluster_io_req_ready),
.io_req_valid_in (per_cluster_io_req_valid),
.io_req_rw_in (per_cluster_io_req_rw),
.io_req_byteen_in (per_cluster_io_req_byteen),
.io_req_addr_in (per_cluster_io_req_addr),
.io_req_data_in (per_cluster_io_req_data),
.io_req_tag_in (per_cluster_io_req_tag),
.io_req_ready_in (per_cluster_io_req_ready),
// input responses
.in_io_rsp_valid (per_cluster_io_rsp_valid),
.in_io_rsp_data (per_cluster_io_rsp_data),
.in_io_rsp_tag (per_cluster_io_rsp_tag),
.in_io_rsp_ready (per_cluster_io_rsp_ready),
.io_rsp_valid_in (per_cluster_io_rsp_valid),
.io_rsp_data_in (per_cluster_io_rsp_data),
.io_rsp_tag_in (per_cluster_io_rsp_tag),
.io_rsp_ready_in (per_cluster_io_rsp_ready),
// output request
.out_io_req_valid (io_req_valid),
.out_io_req_rw (io_req_rw),
.out_io_req_byteen (io_req_byteen),
.out_io_req_addr (io_req_addr),
.out_io_req_data (io_req_data),
.out_io_req_tag (io_req_tag),
.out_io_req_ready (io_req_ready),
.io_req_valid_out (io_req_valid),
.io_req_rw_out (io_req_rw),
.io_req_byteen_out (io_req_byteen),
.io_req_addr_out (io_req_addr),
.io_req_data_out (io_req_data),
.io_req_tag_out (io_req_tag),
.io_req_ready_out (io_req_ready),
// output response
.out_io_rsp_valid (io_rsp_valid),
.out_io_rsp_tag (io_rsp_tag),
.out_io_rsp_data (io_rsp_data),
.out_io_rsp_ready (io_rsp_ready)
.io_rsp_valid_out (io_rsp_valid),
.io_rsp_tag_out (io_rsp_tag),
.io_rsp_data_out (io_rsp_data),
.io_rsp_ready_out (io_rsp_ready)
);
VX_csr_io_arb #(
@@ -291,28 +291,28 @@ module Vortex (
.request_id (csr_io_request_id),
// input requests
.in_csr_io_req_valid (csr_io_req_valid),
.in_csr_io_req_addr (csr_io_req_addr),
.in_csr_io_req_rw (csr_io_req_rw),
.in_csr_io_req_data (csr_io_req_data),
.in_csr_io_req_ready (csr_io_req_ready),
.csr_io_req_valid_in (csr_io_req_valid),
.csr_io_req_addr_in (csr_io_req_addr),
.csr_io_req_rw_in (csr_io_req_rw),
.csr_io_req_data_in (csr_io_req_data),
.csr_io_req_ready_in (csr_io_req_ready),
// input responses
.in_csr_io_rsp_valid (per_cluster_csr_io_rsp_valid),
.in_csr_io_rsp_data (per_cluster_csr_io_rsp_data),
.in_csr_io_rsp_ready (per_cluster_csr_io_rsp_ready),
.csr_io_rsp_valid_in (per_cluster_csr_io_rsp_valid),
.csr_io_rsp_data_in (per_cluster_csr_io_rsp_data),
.csr_io_rsp_ready_in (per_cluster_csr_io_rsp_ready),
// output request
.out_csr_io_req_valid (per_cluster_csr_io_req_valid),
.out_csr_io_req_addr (per_cluster_csr_io_req_addr),
.out_csr_io_req_rw (per_cluster_csr_io_req_rw),
.out_csr_io_req_data (per_cluster_csr_io_req_data),
.out_csr_io_req_ready (per_cluster_csr_io_req_ready),
.csr_io_req_valid_out (per_cluster_csr_io_req_valid),
.csr_io_req_addr_out (per_cluster_csr_io_req_addr),
.csr_io_req_rw_out (per_cluster_csr_io_req_rw),
.csr_io_req_data_out (per_cluster_csr_io_req_data),
.csr_io_req_ready_out (per_cluster_csr_io_req_ready),
// output response
.out_csr_io_rsp_valid (csr_io_rsp_valid),
.out_csr_io_rsp_data (csr_io_rsp_data),
.out_csr_io_rsp_ready (csr_io_rsp_ready)
.csr_io_rsp_valid_out (csr_io_rsp_valid),
.csr_io_rsp_data_out (csr_io_rsp_data),
.csr_io_rsp_ready_out (csr_io_rsp_ready)
);
assign busy = (| per_cluster_busy);
@@ -320,56 +320,56 @@ module Vortex (
// L3 Cache ///////////////////////////////////////////////////////////
wire [`L3NUM_REQUESTS-1:0] l3_core_req_valid;
wire [`L3NUM_REQUESTS-1:0] l3_core_req_rw;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_BYTEEN_WIDTH-1:0] l3_core_req_byteen;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_ADDR_WIDTH-1:0] l3_core_req_addr;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_LINE_WIDTH-1:0] l3_core_req_data;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_TAG_WIDTH-1:0] l3_core_req_tag;
wire [`L3NUM_REQUESTS-1:0] cluster_dram_req_valid;
wire [`L3NUM_REQUESTS-1:0] cluster_dram_req_rw;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_BYTEEN_WIDTH-1:0] cluster_dram_req_byteen;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_ADDR_WIDTH-1:0] cluster_dram_req_addr;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_LINE_WIDTH-1:0] cluster_dram_req_data;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_TAG_WIDTH-1:0] cluster_dram_req_tag;
wire [`L3NUM_REQUESTS-1:0] l3_core_rsp_valid;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_LINE_WIDTH-1:0] l3_core_rsp_data;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_TAG_WIDTH-1:0] l3_core_rsp_tag;
wire l3_core_rsp_ready;
wire [`L3NUM_REQUESTS-1:0] cluster_dram_rsp_valid;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_LINE_WIDTH-1:0] cluster_dram_rsp_data;
wire [`L3NUM_REQUESTS-1:0][`L2DRAM_TAG_WIDTH-1:0] cluster_dram_rsp_tag;
wire cluster_dram_rsp_ready;
wire [`NUM_CLUSTERS-1:0] l3_snp_fwdout_valid;
wire [`NUM_CLUSTERS-1:0][`L2DRAM_ADDR_WIDTH-1:0] l3_snp_fwdout_addr;
wire [`NUM_CLUSTERS-1:0] l3_snp_fwdout_invalidate;
wire [`NUM_CLUSTERS-1:0][`L2SNP_TAG_WIDTH-1:0] l3_snp_fwdout_tag;
wire [`NUM_CLUSTERS-1:0] l3_snp_fwdout_ready;
wire [`NUM_CLUSTERS-1:0] cluster_snp_fwdout_valid;
wire [`NUM_CLUSTERS-1:0][`L2DRAM_ADDR_WIDTH-1:0] cluster_snp_fwdout_addr;
wire [`NUM_CLUSTERS-1:0] cluster_snp_fwdout_invalidate;
wire [`NUM_CLUSTERS-1:0][`L2SNP_TAG_WIDTH-1:0] cluster_snp_fwdout_tag;
wire [`NUM_CLUSTERS-1:0] cluster_snp_fwdout_ready;
wire [`NUM_CLUSTERS-1:0] l3_snp_fwdin_valid;
wire [`NUM_CLUSTERS-1:0][`L2SNP_TAG_WIDTH-1:0] l3_snp_fwdin_tag;
wire [`NUM_CLUSTERS-1:0] l3_snp_fwdin_ready;
wire [`NUM_CLUSTERS-1:0] cluster_snp_fwdin_valid;
wire [`NUM_CLUSTERS-1:0][`L2SNP_TAG_WIDTH-1:0] cluster_snp_fwdin_tag;
wire [`NUM_CLUSTERS-1:0] cluster_snp_fwdin_ready;
for (genvar i = 0; i < `L3NUM_REQUESTS; i++) begin
// Core Request
assign l3_core_req_valid [i] = per_cluster_dram_req_valid [i];
assign l3_core_req_rw [i] = per_cluster_dram_req_rw [i];
assign l3_core_req_byteen [i] = per_cluster_dram_req_byteen[i];
assign l3_core_req_addr [i] = per_cluster_dram_req_addr [i];
assign l3_core_req_tag [i] = per_cluster_dram_req_tag [i];
assign l3_core_req_data [i] = per_cluster_dram_req_data [i];
assign cluster_dram_req_valid [i] = per_cluster_dram_req_valid [i];
assign cluster_dram_req_rw [i] = per_cluster_dram_req_rw [i];
assign cluster_dram_req_byteen [i] = per_cluster_dram_req_byteen[i];
assign cluster_dram_req_addr [i] = per_cluster_dram_req_addr [i];
assign cluster_dram_req_tag [i] = per_cluster_dram_req_tag [i];
assign cluster_dram_req_data [i] = per_cluster_dram_req_data [i];
// Core Response
assign per_cluster_dram_rsp_valid [i] = l3_core_rsp_valid [i] && l3_core_rsp_ready;
assign per_cluster_dram_rsp_data [i] = l3_core_rsp_data [i];
assign per_cluster_dram_rsp_tag [i] = l3_core_rsp_tag [i];
assign per_cluster_dram_rsp_valid [i] = cluster_dram_rsp_valid [i] && cluster_dram_rsp_ready;
assign per_cluster_dram_rsp_data [i] = cluster_dram_rsp_data [i];
assign per_cluster_dram_rsp_tag [i] = cluster_dram_rsp_tag [i];
// Snoop Forwarding out
assign per_cluster_snp_req_valid [i] = l3_snp_fwdout_valid[i];
assign per_cluster_snp_req_addr [i] = l3_snp_fwdout_addr[i];
assign per_cluster_snp_req_invalidate [i] = l3_snp_fwdout_invalidate[i];
assign per_cluster_snp_req_tag [i] = l3_snp_fwdout_tag[i];
assign l3_snp_fwdout_ready [i] = per_cluster_snp_req_ready[i];
assign per_cluster_snp_req_valid [i] = cluster_snp_fwdout_valid[i];
assign per_cluster_snp_req_addr [i] = cluster_snp_fwdout_addr[i];
assign per_cluster_snp_req_invalidate [i] = cluster_snp_fwdout_invalidate[i];
assign per_cluster_snp_req_tag [i] = cluster_snp_fwdout_tag[i];
assign cluster_snp_fwdout_ready [i] = per_cluster_snp_req_ready[i];
// Snoop Forwarding in
assign l3_snp_fwdin_valid [i] = per_cluster_snp_rsp_valid [i];
assign l3_snp_fwdin_tag [i] = per_cluster_snp_rsp_tag [i];
assign per_cluster_snp_rsp_ready [i] = l3_snp_fwdin_ready [i];
assign cluster_snp_fwdin_valid [i] = per_cluster_snp_rsp_valid [i];
assign cluster_snp_fwdin_tag [i] = per_cluster_snp_rsp_tag [i];
assign per_cluster_snp_rsp_ready [i] = cluster_snp_fwdin_ready [i];
end
assign l3_core_rsp_ready = (& per_cluster_dram_rsp_ready);
assign cluster_dram_rsp_ready = (& per_cluster_dram_rsp_ready);
VX_cache #(
.CACHE_ID (`L3CACHE_ID),
@@ -380,11 +380,10 @@ module Vortex (
.NUM_REQUESTS (`L3NUM_REQUESTS),
.CREQ_SIZE (`L3CREQ_SIZE),
.MRVQ_SIZE (`L3MRVQ_SIZE),
.DFPQ_SIZE (`L3DFPQ_SIZE),
.DRPQ_SIZE (`L3DRPQ_SIZE),
.SNRQ_SIZE (`L3SNRQ_SIZE),
.CWBQ_SIZE (`L3CWBQ_SIZE),
.DWBQ_SIZE (`L3DWBQ_SIZE),
.DFQQ_SIZE (`L3DFQQ_SIZE),
.DREQ_SIZE (`L3DREQ_SIZE),
.DRAM_ENABLE (1),
.WRITE_ENABLE (1),
.SNOOP_FORWARDING (1),
@@ -401,19 +400,19 @@ module Vortex (
.reset (reset),
// Core request
.core_req_valid (l3_core_req_valid),
.core_req_rw (l3_core_req_rw),
.core_req_byteen (l3_core_req_byteen),
.core_req_addr (l3_core_req_addr),
.core_req_data (l3_core_req_data),
.core_req_tag (l3_core_req_tag),
.core_req_ready (l3_core_req_ready),
.core_req_valid (cluster_dram_req_valid),
.core_req_rw (cluster_dram_req_rw),
.core_req_byteen (cluster_dram_req_byteen),
.core_req_addr (cluster_dram_req_addr),
.core_req_data (cluster_dram_req_data),
.core_req_tag (cluster_dram_req_tag),
.core_req_ready (cluster_dram_req_ready),
// Core response
.core_rsp_valid (l3_core_rsp_valid),
.core_rsp_data (l3_core_rsp_data),
.core_rsp_tag (l3_core_rsp_tag),
.core_rsp_ready (l3_core_rsp_ready),
.core_rsp_valid (cluster_dram_rsp_valid),
.core_rsp_data (cluster_dram_rsp_data),
.core_rsp_tag (cluster_dram_rsp_tag),
.core_rsp_ready (cluster_dram_rsp_ready),
// DRAM request
.dram_req_valid (dram_req_valid),
@@ -443,16 +442,16 @@ module Vortex (
.snp_rsp_ready (snp_rsp_ready),
// Snoop forwarding out
.snp_fwdout_valid (l3_snp_fwdout_valid),
.snp_fwdout_addr (l3_snp_fwdout_addr),
.snp_fwdout_invalidate(l3_snp_fwdout_invalidate),
.snp_fwdout_tag (l3_snp_fwdout_tag),
.snp_fwdout_ready (l3_snp_fwdout_ready),
.snp_fwdout_valid (cluster_snp_fwdout_valid),
.snp_fwdout_addr (cluster_snp_fwdout_addr),
.snp_fwdout_invalidate(cluster_snp_fwdout_invalidate),
.snp_fwdout_tag (cluster_snp_fwdout_tag),
.snp_fwdout_ready (cluster_snp_fwdout_ready),
// Snoop forwarding in
.snp_fwdin_valid (l3_snp_fwdin_valid),
.snp_fwdin_tag (l3_snp_fwdin_tag),
.snp_fwdin_ready (l3_snp_fwdin_ready)
.snp_fwdin_valid (cluster_snp_fwdin_valid),
.snp_fwdin_tag (cluster_snp_fwdin_tag),
.snp_fwdin_ready (cluster_snp_fwdin_ready)
);
end

533
hw/rtl/cache/VX_bank.v vendored
View File

@@ -19,18 +19,16 @@ module VX_bank #(
parameter CREQ_SIZE = 0,
// Miss Reserv Queue Knob
parameter MRVQ_SIZE = 0,
// Dram Fill Rsp Queue Size
parameter DFPQ_SIZE = 0,
// DRAM Response Queue Size
parameter DRPQ_SIZE = 0,
// Snoop Req Queue Size
parameter SNRQ_SIZE = 0,
// Queues for writebacks Knobs {1, 2, 4, 8, ...}
// Core Writeback Queue Size
parameter CWBQ_SIZE = 0,
// Dram Writeback Queue Size
parameter DWBQ_SIZE = 0,
// Dram Fill Req Queue Size
parameter DFQQ_SIZE = 0,
// DRAM Request Queue Size
parameter DREQ_SIZE = 0,
// Enable cache writeable
parameter WRITE_ENABLE = 0,
@@ -71,31 +69,28 @@ module VX_bank #(
output wire [CORE_TAG_WIDTH-1:0] core_rsp_tag,
input wire core_rsp_ready,
// Dram Fill Requests
output wire dram_fill_req_valid,
output wire[`LINE_ADDR_WIDTH-1:0] dram_fill_req_addr,
input wire dram_fill_req_ready,
// DRAM request
output wire dram_req_valid,
output wire dram_req_rw,
output wire [BANK_LINE_SIZE-1:0] dram_req_byteen,
output wire [`LINE_ADDR_WIDTH-1:0] dram_req_addr,
output wire [`BANK_LINE_WIDTH-1:0] dram_req_data,
input wire dram_req_ready,
// Dram Fill Response
input wire dram_fill_rsp_valid,
input wire [`BANK_LINE_WIDTH-1:0] dram_fill_rsp_data,
input wire [`LINE_ADDR_WIDTH-1:0] dram_fill_rsp_addr,
output wire dram_fill_rsp_ready,
// DRAM response
input wire dram_rsp_valid,
input wire [`LINE_ADDR_WIDTH-1:0] dram_rsp_addr,
input wire [`BANK_LINE_WIDTH-1:0] dram_rsp_data,
output wire dram_rsp_ready,
// Dram WB Requests
output wire dram_wb_req_valid,
output wire [BANK_LINE_SIZE-1:0] dram_wb_req_byteen,
output wire [`LINE_ADDR_WIDTH-1:0] dram_wb_req_addr,
output wire [`BANK_LINE_WIDTH-1:0] dram_wb_req_data,
input wire dram_wb_req_ready,
// Snp Request
// Snoop Request
input wire snp_req_valid,
input wire [`LINE_ADDR_WIDTH-1:0] snp_req_addr,
input wire snp_req_invalidate,
input wire [SNP_REQ_TAG_WIDTH-1:0] snp_req_tag,
output wire snp_req_ready,
// Snoop Response
output wire snp_rsp_valid,
output wire [SNP_REQ_TAG_WIDTH-1:0] snp_rsp_tag,
input wire snp_rsp_ready
@@ -137,51 +132,51 @@ module VX_bank #(
wire snrq_invalidate_st0;
wire [SNP_REQ_TAG_WIDTH-1:0] snrq_tag_st0;
wire snp_req_fire = snp_req_valid && snp_req_ready;
assign snp_req_ready = !snrq_full;
VX_generic_queue #(
.DATAW(`LINE_ADDR_WIDTH + 1 + SNP_REQ_TAG_WIDTH),
.SIZE(SNRQ_SIZE)
) snp_req_queue (
.clk (clk),
.reset (reset),
.push (snp_req_valid && snp_req_ready),
.data_in ({snp_req_addr, snp_req_invalidate, snp_req_tag}),
.push (snp_req_fire),
.pop (snrq_pop),
.data_in ({snp_req_addr, snp_req_invalidate, snp_req_tag}),
.data_out({snrq_addr_st0, snrq_invalidate_st0, snrq_tag_st0}),
.empty (snrq_empty),
.full (snrq_full),
`UNUSED_PIN (size)
);
assign snp_req_ready = !snrq_full;
wire dfpq_pop;
wire dfpq_empty;
wire dfpq_full;
wire [`LINE_ADDR_WIDTH-1:0] dfpq_addr_st0;
wire [`BANK_LINE_WIDTH-1:0] dfpq_filldata_st0;
wire dram_rsp_fire = dram_rsp_valid && dram_rsp_ready;
assign dram_rsp_ready = !dfpq_full;
VX_generic_queue #(
.DATAW(`LINE_ADDR_WIDTH + $bits(dram_fill_rsp_data)),
.SIZE(DFPQ_SIZE)
.DATAW(`LINE_ADDR_WIDTH + $bits(dram_rsp_data)),
.SIZE(DRPQ_SIZE)
) dfp_queue (
.clk (clk),
.reset (reset),
.push (dram_fill_rsp_valid && dram_fill_rsp_ready),
.data_in ({dram_fill_rsp_addr, dram_fill_rsp_data}),
.push (dram_rsp_fire),
.pop (dfpq_pop),
.data_in ({dram_rsp_addr, dram_rsp_data}),
.data_out({dfpq_addr_st0, dfpq_filldata_st0}),
.empty (dfpq_empty),
.full (dfpq_full),
`UNUSED_PIN (size)
);
assign dram_fill_rsp_ready = !dfpq_full;
wire reqq_pop;
wire reqq_push;
wire reqq_empty;
wire reqq_full;
wire reqq_req_st0;
wire [`REQS_BITS-1:0] reqq_req_tid_st0;
wire reqq_req_rw_st0;
wire [WORD_SIZE-1:0] reqq_req_byteen_st0;
@@ -191,6 +186,9 @@ module VX_bank #(
wire [`WORD_WIDTH-1:0] reqq_req_writeword_st0;
wire [CORE_TAG_WIDTH-1:0] reqq_req_tag_st0;
wire core_req_fire = (| core_req_valid) && core_req_ready;
assign core_req_ready = !reqq_full;
VX_bank_core_req_arb #(
.WORD_SIZE (WORD_SIZE),
.NUM_REQUESTS (NUM_REQUESTS),
@@ -201,7 +199,7 @@ module VX_bank #(
.clk (clk),
.reset (reset),
// Enqueue
.reqq_push (reqq_push),
.reqq_push (core_req_fire),
.bank_valids (core_req_valid),
.bank_rw (core_req_rw),
.bank_byteen (core_req_byteen),
@@ -211,7 +209,6 @@ module VX_bank #(
// Dequeue
.reqq_pop (reqq_pop),
.reqq_req_st0 (reqq_req_st0),
.reqq_req_tid_st0 (reqq_req_tid_st0),
.reqq_req_rw_st0 (reqq_req_rw_st0),
.reqq_req_byteen_st0 (reqq_req_byteen_st0),
@@ -222,26 +219,21 @@ module VX_bank #(
.reqq_full (reqq_full)
);
assign core_req_ready = !reqq_full;
assign reqq_push = (| core_req_valid) && core_req_ready;
wire mrvq_pop;
wire mrvq_full;
wire mrvq_stop;
wire mrvq_valid_st0;
wire[`REQS_BITS-1:0] mrvq_tid_st0;
wire [`LINE_ADDR_WIDTH-1:0] mrvq_addr_st0;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] mrvq_wsel_st0;
wire [`WORD_WIDTH-1:0] mrvq_writeword_st0;
wire [`REQ_TAG_WIDTH-1:0] mrvq_tag_st0;
wire mrvq_rw_st0;
wire [WORD_SIZE-1:0] mrvq_byteen_st0;
wire mrvq_is_snp_st0;
wire mrvq_snp_invalidate_st0;
wire mrvq_pending_hazard_st1;
wire st2_pending_hazard_st1;
wire force_request_miss_st1;
wire msrq_pop;
wire msrq_full;
wire msrq_almfull;
wire msrq_valid_st0;
wire[`REQS_BITS-1:0] msrq_tid_st0;
wire [`LINE_ADDR_WIDTH-1:0] msrq_addr_st0;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] msrq_wsel_st0;
wire [`WORD_WIDTH-1:0] msrq_writeword_st0;
wire [`REQ_TAG_WIDTH-1:0] msrq_tag_st0;
wire msrq_rw_st0;
wire [WORD_SIZE-1:0] msrq_byteen_st0;
wire msrq_is_snp_st0;
wire msrq_snp_invalidate_st0;
wire msrq_pending_hazard_st0;
wire msrq_pending_hazard_st1;
wire[`REQS_BITS-1:0] miss_add_tid;
wire[`REQ_TAG_WIDTH-1:0] miss_add_tag;
@@ -249,48 +241,41 @@ module VX_bank #(
wire[WORD_SIZE-1:0] miss_add_byteen;
wire[`LINE_ADDR_WIDTH-1:0] addr_st2;
wire is_fill_st2;
wire recover_mrvq_state_st2;
wire is_msrq_miss_st2;
wire mrvq_push_stall;
wire msrq_push_stall;
wire cwbq_push_stall;
wire dwbq_push_stall;
wire dram_fill_req_stall;
wire stall_bank_pipe;
wire is_fill_st1;
`DEBUG_BEGIN
wire going_to_write_st1;
`DEBUG_END
//determines if the if it is time to pop a req from the queues
//unqual - the req does NOT qualify for execution in the bank.
wire mrvq_pop_unqual = mrvq_valid_st0;
wire dfpq_pop_unqual = !mrvq_pop_unqual && !dfpq_empty;
wire reqq_pop_unqual = !mrvq_stop && !mrvq_pop_unqual && !dfpq_pop_unqual && !reqq_empty && reqq_req_st0 && !is_fill_st1 && !is_fill_st1;
wire snrq_pop_unqual = !mrvq_stop && !reqq_pop_unqual && !reqq_pop_unqual && !mrvq_pop_unqual && !dfpq_pop_unqual && !snrq_empty && !reqq_req_st0; // if there's any reqq_req, don't schedule snrq.
// determine which queue to pop next in piority order
wire msrq_pop_unqual = msrq_valid_st0;
wire dfpq_pop_unqual = !msrq_pop_unqual && !dfpq_empty;
wire reqq_pop_unqual = !msrq_pop_unqual && !dfpq_pop_unqual && !reqq_empty && !msrq_almfull;
wire snrq_pop_unqual = !msrq_pop_unqual && !dfpq_pop_unqual && !reqq_pop_unqual && !snrq_empty && !msrq_almfull;
assign mrvq_pop = mrvq_pop_unqual && !stall_bank_pipe && !recover_mrvq_state_st2;
assign msrq_pop = msrq_pop_unqual && !stall_bank_pipe
&& !is_msrq_miss_st2; // stop if previous request was a miss
assign dfpq_pop = dfpq_pop_unqual && !stall_bank_pipe;
assign reqq_pop = reqq_pop_unqual && !stall_bank_pipe;
assign snrq_pop = snrq_pop_unqual && !stall_bank_pipe;
//signals to progress to the next stage
wire qual_is_fill_st0;
wire qual_valid_st0;
wire [`LINE_ADDR_WIDTH-1:0] qual_addr_st0;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] qual_wsel_st0;
wire qual_is_mrvq_st0;
wire is_fill_st0;
wire valid_st0;
wire [`LINE_ADDR_WIDTH-1:0] addr_st0;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] wsel_st0;
wire is_msrq_st0;
wire [`WORD_WIDTH-1:0] qual_writeword_st0;
wire [`BANK_LINE_WIDTH-1:0] qual_writedata_st0;
wire [`REQ_INST_META_WIDTH-1:0] qual_inst_meta_st0;
wire qual_going_to_write_st0;
wire qual_is_snp_st0;
wire qual_snp_invalidate_st0;
wire [`WORD_WIDTH-1:0] writeword_st0;
wire [`BANK_LINE_WIDTH-1:0] writedata_st0;
wire [`REQ_INST_META_WIDTH-1:0] inst_meta_st0;
wire is_snp_st0;
wire snp_invalidate_st0;
wire msrq_pending_hazard_unqual_st0;
//signals to be *used* in the next stage
wire valid_st1;
wire [`LINE_ADDR_WIDTH-1:0] addr_st1;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] wsel_st1;
@@ -299,64 +284,56 @@ module VX_bank #(
wire [`BANK_LINE_WIDTH-1:0] writedata_st1;
wire is_snp_st1;
wire snp_invalidate_st1;
wire is_mrvq_st1;
wire is_msrq_st1;
wire msrq_pending_hazard_st1;
//Determine which req will progress to the next stage
assign qual_is_fill_st0 = dfpq_pop_unqual; //dram is filling a request
assign is_msrq_st0 = msrq_pop_unqual;
assign qual_valid_st0 = dfpq_pop || mrvq_pop || reqq_pop || snrq_pop; //valid if something is being popped
assign is_fill_st0 = dfpq_pop_unqual;
//Decides which request to deal with. Priority: 1) Miss reserve 2) DRAM fill 3) Core req 4) Snp req
assign qual_addr_st0 = mrvq_pop_unqual ? mrvq_addr_st0 :
assign valid_st0 = dfpq_pop || msrq_pop || reqq_pop || snrq_pop;
assign addr_st0 = msrq_pop_unqual ? msrq_addr_st0 :
dfpq_pop_unqual ? dfpq_addr_st0 :
reqq_pop_unqual ? reqq_req_addr_st0[`LINE_SELECT_ADDR_RNG] :
snrq_pop_unqual ? snrq_addr_st0 :
0;
//Word select does ? Does this just pick a specific word from the line instead of the whole line?
if (`WORD_SELECT_WIDTH != 0) begin
assign qual_wsel_st0 = reqq_pop_unqual ? reqq_req_addr_st0[`WORD_SELECT_WIDTH-1:0] :
mrvq_pop_unqual ? mrvq_wsel_st0 :
assign wsel_st0 = reqq_pop_unqual ? reqq_req_addr_st0[`WORD_SELECT_WIDTH-1:0] :
msrq_pop_unqual ? msrq_wsel_st0 :
0;
end else begin
`UNUSED_VAR(mrvq_wsel_st0)
assign qual_wsel_st0 = 0;
`UNUSED_VAR(msrq_wsel_st0)
assign wsel_st0 = 0;
end
//if you are filling from dram then that is the write data? What about core? What is 57?
assign qual_writedata_st0 = dfpq_pop_unqual ? dfpq_filldata_st0 : 57;
assign writedata_st0 = dfpq_filldata_st0;
//note that this is stored even if a DRAM fill is processed
assign qual_inst_meta_st0 = mrvq_pop_unqual ? {`REQ_TAG_WIDTH'(mrvq_tag_st0) , mrvq_rw_st0, mrvq_byteen_st0, mrvq_tid_st0} :
assign inst_meta_st0 = msrq_pop_unqual ? {`REQ_TAG_WIDTH'(msrq_tag_st0) , msrq_rw_st0, msrq_byteen_st0, msrq_tid_st0} :
reqq_pop_unqual ? {`REQ_TAG_WIDTH'(reqq_req_tag_st0), reqq_req_rw_st0, reqq_req_byteen_st0, reqq_req_tid_st0} :
snrq_pop_unqual ? {`REQ_TAG_WIDTH'(snrq_tag_st0), 1'b0, WORD_SIZE'(0), `REQS_BITS'(0)} :
0;
assign qual_going_to_write_st0 = dfpq_pop_unqual ? 1 :
(mrvq_pop_unqual && mrvq_rw_st0) ? 1 :
(reqq_pop_unqual && reqq_req_rw_st0) ? 1 :
0;
//snp signals check to see if the miss reserve as a snp in it first.
assign qual_is_snp_st0 = mrvq_pop_unqual ? mrvq_is_snp_st0 :
assign is_snp_st0 = msrq_pop_unqual ? msrq_is_snp_st0 :
snrq_pop_unqual ? 1 :
0;
//if we are popping from the miss reserve then assign to the mrvq invalidate. If not and popping from the snoop queue use the snoop invalidate. Else this is 0
assign qual_snp_invalidate_st0 = mrvq_pop_unqual ? mrvq_snp_invalidate_st0 :
assign snp_invalidate_st0 = msrq_pop_unqual ? msrq_snp_invalidate_st0 :
snrq_pop_unqual ? snrq_invalidate_st0 :
0;
//choose which word of the lien is being written to
assign qual_writeword_st0 = mrvq_pop_unqual ? mrvq_writeword_st0 :
assign writeword_st0 = msrq_pop_unqual ? msrq_writeword_st0 :
reqq_pop_unqual ? reqq_req_writeword_st0 :
0;
assign qual_is_mrvq_st0 = mrvq_pop_unqual;
// we have a miss in msrq or going into it for the current address
wire msrq_pending_hazard_st0 = msrq_pending_hazard_unqual_st0
|| (miss_add_unqual && (addr_st2 == addr_st0));
`ifdef DBG_CORE_REQ_INFO
if (WORD_SIZE != `GLOBAL_BLOCK_SIZE) begin
assign {debug_pc_st0, debug_rd_st0, debug_wid_st0, debug_tagid_st0, debug_rw_st0, debug_byteen_st0, debug_tid_st0} = qual_inst_meta_st0;
assign {debug_pc_st0, debug_rd_st0, debug_wid_st0, debug_tagid_st0, debug_rw_st0, debug_byteen_st0, debug_tid_st0} = inst_meta_st0;
end
`endif
@@ -367,10 +344,16 @@ module VX_bank #(
.reset (reset),
.stall (stall_bank_pipe),
.flush (1'b0),
.in ({qual_is_mrvq_st0, qual_is_snp_st0, qual_snp_invalidate_st0, qual_going_to_write_st0, qual_valid_st0, qual_addr_st0, qual_wsel_st0, qual_writeword_st0, qual_inst_meta_st0, qual_is_fill_st0, qual_writedata_st0}),
.out ({is_mrvq_st1 , is_snp_st1, snp_invalidate_st1, going_to_write_st1, valid_st1, addr_st1, wsel_st1, writeword_st1, inst_meta_st1, is_fill_st1, writedata_st1})
.in ({is_msrq_st0, is_snp_st0, snp_invalidate_st0, msrq_pending_hazard_st0, valid_st0, addr_st0, wsel_st0, writeword_st0, inst_meta_st0, is_fill_st0, writedata_st0}),
.out ({is_msrq_st1, is_snp_st1, snp_invalidate_st1, msrq_pending_hazard_st1, valid_st1, addr_st1, wsel_st1, writeword_st1, inst_meta_st1, is_fill_st1, writedata_st1})
);
`ifdef DBG_CORE_REQ_INFO
if (WORD_SIZE != `GLOBAL_BLOCK_SIZE) begin
assign {debug_pc_st1, debug_rd_st1, debug_wid_st1, debug_tagid_st1, debug_rw_st1, debug_byteen_st1, debug_tid_st1} = inst_meta_st1;
end
`endif
wire[`WORD_WIDTH-1:0] readword_st1;
wire[`BANK_LINE_WIDTH-1:0] readdata_st1;
wire[`TAG_SELECT_BITS-1:0] readtag_st1;
@@ -383,24 +366,20 @@ module VX_bank #(
`DEBUG_END
wire mem_rw_st1;
wire [WORD_SIZE-1:0] mem_byteen_st1;
wire fill_saw_dirty_st1;
wire snp_to_mrvq_st1;
wire mrvq_init_ready_state_st1;
wire miss_add_because_miss;
wire mrvq_recover_ready_state_st1;
wire miss_add_unqual;
assign {tag_st1, mem_rw_st1, mem_byteen_st1, tid_st1} = inst_meta_st1;
assign st2_pending_hazard_st1 = (miss_add_because_miss)
&& ((addr_st2 == addr_st1) && !is_fill_st2);
// we have a miss in st2 for the current address
wire st2_pending_hazard_st1 = miss_add_unqual && (addr_st2 == addr_st1);
assign force_request_miss_st1 = (valid_st1 && !is_mrvq_st1 && (mrvq_pending_hazard_st1 || st2_pending_hazard_st1))
|| (valid_st1 && is_mrvq_st1 && recover_mrvq_state_st2);
// force miss to ensure commit order when a new request has pending previous requests to same block
// also force a miss for msrq requests when previous request in st2 got a miss
wire force_miss_st1 = (valid_st1 && !is_msrq_st1 && ~is_fill_st1 && (msrq_pending_hazard_st1 || st2_pending_hazard_st1))
|| (valid_st1 && is_msrq_st1 && is_msrq_miss_st2);
assign mrvq_recover_ready_state_st1 = valid_st1
&& is_mrvq_st1
&& recover_mrvq_state_st2
&& (addr_st2 == addr_st1);
// access the tag data store
wire tag_data_fire = valid_st1 && !stall_bank_pipe;
VX_tag_data_access #(
.BANK_ID (BANK_ID),
@@ -423,27 +402,19 @@ module VX_bank #(
.debug_tagid_st1(debug_tagid_st1),
`endif
.stall (stall_bank_pipe),
.stall_bank_pipe(stall_bank_pipe),
.force_request_miss_st1(force_request_miss_st1),
// Initial Read
.readaddr_st1(addr_st1[`LINE_SELECT_BITS-1:0]),
// Actual Read/Write
.valid_req_st1 (valid_st1),
.valid_req_st1 (tag_data_fire),
.writefill_st1 (is_fill_st1),
.writeaddr_st1 (addr_st1),
.addr_st1 (addr_st1),
.wordsel_st1 (wsel_st1),
.writeword_st1 (writeword_st1),
.writedata_st1 (writedata_st1),
.mem_rw_st1 (mem_rw_st1),
.mem_byteen_st1 (mem_byteen_st1),
.is_snp_st1 (is_snp_st1),
.snp_invalidate_st1(snp_invalidate_st1),
.force_miss_st1 (force_miss_st1),
// Read Data
.readword_st1 (readword_st1),
@@ -451,23 +422,9 @@ module VX_bank #(
.readtag_st1 (readtag_st1),
.miss_st1 (miss_st1),
.dirty_st1 (dirty_st1),
.dirtyb_st1 (dirtyb_st1),
.fill_saw_dirty_st1(fill_saw_dirty_st1),
.snp_to_mrvq_st1(snp_to_mrvq_st1),
.mrvq_init_ready_state_st1(mrvq_init_ready_state_st1)
.dirtyb_st1 (dirtyb_st1)
);
`ifdef DBG_CORE_REQ_INFO
if (WORD_SIZE != `GLOBAL_BLOCK_SIZE) begin
assign {debug_pc_st1, debug_rd_st1, debug_wid_st1, debug_tagid_st1, debug_rw_st1, debug_byteen_st1, debug_tid_st1} = inst_meta_st1;
end else begin
assign {debug_pc_st1, debug_rd_st1, debug_wid_st1, debug_tagid_st1, debug_rw_st1, debug_byteen_st1, debug_tid_st1} = 0;
end
`endif
wire qual_valid_st1_2 = valid_st1 && !is_fill_st1;
wire is_mrvq_st1_st2 = is_mrvq_st1;
wire valid_st2;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] wsel_st2;
wire [`WORD_WIDTH-1:0] writeword_st2;
@@ -478,26 +435,21 @@ module VX_bank #(
wire [BANK_LINE_SIZE-1:0] dirtyb_st2;
wire [`REQ_INST_META_WIDTH-1:0] inst_meta_st2;
wire [`TAG_SELECT_BITS-1:0] readtag_st2;
wire fill_saw_dirty_st2;
wire is_fill_st2;
wire is_snp_st2;
wire snp_invalidate_st2;
wire snp_to_mrvq_st2;
wire is_mrvq_st2;
wire mrvq_init_ready_state_st2;
wire mrvq_recover_ready_state_st2;
wire mrvq_init_ready_state_unqual_st2;
wire mrvq_init_ready_state_hazard_st0_st1;
wire mrvq_init_ready_state_hazard_st1_st1;
wire force_miss_st2;
wire is_msrq_st2;
VX_generic_register #(
.N(1+ 1+ 1 + 1 + 1 + 1 + 1 + 1 + 1 + `LINE_ADDR_WIDTH + `UP(`WORD_SELECT_WIDTH) + `WORD_WIDTH + `WORD_WIDTH + `BANK_LINE_WIDTH + `TAG_SELECT_BITS + 1 + 1 + BANK_LINE_SIZE + `REQ_INST_META_WIDTH)
.N(1+ 1+ 1 + 1 + 1 + 1 + `LINE_ADDR_WIDTH + `UP(`WORD_SELECT_WIDTH) + `WORD_WIDTH + `WORD_WIDTH + `BANK_LINE_WIDTH + `TAG_SELECT_BITS + 1 + 1 + BANK_LINE_SIZE + `REQ_INST_META_WIDTH)
) pipe_reg1 (
.clk (clk),
.reset (reset),
.stall (stall_bank_pipe),
.flush (1'b0),
.in ({mrvq_recover_ready_state_st1, is_mrvq_st1_st2, mrvq_init_ready_state_st1, snp_to_mrvq_st1, is_snp_st1, snp_invalidate_st1, fill_saw_dirty_st1, is_fill_st1, qual_valid_st1_2, addr_st1, wsel_st1, writeword_st1, readword_st1, readdata_st1, readtag_st1, miss_st1, dirty_st1, dirtyb_st1, inst_meta_st1}),
.out ({mrvq_recover_ready_state_st2 , is_mrvq_st2 , mrvq_init_ready_state_unqual_st2, snp_to_mrvq_st2 , is_snp_st2 , snp_invalidate_st2, fill_saw_dirty_st2 , is_fill_st2 , valid_st2 , addr_st2, wsel_st2, writeword_st2, readword_st2, readdata_st2, readtag_st2, miss_st2, dirty_st2, dirtyb_st2, inst_meta_st2})
.in ({is_msrq_st1, force_miss_st1, is_snp_st1, snp_invalidate_st1, is_fill_st1, valid_st1, addr_st1, wsel_st1, writeword_st1, readword_st1, readdata_st1, readtag_st1, miss_st1, dirty_st1, dirtyb_st1, inst_meta_st1}),
.out ({is_msrq_st2, force_miss_st2, is_snp_st2, snp_invalidate_st2, is_fill_st2, valid_st2, addr_st2, wsel_st2, writeword_st2, readword_st2, readdata_st2, readtag_st2, miss_st2, dirty_st2, dirtyb_st2, inst_meta_st2})
);
`ifdef DBG_CORE_REQ_INFO
@@ -507,19 +459,17 @@ module VX_bank #(
`endif
// Enqueue to miss reserv if it's a valid miss
assign miss_add_because_miss = valid_st2 && !is_snp_st2 && miss_st2;
wire miss_add_because_pending = snp_to_mrvq_st2;
wire miss_add_unqual = (miss_add_because_miss || miss_add_because_pending);
assign mrvq_push_stall = miss_add_unqual && mrvq_full;
assign miss_add_unqual = miss_st2 || force_miss_st2;
assign msrq_push_stall = miss_add_unqual && msrq_full;
wire miss_add = miss_add_unqual
&& !mrvq_full
&& !(cwbq_push_stall
|| dwbq_push_stall
|| dram_fill_req_stall);
&& !msrq_full
&& !cwbq_push_stall
&& !dwbq_push_stall;
assign recover_mrvq_state_st2 = miss_add_unqual && is_mrvq_st2; // Doesn't need to include the stalls
// we have a recurrent msrq miss
assign is_msrq_miss_st2 = miss_add_unqual && is_msrq_st2;
wire [`LINE_ADDR_WIDTH-1:0] miss_add_addr = addr_st2;
wire [`UP(`WORD_SELECT_WIDTH)-1:0] miss_add_wsel = wsel_st2;
@@ -528,18 +478,31 @@ module VX_bank #(
wire miss_add_is_snp = is_snp_st2;
wire miss_add_snp_invalidate = snp_invalidate_st2;
wire miss_add_is_mrvq = valid_st2 && is_mrvq_st2 && !stall_bank_pipe;
wire msrq_real_pop_st2 = valid_st2 && is_msrq_st2 && !miss_add_unqual && !stall_bank_pipe;
assign mrvq_init_ready_state_hazard_st0_st1 = miss_add_unqual && qual_is_fill_st0 && (miss_add_addr == dfpq_addr_st0); // Doesn't need to be muxed to qual, only care about fills
assign mrvq_init_ready_state_hazard_st1_st1 = miss_add_unqual && is_fill_st1 && (miss_add_addr == addr_st1);
// mark msrq entry that match DRAM fill as 'ready'
wire update_ready_st0 = dfpq_pop;
assign mrvq_init_ready_state_st2 = mrvq_init_ready_state_unqual_st2 // When req was in st1e, either matched with an mrvq entery OR mrvq recovering state
|| mrvq_init_ready_state_hazard_st0_st1 // If there's a fill in st0 that has the same address as miss_add_addr
|| mrvq_init_ready_state_hazard_st1_st1; // If there's a fill in st1 that has the same address as miss_add_addr
// push missed requests as 'ready'
// if it didn't actually missed but had to abort because of pending requets in msrq
// if matching fill request to the block is in stage 0
// if matching fill request to the block is in stage 1
wire match_st0_fill_st2 = is_fill_st0 && (miss_add_addr == addr_st0);
wire match_st1_fill_st2 = is_fill_st1 && (miss_add_addr == addr_st1);
wire msrq_init_ready_state_st2 = !miss_st2
|| match_st0_fill_st2
|| match_st1_fill_st2;
always @(*) begin
if (miss_st2 && (match_st0_fill_st2 || match_st1_fill_st2)) begin
$display("%t: incoming fill - addr=%0h, st0=%b, st1=%b", $time, `LINE_TO_BYTE_ADDR(miss_add_addr, BANK_ID), match_st0_fill_st2, match_st1_fill_st2);
end
end
VX_cache_miss_resrv #(
.BANK_ID (BANK_ID),
.CACHE_ID (CACHE_ID),
.CORE_TAG_ID_BITS (CORE_TAG_ID_BITS),
.BANK_LINE_SIZE (BANK_LINE_SIZE),
.NUM_BANKS (NUM_BANKS),
.WORD_SIZE (WORD_SIZE),
@@ -551,9 +514,19 @@ module VX_bank #(
.clk (clk),
.reset (reset),
// Enqueue
`ifdef DBG_CORE_REQ_INFO
.debug_pc_st0 (debug_pc_st0),
.debug_rd_st0 (debug_rd_st0),
.debug_wid_st0 (debug_wid_st0),
.debug_tagid_st0(debug_tagid_st0),
.debug_pc_st2 (debug_pc_st2),
.debug_rd_st2 (debug_rd_st2),
.debug_wid_st2 (debug_wid_st2),
.debug_tagid_st2(debug_tagid_st2),
`endif
// enqueue
.miss_add (miss_add),
.is_mrvq (miss_add_is_mrvq),
.miss_add_addr (miss_add_addr),
.miss_add_wsel (miss_add_wsel),
.miss_add_data (miss_add_data),
@@ -563,27 +536,30 @@ module VX_bank #(
.miss_add_byteen (miss_add_byteen),
.miss_add_is_snp (miss_add_is_snp),
.miss_add_snp_invalidate (miss_add_snp_invalidate),
.miss_resrv_full (mrvq_full),
.miss_resrv_stop (mrvq_stop),
.mrvq_init_ready_state (mrvq_init_ready_state_st2),
.is_msrq_st2 (is_msrq_st2),
.init_ready_state_st2 (msrq_init_ready_state_st2),
// Broadcast
.is_fill_st1 (is_fill_st1),
.fill_addr_st1 (addr_st1),
.pending_hazard_st1 (mrvq_pending_hazard_st1),
.miss_resrv_full (msrq_full),
.miss_resrv_almfull (msrq_almfull),
// Dequeue
.miss_resrv_pop (mrvq_pop),
.miss_resrv_valid_st0 (mrvq_valid_st0),
.miss_resrv_addr_st0 (mrvq_addr_st0),
.miss_resrv_wsel_st0 (mrvq_wsel_st0),
.miss_resrv_data_st0 (mrvq_writeword_st0),
.miss_resrv_tid_st0 (mrvq_tid_st0),
.miss_resrv_tag_st0 (mrvq_tag_st0),
.miss_resrv_rw_st0 (mrvq_rw_st0),
.miss_resrv_byteen_st0 (mrvq_byteen_st0),
.miss_resrv_is_snp_st0 (mrvq_is_snp_st0),
.miss_resrv_snp_invalidate_st0 (mrvq_snp_invalidate_st0)
// fill
.update_ready_st0 (update_ready_st0),
.fill_addr_st0 (addr_st0),
.pending_hazard_st0 (msrq_pending_hazard_unqual_st0),
// dequeue
.miss_resrv_schedule_st0 (msrq_pop),
.miss_resrv_valid_st0 (msrq_valid_st0),
.miss_resrv_addr_st0 (msrq_addr_st0),
.miss_resrv_wsel_st0 (msrq_wsel_st0),
.miss_resrv_data_st0 (msrq_writeword_st0),
.miss_resrv_tid_st0 (msrq_tid_st0),
.miss_resrv_tag_st0 (msrq_tag_st0),
.miss_resrv_rw_st0 (msrq_rw_st0),
.miss_resrv_byteen_st0 (msrq_byteen_st0),
.miss_resrv_is_snp_st0 (msrq_is_snp_st0),
.miss_resrv_snp_invalidate_st0 (msrq_snp_invalidate_st0),
.miss_resrv_pop_st2 (msrq_real_pop_st2)
);
// Enqueue core response
@@ -591,15 +567,13 @@ module VX_bank #(
wire cwbq_push, cwbq_pop;
wire cwbq_empty, cwbq_full;
wire cwbq_push_unqual = valid_st2 && !miss_st2 && !is_fill_st2 && !is_snp_st2;
wire cwbq_push_unqual = valid_st2 && !is_fill_st2 && !is_snp_st2 && !miss_st2 && !force_miss_st2 && !miss_add_rw;
assign cwbq_push_stall = cwbq_push_unqual && cwbq_full;
assign cwbq_push = cwbq_push_unqual
&& !cwbq_full
&& (miss_add_rw == 0)
&& !(dwbq_push_stall
|| mrvq_push_stall
|| dram_fill_req_stall);
&& !dwbq_push_stall
&& !msrq_push_stall;
assign cwbq_pop = core_rsp_valid && core_rsp_ready;
@@ -613,11 +587,9 @@ module VX_bank #(
) cwb_queue (
.clk (clk),
.reset (reset),
.push (cwbq_push),
.data_in ({cwbq_tid, cwbq_tag, cwbq_data}),
.pop (cwbq_pop),
.data_in ({cwbq_tid, cwbq_tag, cwbq_data}),
.data_out({core_rsp_tid, core_rsp_tag, core_rsp_data}),
.empty (cwbq_empty),
.full (cwbq_full),
@@ -626,90 +598,92 @@ module VX_bank #(
assign core_rsp_valid = !cwbq_empty;
// Enqueue DRAM fill request
wire dram_fill_req_fast = miss_add_unqual; // Completely unqualified hint that we might send a dram_fill_req
wire dram_fill_req_unqual = dram_fill_req_fast
&& (!mrvq_init_ready_state_st2
|| (is_mrvq_st2 && !mrvq_recover_ready_state_st2)); // If this is set, then we are sure we will be sending a dram_fill_req
assign dram_fill_req_valid = dram_fill_req_unqual
&& !(dwbq_push_stall
|| mrvq_push_stall
|| cwbq_push_stall);
assign dram_fill_req_addr = addr_st2;
assign dram_fill_req_stall = dram_fill_req_fast && !dram_fill_req_ready; // Uses dram_fill_req_fast for critical path
// Enqueue DRAM writeback request
// Enqueue DRAM / Snoop request
wire dwbq_push, dwbq_pop;
wire dwbq_empty, dwbq_full;
wire dwbq_is_dwb_in, dwbq_is_snp_in;
wire dwbq_is_dwb_out, dwbq_is_snp_out;
wire dwbq_is_dram_out, dwbq_is_snp_out;
assign dwbq_is_snp_in = is_snp_st2 && valid_st2 && !snp_to_mrvq_st2;
assign dwbq_is_dwb_in = (valid_st2 && miss_st2 && dirty_st2) || fill_saw_dirty_st2;
wire dwbq_push_unqual = dwbq_is_dwb_in || dwbq_is_snp_in;
wire [`LINE_ADDR_WIDTH-1:0] dwbq_req_addr;
wire [SNP_REQ_TAG_WIDTH-1:0] dwbq_snp_tag;
wire dwbq_is_dfl_in = miss_st2 && !msrq_init_ready_state_st2 && (!force_miss_st2 || is_msrq_st2);
wire dwbq_is_dwb_in = dirty_st2 && !force_miss_st2 && (is_fill_st2 || is_snp_st2);
wire dwbq_is_snp_in = valid_st2 && !force_miss_st2 && is_snp_st2;
wire dwbq_is_dram_in = dwbq_is_dfl_in || dwbq_is_dwb_in;
always @(posedge clk) begin
assert(!is_msrq_st2 || !is_fill_st2);
assert(!dwbq_is_dfl_in || !dwbq_is_dwb_in);
end
wire dwbq_push_unqual = dwbq_is_dram_in || dwbq_is_snp_in;
assign dwbq_push_stall = dwbq_push_unqual && dwbq_full;
assign dwbq_push = dwbq_push_unqual
&& !dwbq_full
&& !(cwbq_push_stall
|| mrvq_push_stall
|| dram_fill_req_stall);
&& !cwbq_push_stall
&& !msrq_push_stall;
wire [`LINE_ADDR_WIDTH-1:0] dwbq_req_addr = {readtag_st2, addr_st2[`LINE_SELECT_BITS-1:0]};
wire [SNP_REQ_TAG_WIDTH-1:0] snrq_tag_st2 = SNP_REQ_TAG_WIDTH'(miss_add_tag);
assign dwbq_req_addr = dwbq_is_dwb_in ? {readtag_st2, addr_st2[`LINE_SELECT_BITS-1:0]} : addr_st2;
assign dwbq_snp_tag = SNP_REQ_TAG_WIDTH'(miss_add_tag);
VX_generic_queue #(
.DATAW(1 + 1 + BANK_LINE_SIZE + `LINE_ADDR_WIDTH + `BANK_LINE_WIDTH + SNP_REQ_TAG_WIDTH),
.SIZE(DWBQ_SIZE)
.DATAW(1 + 1 + 1 + BANK_LINE_SIZE + `LINE_ADDR_WIDTH + `BANK_LINE_WIDTH + SNP_REQ_TAG_WIDTH),
.SIZE(DREQ_SIZE)
) dwb_queue (
.clk (clk),
.reset (reset),
.push (dwbq_push),
.data_in ({dwbq_is_dwb_in, dwbq_is_snp_in, dirtyb_st2, dwbq_req_addr, readdata_st2, snrq_tag_st2}),
.pop (dwbq_pop),
.data_out({dwbq_is_dwb_out, dwbq_is_snp_out, dram_wb_req_byteen, dram_wb_req_addr, dram_wb_req_data, snp_rsp_tag}),
.data_in ({dwbq_is_dram_in, dwbq_is_snp_in, dwbq_is_dwb_in, dirtyb_st2, dwbq_req_addr, readdata_st2, dwbq_snp_tag}),
.data_out({dwbq_is_dram_out, dwbq_is_snp_out, dram_req_rw, dram_req_byteen, dram_req_addr, dram_req_data, snp_rsp_tag}),
.empty (dwbq_empty),
.full (dwbq_full),
`UNUSED_PIN (size)
);
wire dram_wb_req_fire = dram_wb_req_valid && dram_wb_req_ready;
wire dram_req_fire = dram_req_valid && dram_req_ready;
wire snp_rsp_fire = snp_rsp_valid && snp_rsp_ready;
reg dwbq_dual_valid_sel;
reg dwbq_out_sel_snp;
always @(posedge clk) begin
if (reset) begin
dwbq_dual_valid_sel <= 0;
end else if (dwbq_is_dwb_out
dwbq_out_sel_snp <= 0;
end else if (dwbq_is_dram_out
&& dwbq_is_snp_out
&& (dram_wb_req_fire || snp_rsp_fire)) begin
dwbq_dual_valid_sel <= ~dwbq_dual_valid_sel;
&& (dram_req_fire || snp_rsp_fire)) begin
dwbq_out_sel_snp <= ~dwbq_out_sel_snp;
end
end
// when both dwb and snp are asserted, first release the cwb, then release the snp.
assign dram_wb_req_valid = !dwbq_empty && dwbq_is_dwb_out && (~dwbq_is_snp_out || dwbq_dual_valid_sel == 0);
assign snp_rsp_valid = !dwbq_empty && dwbq_is_snp_out && (~dwbq_is_dwb_out || dwbq_dual_valid_sel == 1);
assign dram_req_valid = !dwbq_empty && dwbq_is_dram_out && (~dwbq_is_snp_out || !dwbq_out_sel_snp);
assign snp_rsp_valid = !dwbq_empty && dwbq_is_snp_out && (~dwbq_is_dram_out || dwbq_out_sel_snp);
assign dwbq_pop = (dwbq_is_dwb_out && !dwbq_is_snp_out && dram_wb_req_fire)
|| (dwbq_is_snp_out && !dwbq_is_dwb_out && snp_rsp_fire)
|| (dwbq_is_dwb_out && dwbq_is_snp_out && snp_rsp_fire);
assign dwbq_pop = (dwbq_is_dram_out && !dwbq_is_snp_out && dram_req_fire)
|| (dwbq_is_snp_out && snp_rsp_fire);
// bank pipeline stall
assign stall_bank_pipe = cwbq_push_stall
|| dwbq_push_stall
|| mrvq_push_stall
|| dram_fill_req_stall;
assign stall_bank_pipe = (cwbq_push_stall || dwbq_push_stall || msrq_push_stall);
`SCOPE_ASSIGN (valid_st0, valid_st0);
`SCOPE_ASSIGN (valid_st1, valid_st1);
`SCOPE_ASSIGN (valid_st2, valid_st2);
`SCOPE_ASSIGN (is_msrq_st1, is_msrq_st1);
`SCOPE_ASSIGN (miss_st1, miss_st1);
`SCOPE_ASSIGN (dirty_st1, dirty_st1);
`SCOPE_ASSIGN (force_miss_st1, force_miss_st1);
`SCOPE_ASSIGN (stall_pipe, stall_bank_pipe);
`SCOPE_ASSIGN (addr_st0, `LINE_TO_BYTE_ADDR(addr_st0, BANK_ID));
`SCOPE_ASSIGN (addr_st1, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID));
`SCOPE_ASSIGN (addr_st2, `LINE_TO_BYTE_ADDR(addr_st2, BANK_ID));
`ifdef DBG_PRINT_CACHE_BANK
always @(posedge clk) begin
@@ -719,14 +693,11 @@ module VX_bank #(
if (core_rsp_valid && core_rsp_ready) begin
$display("%t: cache%0d:%0d core rsp: tag=%0h, data=%0h", $time, CACHE_ID, BANK_ID, core_rsp_tag, core_rsp_data);
end
if (dram_fill_req_valid && dram_fill_req_ready) begin
$display("%t: cache%0d:%0d dram_fill req: addr=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(dram_fill_req_addr, BANK_ID));
if (dram_req_valid && dram_req_ready) begin
$display("%t: cache%0d:%0d dram req: addr=%0h, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(dram_req_addr, BANK_ID), dram_req_data);
end
if (dram_wb_req_valid && dram_wb_req_ready) begin
$display("%t: cache%0d:%0d dram_wb req: addr=%0h, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(dram_wb_req_addr, BANK_ID), dram_wb_req_data);
end
if (dram_fill_rsp_valid && dram_fill_rsp_ready) begin
$display("%t: cache%0d:%0d dram_fill rsp: addr=%0h, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(dram_fill_rsp_addr, BANK_ID), dram_fill_rsp_data);
if (dram_rsp_valid && dram_rsp_ready) begin
$display("%t: cache%0d:%0d dram rsp: addr=%0h, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(dram_rsp_addr, BANK_ID), dram_rsp_data);
end
if (snp_req_valid && snp_req_ready) begin
$display("%t: cache%0d:%0d snp req: addr=%0h, invalidate=%0d, tag=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(snp_req_addr, BANK_ID), snp_req_invalidate, snp_req_tag);
@@ -734,21 +705,25 @@ module VX_bank #(
if (snp_rsp_valid && snp_rsp_ready) begin
$display("%t: cache%0d:%0d snp rsp: tag=%0h", $time, CACHE_ID, BANK_ID, snp_rsp_tag);
end
if (msrq_pop) begin
$display("%t: cache%0d:%0d msrq_pop: addr=%0h wid=%0d, PC=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st0, BANK_ID), debug_wid_st0, debug_pc_st0);
end
if (dfpq_pop) begin
$display("%t: cache%0d:%0d dfpq_pop: addr=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st0, BANK_ID));
end
if (reqq_pop) begin
$display("%t: cache%0d:%0d reqq_pop: addr=%0h wid=%0d, PC=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st0, BANK_ID), debug_wid_st0, debug_pc_st0);
end
if (snrq_pop) begin
$display("%t: cache%0d:%0d snrq_pop: addr=%0h wid=%0d, PC=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st0, BANK_ID), debug_wid_st0, debug_pc_st0);
end
if (cwbq_push) begin
$display("%t: cache%0d:%0d cwbq_push: addr=%0h wid=%0d, PC=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st2, BANK_ID), debug_wid_st2, debug_pc_st2);
end
if (dwbq_push) begin
$display("%t: cache%0d:%0d dwbq_push: addr=%0h wid=%0d, PC=%0h, fill=%b, wb=%b, snp=%b", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st2, BANK_ID), debug_wid_st2, debug_pc_st2, dwbq_is_dfl_in, dwbq_is_dwb_in, dwbq_is_snp_in);
end
end
`endif
`SCOPE_ASSIGN (valid_st0, qual_valid_st0);
`SCOPE_ASSIGN (valid_st1, valid_st1);
`SCOPE_ASSIGN (valid_st2, valid_st2);
`SCOPE_ASSIGN (is_mrvq_st1, is_mrvq_st1);
`SCOPE_ASSIGN (miss_st1, miss_st1);
`SCOPE_ASSIGN (dirty_st1, dirty_st1);
`SCOPE_ASSIGN (force_miss_st1, force_request_miss_st1);
`SCOPE_ASSIGN (stall_pipe, stall_bank_pipe);
`SCOPE_ASSIGN (addr_st0, `LINE_TO_BYTE_ADDR(qual_addr_st0, BANK_ID));
`SCOPE_ASSIGN (addr_st1, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID));
`SCOPE_ASSIGN (addr_st2, `LINE_TO_BYTE_ADDR(addr_st2, BANK_ID));
endmodule

2
hw/rtl/cache/VX_bank_core_req_arb.v vendored
View File

@@ -26,7 +26,6 @@ module VX_bank_core_req_arb #(
// Dequeue Data
input wire reqq_pop,
output wire reqq_req_st0,
output wire [`REQS_BITS-1:0] reqq_req_tid_st0,
output wire reqq_req_rw_st0,
output wire [WORD_SIZE-1:0] reqq_req_byteen_st0,
@@ -107,7 +106,6 @@ module VX_bank_core_req_arb #(
);
assign reqq_empty = !qual_has_request;
assign reqq_req_st0 = qual_has_request;
assign reqq_req_tid_st0 = qual_request_index;
assign reqq_req_byteen_st0 = qual_byteen[qual_request_index];
assign reqq_req_addr_st0 = qual_addr[qual_request_index];

171
hw/rtl/cache/VX_cache.v vendored
View File

@@ -19,18 +19,16 @@ module VX_cache #(
parameter CREQ_SIZE = 8,
// Miss Reserv Queue Knob
parameter MRVQ_SIZE = 16,
// Dram Fill Rsp Queue Size
parameter DFPQ_SIZE = 16,
// DRAM Response Queue Size
parameter DRPQ_SIZE = 16,
// Snoop Req Queue Size
parameter SNRQ_SIZE = 16,
// Queues for writebacks Knobs {1, 2, 4, 8, ...}
// Core Writeback Queue Size
parameter CWBQ_SIZE = 8,
// Dram Writeback Queue Size
parameter DWBQ_SIZE = 4,
// Dram Fill Req Queue Size
parameter DFQQ_SIZE = 8,
// DRAM Request Queue Size
parameter DREQ_SIZE = 4,
// Enable cache writeable
parameter WRITE_ENABLE = 1,
@@ -144,17 +142,14 @@ module VX_cache #(
wire [NUM_BANKS-1:0][CORE_TAG_WIDTH-1:0] per_bank_core_rsp_tag;
wire [NUM_BANKS-1:0] per_bank_core_rsp_ready;
wire [NUM_BANKS-1:0] per_bank_dram_fill_req_valid;
wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_fill_req_addr;
wire dram_fill_req_ready;
wire [NUM_BANKS-1:0] per_bank_dram_req_valid;
wire [NUM_BANKS-1:0] per_bank_dram_req_rw;
wire [NUM_BANKS-1:0][BANK_LINE_SIZE-1:0] per_bank_dram_req_byteen;
wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_req_addr;
wire [NUM_BANKS-1:0][`BANK_LINE_WIDTH-1:0] per_bank_dram_req_data;
wire [NUM_BANKS-1:0] per_bank_dram_req_ready;
wire [NUM_BANKS-1:0] per_bank_dram_fill_rsp_ready;
wire [NUM_BANKS-1:0] per_bank_dram_wb_req_ready;
wire [NUM_BANKS-1:0] per_bank_dram_wb_req_valid;
wire [NUM_BANKS-1:0][BANK_LINE_SIZE-1:0] per_bank_dram_wb_req_byteen;
wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_wb_req_addr;
wire [NUM_BANKS-1:0][`BANK_LINE_WIDTH-1:0] per_bank_dram_wb_req_data;
wire [NUM_BANKS-1:0] per_bank_dram_rsp_ready;
wire [NUM_BANKS-1:0] per_bank_snp_req_ready;
@@ -236,7 +231,7 @@ module VX_cache #(
);
assign dram_req_tag = dram_req_addr;
assign dram_rsp_ready = (| per_bank_dram_fill_rsp_ready);
assign dram_rsp_ready = (| per_bank_dram_rsp_ready);
for (genvar i = 0; i < NUM_BANKS; i++) begin
wire [NUM_REQUESTS-1:0] curr_bank_core_req_valid;
@@ -245,6 +240,7 @@ module VX_cache #(
wire [NUM_REQUESTS-1:0][`WORD_ADDR_WIDTH-1:0] curr_bank_core_req_addr;
wire [`CORE_REQ_TAG_COUNT-1:0][CORE_TAG_WIDTH-1:0] curr_bank_core_req_tag;
wire [NUM_REQUESTS-1:0][`WORD_WIDTH-1:0] curr_bank_core_req_data;
wire curr_bank_core_req_ready;
wire curr_bank_core_rsp_valid;
wire [`REQS_BITS-1:0] curr_bank_core_rsp_tid;
@@ -252,20 +248,17 @@ module VX_cache #(
wire [CORE_TAG_WIDTH-1:0] curr_bank_core_rsp_tag;
wire curr_bank_core_rsp_ready;
wire curr_bank_dram_fill_rsp_valid;
wire [`BANK_LINE_WIDTH-1:0] curr_bank_dram_fill_rsp_data;
wire [`LINE_ADDR_WIDTH-1:0] curr_bank_dram_fill_rsp_addr;
wire curr_bank_dram_fill_rsp_ready;
wire curr_bank_dram_req_valid;
wire curr_bank_dram_req_rw;
wire [BANK_LINE_SIZE-1:0] curr_bank_dram_req_byteen;
wire [`LINE_ADDR_WIDTH-1:0] curr_bank_dram_req_addr;
wire[`BANK_LINE_WIDTH-1:0] curr_bank_dram_req_data;
wire curr_bank_dram_req_ready;
wire curr_bank_dram_fill_req_valid;
wire [`LINE_ADDR_WIDTH-1:0] curr_bank_dram_fill_req_addr;
wire curr_bank_dram_fill_req_ready;
wire curr_bank_dram_wb_req_valid;
wire [BANK_LINE_SIZE-1:0] curr_bank_dram_wb_req_byteen;
wire [`LINE_ADDR_WIDTH-1:0] curr_bank_dram_wb_req_addr;
wire[`BANK_LINE_WIDTH-1:0] curr_bank_dram_wb_req_data;
wire curr_bank_dram_wb_req_ready;
wire curr_bank_dram_rsp_valid;
wire [`BANK_LINE_WIDTH-1:0] curr_bank_dram_rsp_data;
wire [`LINE_ADDR_WIDTH-1:0] curr_bank_dram_rsp_addr;
wire curr_bank_dram_rsp_ready;
wire curr_bank_snp_req_valid;
wire [`LINE_ADDR_WIDTH-1:0] curr_bank_snp_req_addr;
@@ -277,8 +270,6 @@ module VX_cache #(
wire [SNP_REQ_TAG_WIDTH-1:0] curr_bank_snp_rsp_tag;
wire curr_bank_snp_rsp_ready;
wire curr_bank_core_req_ready;
// Core Req
assign curr_bank_core_req_valid = (per_bank_valid[i] & {NUM_REQUESTS{core_req_ready}});
assign curr_bank_core_req_addr = core_req_addr;
@@ -295,36 +286,28 @@ module VX_cache #(
assign per_bank_core_rsp_tag [i] = curr_bank_core_rsp_tag;
assign per_bank_core_rsp_data [i] = curr_bank_core_rsp_data;
// Dram fill request
assign per_bank_dram_fill_req_valid[i] = curr_bank_dram_fill_req_valid;
// DRAM request
assign per_bank_dram_req_valid[i] = curr_bank_dram_req_valid;
assign per_bank_dram_req_rw[i] = curr_bank_dram_req_rw;
assign per_bank_dram_req_byteen[i] = curr_bank_dram_req_byteen;
if (NUM_BANKS == 1) begin
assign per_bank_dram_fill_req_addr[i] = curr_bank_dram_fill_req_addr;
assign per_bank_dram_req_addr[i] = curr_bank_dram_req_addr;
end else begin
assign per_bank_dram_fill_req_addr[i] = `LINE_TO_DRAM_ADDR(curr_bank_dram_fill_req_addr, i);
assign per_bank_dram_req_addr[i] = `LINE_TO_DRAM_ADDR(curr_bank_dram_req_addr, i);
end
assign curr_bank_dram_fill_req_ready = dram_fill_req_ready;
assign per_bank_dram_req_data[i] = curr_bank_dram_req_data;
assign curr_bank_dram_req_ready = per_bank_dram_req_ready[i];
// Dram fill response
// DRAM response
if (NUM_BANKS == 1) begin
assign curr_bank_dram_fill_rsp_valid = dram_rsp_valid;
assign curr_bank_dram_fill_rsp_addr = dram_rsp_tag;
assign curr_bank_dram_rsp_valid = dram_rsp_valid;
assign curr_bank_dram_rsp_addr = dram_rsp_tag;
end else begin
assign curr_bank_dram_fill_rsp_valid = dram_rsp_valid && (`DRAM_ADDR_BANK(dram_rsp_tag) == i);
assign curr_bank_dram_fill_rsp_addr = `DRAM_TO_LINE_ADDR(dram_rsp_tag);
assign curr_bank_dram_rsp_valid = dram_rsp_valid && (`DRAM_ADDR_BANK(dram_rsp_tag) == i);
assign curr_bank_dram_rsp_addr = `DRAM_TO_LINE_ADDR(dram_rsp_tag);
end
assign curr_bank_dram_fill_rsp_data = dram_rsp_data;
assign per_bank_dram_fill_rsp_ready[i] = curr_bank_dram_fill_rsp_ready;
// Dram writeback request
assign per_bank_dram_wb_req_valid[i] = curr_bank_dram_wb_req_valid;
assign per_bank_dram_wb_req_byteen[i] = curr_bank_dram_wb_req_byteen;
if (NUM_BANKS == 1) begin
assign per_bank_dram_wb_req_addr[i] = curr_bank_dram_wb_req_addr;
end else begin
assign per_bank_dram_wb_req_addr[i] = `LINE_TO_DRAM_ADDR(curr_bank_dram_wb_req_addr, i);
end
assign per_bank_dram_wb_req_data[i] = curr_bank_dram_wb_req_data;
assign curr_bank_dram_wb_req_ready = per_bank_dram_wb_req_ready[i];
assign curr_bank_dram_rsp_data = dram_rsp_data;
assign per_bank_dram_rsp_ready[i] = curr_bank_dram_rsp_ready;
// Snoop request
if (NUM_BANKS == 1) begin
@@ -353,11 +336,10 @@ module VX_cache #(
.NUM_REQUESTS (NUM_REQUESTS),
.CREQ_SIZE (CREQ_SIZE),
.MRVQ_SIZE (MRVQ_SIZE),
.DFPQ_SIZE (DFPQ_SIZE),
.DRPQ_SIZE (DRPQ_SIZE),
.SNRQ_SIZE (SNRQ_SIZE),
.CWBQ_SIZE (CWBQ_SIZE),
.DWBQ_SIZE (DWBQ_SIZE),
.DFQQ_SIZE (DFQQ_SIZE),
.DREQ_SIZE (DREQ_SIZE),
.DRAM_ENABLE (DRAM_ENABLE),
.WRITE_ENABLE (WRITE_ENABLE),
.SNOOP_FORWARDING (SNOOP_FORWARDING),
@@ -385,23 +367,19 @@ module VX_cache #(
.core_rsp_tag (curr_bank_core_rsp_tag),
.core_rsp_ready (curr_bank_core_rsp_ready),
// Dram fill request
.dram_fill_req_valid (curr_bank_dram_fill_req_valid),
.dram_fill_req_addr (curr_bank_dram_fill_req_addr),
.dram_fill_req_ready (curr_bank_dram_fill_req_ready),
// DRAM request
.dram_req_valid (curr_bank_dram_req_valid),
.dram_req_rw (curr_bank_dram_req_rw),
.dram_req_byteen (curr_bank_dram_req_byteen),
.dram_req_addr (curr_bank_dram_req_addr),
.dram_req_data (curr_bank_dram_req_data),
.dram_req_ready (curr_bank_dram_req_ready),
// Dram fill response
.dram_fill_rsp_valid (curr_bank_dram_fill_rsp_valid),
.dram_fill_rsp_data (curr_bank_dram_fill_rsp_data),
.dram_fill_rsp_addr (curr_bank_dram_fill_rsp_addr),
.dram_fill_rsp_ready (curr_bank_dram_fill_rsp_ready),
// Dram writeback request
.dram_wb_req_valid (curr_bank_dram_wb_req_valid),
.dram_wb_req_byteen (curr_bank_dram_wb_req_byteen),
.dram_wb_req_addr (curr_bank_dram_wb_req_addr),
.dram_wb_req_data (curr_bank_dram_wb_req_data),
.dram_wb_req_ready (curr_bank_dram_wb_req_ready),
// DRAM response
.dram_rsp_valid (curr_bank_dram_rsp_valid),
.dram_rsp_data (curr_bank_dram_rsp_data),
.dram_rsp_addr (curr_bank_dram_rsp_addr),
.dram_rsp_ready (curr_bank_dram_rsp_ready),
// Snoop request
.snp_req_valid (curr_bank_snp_req_valid),
@@ -417,30 +395,6 @@ module VX_cache #(
);
end
VX_cache_dram_req_arb #(
.BANK_LINE_SIZE (BANK_LINE_SIZE),
.NUM_BANKS (NUM_BANKS),
.WORD_SIZE (WORD_SIZE),
.DFQQ_SIZE (DFQQ_SIZE)
) cache_dram_req_arb (
.clk (clk),
.reset (reset),
.per_bank_dram_fill_req_valid (per_bank_dram_fill_req_valid),
.per_bank_dram_fill_req_addr (per_bank_dram_fill_req_addr),
.dram_fill_req_ready (dram_fill_req_ready),
.per_bank_dram_wb_req_valid (per_bank_dram_wb_req_valid),
.per_bank_dram_wb_req_byteen (per_bank_dram_wb_req_byteen),
.per_bank_dram_wb_req_addr (per_bank_dram_wb_req_addr),
.per_bank_dram_wb_req_data (per_bank_dram_wb_req_data),
.per_bank_dram_wb_req_ready (per_bank_dram_wb_req_ready),
.dram_req_valid (dram_req_valid),
.dram_req_rw (dram_req_rw),
.dram_req_byteen (dram_req_byteen),
.dram_req_addr (dram_req_addr),
.dram_req_data (dram_req_data),
.dram_req_ready (dram_req_ready)
);
VX_cache_core_rsp_merge #(
.NUM_BANKS (NUM_BANKS),
.WORD_SIZE (WORD_SIZE),
@@ -461,6 +415,27 @@ module VX_cache #(
.core_rsp_ready (core_rsp_ready)
);
VX_cache_dram_req_arb #(
.BANK_LINE_SIZE (BANK_LINE_SIZE),
.NUM_BANKS (NUM_BANKS),
.WORD_SIZE (WORD_SIZE)
) cache_dram_req_arb (
.clk (clk),
.reset (reset),
.per_bank_dram_req_valid (per_bank_dram_req_valid),
.per_bank_dram_req_rw (per_bank_dram_req_rw),
.per_bank_dram_req_byteen (per_bank_dram_req_byteen),
.per_bank_dram_req_addr (per_bank_dram_req_addr),
.per_bank_dram_req_data (per_bank_dram_req_data),
.per_bank_dram_req_ready (per_bank_dram_req_ready),
.dram_req_valid (dram_req_valid),
.dram_req_rw (dram_req_rw),
.dram_req_byteen (dram_req_byteen),
.dram_req_addr (dram_req_addr),
.dram_req_data (dram_req_data),
.dram_req_ready (dram_req_ready)
);
VX_snp_rsp_arb #(
.NUM_BANKS (NUM_BANKS),
.BANK_LINE_SIZE (BANK_LINE_SIZE),

92
hw/rtl/cache/VX_cache_dram_fill_arb.v vendored
View File

@@ -1,92 +0,0 @@
`include "VX_cache_config.vh"
module VX_cache_dram_fill_arb #(
// Size of line inside a bank in bytes
parameter BANK_LINE_SIZE = 0,
// Number of banks {1, 2, 4, 8,...}
parameter NUM_BANKS = 0,
// Dram Fill Req Queue Size
parameter DFQQ_SIZE = 0
) (
input wire clk,
input wire reset,
input wire dfqq_push,
input wire[NUM_BANKS-1:0] per_bank_dram_fill_req_valid,
input wire[NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_fill_req_addr,
input wire dfqq_pop,
output wire dfqq_req,
output wire[`DRAM_ADDR_WIDTH-1:0] dfqq_req_addr,
output wire dfqq_empty,
output wire dfqq_full
);
reg [NUM_BANKS-1:0] use_per_bank_dram_fill_req_valid;
reg [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] use_per_bank_dram_fill_req_addr;
wire [NUM_BANKS-1:0] out_per_bank_dram_fill_req_valid;
wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] out_per_bank_dram_fill_req_addr;
wire [NUM_BANKS-1:0] use_per_bqual_bank_dram_fill_req_valid;
wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] qual_bank_dram_fill_req_addr;
wire [NUM_BANKS-1:0] updated_bank_dram_fill_req_valid;
wire o_empty;
wire use_empty = !(| use_per_bank_dram_fill_req_valid);
wire out_empty = !(| out_per_bank_dram_fill_req_valid) || o_empty;
wire push_qual = dfqq_push && !dfqq_full;
wire pop_qual = dfqq_pop && use_empty && !out_empty;
VX_generic_queue #(
.DATAW(NUM_BANKS * (1+`DRAM_ADDR_WIDTH)),
.SIZE(DFQQ_SIZE)
) dfqq_queue (
.clk (clk),
.reset (reset),
.push (push_qual),
.data_in ({per_bank_dram_fill_req_valid, per_bank_dram_fill_req_addr}),
.pop (pop_qual),
.data_out({out_per_bank_dram_fill_req_valid, out_per_bank_dram_fill_req_addr}),
.empty (o_empty),
.full (dfqq_full),
`UNUSED_PIN (size)
);
assign use_per_bqual_bank_dram_fill_req_valid = use_empty ? (out_per_bank_dram_fill_req_valid & {NUM_BANKS{!o_empty}}) : (use_per_bank_dram_fill_req_valid & {NUM_BANKS{!use_empty}});
assign qual_bank_dram_fill_req_addr = use_empty ? out_per_bank_dram_fill_req_addr : use_per_bank_dram_fill_req_addr;
wire[`BANK_BITS-1:0] qual_request_index;
wire qual_has_request;
VX_fixed_arbiter #(
.N(NUM_BANKS)
) sel_bank (
.clk (clk),
.reset (reset),
.requests (use_per_bqual_bank_dram_fill_req_valid),
.grant_index (qual_request_index),
.grant_valid (qual_has_request),
`UNUSED_PIN (grant_onehot)
);
assign dfqq_empty = !qual_has_request;
assign dfqq_req = use_per_bqual_bank_dram_fill_req_valid [qual_request_index];
assign dfqq_req_addr = qual_bank_dram_fill_req_addr[qual_request_index];
assign updated_bank_dram_fill_req_valid = use_per_bqual_bank_dram_fill_req_valid & (~(1 << qual_request_index));
always @(posedge clk) begin
if (reset) begin
use_per_bank_dram_fill_req_valid <= 0;
use_per_bank_dram_fill_req_addr <= 0;
end else begin
if (dfqq_pop && qual_has_request) begin
use_per_bank_dram_fill_req_valid <= updated_bank_dram_fill_req_valid;
use_per_bank_dram_fill_req_addr <= qual_bank_dram_fill_req_addr;
end
end
end
endmodule

82
hw/rtl/cache/VX_cache_dram_req_arb.v vendored
View File

@@ -6,88 +6,50 @@ module VX_cache_dram_req_arb #(
// Number of banks {1, 2, 4, 8,...}
parameter NUM_BANKS = 0,
// Size of a word in bytes
parameter WORD_SIZE = 0,
// Dram Fill Req Queue Size
parameter DFQQ_SIZE = 0
parameter WORD_SIZE = 0
) (
input wire clk,
input wire reset,
// Fill Request
input wire [NUM_BANKS-1:0] per_bank_dram_fill_req_valid,
input wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_fill_req_addr,
output wire dram_fill_req_ready,
// Inputs
input wire [NUM_BANKS-1:0] per_bank_dram_req_valid,
input wire [NUM_BANKS-1:0] per_bank_dram_req_rw,
input wire [NUM_BANKS-1:0][BANK_LINE_SIZE-1:0] per_bank_dram_req_byteen,
input wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_req_addr,
input wire [NUM_BANKS-1:0][`BANK_LINE_WIDTH-1:0] per_bank_dram_req_data,
output wire [NUM_BANKS-1:0] per_bank_dram_req_ready,
// Writeback Request
input wire [NUM_BANKS-1:0] per_bank_dram_wb_req_valid,
input wire [NUM_BANKS-1:0][BANK_LINE_SIZE-1:0] per_bank_dram_wb_req_byteen,
input wire [NUM_BANKS-1:0][`DRAM_ADDR_WIDTH-1:0] per_bank_dram_wb_req_addr,
input wire [NUM_BANKS-1:0][`BANK_LINE_WIDTH-1:0] per_bank_dram_wb_req_data,
output wire [NUM_BANKS-1:0] per_bank_dram_wb_req_ready,
// Merged Request
// Output
output wire dram_req_valid,
output wire dram_req_rw,
output wire [BANK_LINE_SIZE-1:0] dram_req_byteen,
output wire [`DRAM_ADDR_WIDTH-1:0] dram_req_addr,
output wire [`BANK_LINE_WIDTH-1:0] dram_req_data,
input wire dram_req_ready
);
wire dwb_valid;
wire dfqq_req;
wire[`DRAM_ADDR_WIDTH-1:0] dfqq_req_addr;
`DEBUG_BEGIN
wire dfqq_empty;
`DEBUG_END
wire dfqq_pop = !dwb_valid && dfqq_req && dram_req_ready; // If no dwb, and dfqq has valids, then pop
wire dfqq_push = (| per_bank_dram_fill_req_valid);
wire dfqq_full;
VX_cache_dram_fill_arb #(
.BANK_LINE_SIZE(BANK_LINE_SIZE),
.NUM_BANKS(NUM_BANKS),
.DFQQ_SIZE(DFQQ_SIZE)
) dram_fill_arb (
.clk (clk),
.reset (reset),
.dfqq_push (dfqq_push),
.per_bank_dram_fill_req_valid (per_bank_dram_fill_req_valid),
.per_bank_dram_fill_req_addr (per_bank_dram_fill_req_addr),
.dfqq_pop (dfqq_pop),
.dfqq_req (dfqq_req),
.dfqq_req_addr (dfqq_req_addr),
.dfqq_empty (dfqq_empty),
.dfqq_full (dfqq_full)
);
assign dram_fill_req_ready = !dfqq_full;
wire [`BANK_BITS-1:0] dwb_bank;
wire [`BANK_BITS-1:0] sel_bank;
wire sel_valid;
VX_fixed_arbiter #(
.N(NUM_BANKS)
) sel_dwb (
) sel_arb (
.clk (clk),
.reset (reset),
.requests (per_bank_dram_wb_req_valid),
.grant_index (dwb_bank),
.grant_valid (dwb_valid),
.requests (per_bank_dram_req_valid),
.grant_index (sel_bank),
.grant_valid (sel_valid),
`UNUSED_PIN (grant_onehot)
);
assign dram_req_valid = sel_valid;
assign dram_req_rw = per_bank_dram_req_rw[sel_bank];
assign dram_req_byteen = per_bank_dram_req_byteen[sel_bank];
assign dram_req_addr = per_bank_dram_req_addr[sel_bank];
assign dram_req_data = per_bank_dram_req_data[sel_bank];
for (genvar i = 0; i < NUM_BANKS; i++) begin
assign per_bank_dram_wb_req_ready[i] = dram_req_ready && (dwb_bank == `BANK_BITS'(i));
assign per_bank_dram_req_ready[i] = dram_req_ready && (sel_bank == `BANK_BITS'(i));
end
assign dram_req_valid = dwb_valid || dfqq_req;
assign dram_req_rw = dwb_valid;
assign dram_req_byteen = dwb_valid ? per_bank_dram_wb_req_byteen[dwb_bank] : {BANK_LINE_SIZE{1'b1}};
assign dram_req_addr = dwb_valid ? per_bank_dram_wb_req_addr[dwb_bank] : dfqq_req_addr;
assign {dram_req_data} = dwb_valid ? per_bank_dram_wb_req_data[dwb_bank] : 0;
endmodule

159
hw/rtl/cache/VX_cache_miss_resrv.v vendored
View File

@@ -3,6 +3,7 @@
module VX_cache_miss_resrv #(
parameter CACHE_ID = 0,
parameter BANK_ID = 0,
parameter CORE_TAG_ID_BITS = 0,
// Size of line inside a bank in bytes
parameter BANK_LINE_SIZE = 0,
// Number of banks {1, 2, 4, 8,...}
@@ -21,9 +22,21 @@ module VX_cache_miss_resrv #(
input wire clk,
input wire reset,
// Miss enqueue
`ifdef DBG_CORE_REQ_INFO
`IGNORE_WARNINGS_BEGIN
input wire[31:0] debug_pc_st0,
input wire[`NR_BITS-1:0] debug_rd_st0,
input wire[`NW_BITS-1:0] debug_wid_st0,
input wire[`UP(CORE_TAG_ID_BITS)-1:0] debug_tagid_st0,
input wire[31:0] debug_pc_st2,
input wire[`NR_BITS-1:0] debug_rd_st2,
input wire[`NW_BITS-1:0] debug_wid_st2,
input wire[`UP(CORE_TAG_ID_BITS)-1:0] debug_tagid_st2,
`IGNORE_WARNINGS_END
`endif
// enqueue
input wire miss_add,
input wire is_mrvq,
input wire[`LINE_ADDR_WIDTH-1:0] miss_add_addr,
input wire[`UP(`WORD_SELECT_WIDTH)-1:0] miss_add_wsel,
input wire[`WORD_WIDTH-1:0] miss_add_data,
@@ -31,20 +44,21 @@ module VX_cache_miss_resrv #(
input wire[`REQ_TAG_WIDTH-1:0] miss_add_tag,
input wire miss_add_rw,
input wire[WORD_SIZE-1:0] miss_add_byteen,
input wire mrvq_init_ready_state,
input wire miss_add_is_snp,
input wire miss_add_snp_invalidate,
input wire is_msrq_st2,
input wire init_ready_state_st2,
output wire miss_resrv_full,
output wire miss_resrv_stop,
output wire miss_resrv_almfull,
// Broadcast Address
input wire is_fill_st1,
input wire[`LINE_ADDR_WIDTH-1:0] fill_addr_st1,
// fill
input wire update_ready_st0,
input wire[`LINE_ADDR_WIDTH-1:0] fill_addr_st0,
output wire pending_hazard_st0,
output wire pending_hazard_st1,
// Miss dequeue
input wire miss_resrv_pop,
// dequeue
input wire miss_resrv_schedule_st0,
output wire miss_resrv_valid_st0,
output wire[`LINE_ADDR_WIDTH-1:0] miss_resrv_addr_st0,
output wire[`UP(`WORD_SELECT_WIDTH)-1:0] miss_resrv_wsel_st0,
@@ -54,43 +68,38 @@ module VX_cache_miss_resrv #(
output wire miss_resrv_rw_st0,
output wire[WORD_SIZE-1:0] miss_resrv_byteen_st0,
output wire miss_resrv_is_snp_st0,
output wire miss_resrv_snp_invalidate_st0
output wire miss_resrv_snp_invalidate_st0,
input wire miss_resrv_pop_st2
);
localparam FULL_DISTANCE = 2; // need 2 cycles window to prevent pipeline lock
wire [`MRVQ_METADATA_WIDTH-1:0] metadata_table;
reg [MRVQ_SIZE-1:0][`LINE_ADDR_WIDTH-1:0] addr_table;
`NO_RW_RAM_CHECK reg [`LINE_ADDR_WIDTH-1:0] addr_table [MRVQ_SIZE-1:0];
reg [MRVQ_SIZE-1:0] valid_table;
reg [MRVQ_SIZE-1:0] ready_table;
reg [`LOG2UP(MRVQ_SIZE)-1:0] schedule_ptr;
reg [`LOG2UP(MRVQ_SIZE)-1:0] schedule_ptr, restore_ptr;
reg [`LOG2UP(MRVQ_SIZE)-1:0] head_ptr;
reg [`LOG2UP(MRVQ_SIZE)-1:0] tail_ptr;
reg [`LOG2UP(MRVQ_SIZE+1)-1:0] size;
`STATIC_ASSERT(MRVQ_SIZE > 5, ("invalid size"))
`STATIC_ASSERT(MRVQ_SIZE > FULL_DISTANCE, ("invalid size"))
assign miss_resrv_full = (size == $bits(size)'(MRVQ_SIZE));
assign miss_resrv_stop = (size > $bits(size)'(MRVQ_SIZE-5)); // need to add 5 cycles to prevent pipeline lock
wire enqueue_possible = !miss_resrv_full;
wire [`LOG2UP(MRVQ_SIZE)-1:0] enqueue_index = tail_ptr;
reg [MRVQ_SIZE-1:0] make_ready;
reg [MRVQ_SIZE-1:0] make_ready_push;
reg [MRVQ_SIZE-1:0] valid_address_match;
assign miss_resrv_almfull = (size >= $bits(size)'(MRVQ_SIZE-FULL_DISTANCE));
wire [MRVQ_SIZE-1:0] valid_address_match;
for (genvar i = 0; i < MRVQ_SIZE; i++) begin
assign valid_address_match[i] = valid_table[i] ? (addr_table[i] == fill_addr_st1) : 0;
assign make_ready[i] = is_fill_st1 && valid_address_match[i];
assign valid_address_match[i] = valid_table[i] && (addr_table[i] == fill_addr_st0);
end
assign pending_hazard_st1 = |(valid_address_match);
assign pending_hazard_st0 = (| valid_address_match);
wire dequeue_possible = valid_table[schedule_ptr] && ready_table[schedule_ptr];
wire [`LOG2UP(MRVQ_SIZE)-1:0] dequeue_index = schedule_ptr;
wire dequeue_ready = valid_table[schedule_ptr] && ready_table[schedule_ptr];
assign miss_resrv_valid_st0 = dequeue_possible;
assign miss_resrv_addr_st0 = addr_table[dequeue_index];
assign miss_resrv_valid_st0 = dequeue_ready;
assign miss_resrv_addr_st0 = addr_table[schedule_ptr];
assign {miss_resrv_data_st0,
miss_resrv_tid_st0,
miss_resrv_tag_st0,
@@ -100,56 +109,51 @@ module VX_cache_miss_resrv #(
miss_resrv_is_snp_st0,
miss_resrv_snp_invalidate_st0} = metadata_table;
wire mrvq_push = miss_add && enqueue_possible && !is_mrvq;
wire mrvq_pop = miss_resrv_pop && dequeue_possible;
wire msrq_push = miss_add && !is_msrq_st2;
wire recover_state = miss_add && is_mrvq;
wire increment_head = !miss_add && is_mrvq;
wire update_ready = (|make_ready);
wire qual_mrvq_init = mrvq_push && mrvq_init_ready_state;
assign make_ready_push = (MRVQ_SIZE'(qual_mrvq_init)) << enqueue_index;
wire [`LOG2UP(MRVQ_SIZE)-1:0] head_ptr_n = head_ptr + $bits(head_ptr)'(1);
always @(posedge clk) begin
if (reset) begin
valid_table <= 0;
ready_table <= 0;
size <= 0;
schedule_ptr <= 0;
restore_ptr <= 0;
head_ptr <= 0;
tail_ptr <= 0;
size <= 0;
end else begin
if (mrvq_push) begin
valid_table[enqueue_index] <= 1;
ready_table[enqueue_index] <= mrvq_init_ready_state;
addr_table[enqueue_index] <= miss_add_addr;
if (update_ready_st0) begin
ready_table <= ready_table | valid_address_match;
end
if (miss_add) begin
assert(!miss_resrv_full);
if (is_msrq_st2) begin
// returning missed msrq entry, restore schedule
valid_table[restore_ptr] <= 1;
ready_table[restore_ptr] <= init_ready_state_st2;
restore_ptr <= restore_ptr + $bits(restore_ptr)'(1);
schedule_ptr <= head_ptr;
end else begin
valid_table[tail_ptr] <= 1;
ready_table[tail_ptr] <= init_ready_state_st2;
addr_table[tail_ptr] <= miss_add_addr;
tail_ptr <= tail_ptr + $bits(tail_ptr)'(1);
end else if (increment_head) begin
valid_table[head_ptr] <= 0;
head_ptr <= head_ptr + $bits(head_ptr)'(1);
end else if (recover_state) begin
schedule_ptr <= schedule_ptr - $bits(schedule_ptr)'(1);
end
// update entry as 'ready' during DRAM fill response
if (update_ready) begin
ready_table <= ready_table | make_ready | make_ready_push;
end
if (mrvq_pop) begin
ready_table[dequeue_index] <= 0;
schedule_ptr <= schedule_ptr + $bits(schedule_ptr)'(1);
end
if (!(mrvq_push && increment_head)) begin
if (mrvq_push) begin
size <= size + $bits(size)'(1);
end
if (increment_head) begin
end else if (miss_resrv_pop_st2) begin
head_ptr <= head_ptr_n;
restore_ptr <= head_ptr_n;
valid_table[head_ptr] <= 0;
size <= size - $bits(size)'(1);
end
if (miss_resrv_schedule_st0) begin
assert(miss_resrv_valid_st0);
valid_table[schedule_ptr] <= 0;
schedule_ptr <= schedule_ptr + $bits(schedule_ptr)'(1);
end
end
end
@@ -160,11 +164,11 @@ module VX_cache_miss_resrv #(
.BYTEENW(1),
.BUFFERED(0),
.RWCHECK(1)
) metadata_ram (
) metadata (
.clk(clk),
.waddr(enqueue_index),
.raddr(dequeue_index),
.wren(mrvq_push),
.waddr(tail_ptr),
.raddr(schedule_ptr),
.wren(msrq_push),
.rden(1'b1),
.din({miss_add_data, miss_add_tid, miss_add_tag, miss_add_rw, miss_add_byteen, miss_add_wsel, miss_add_is_snp, miss_add_snp_invalidate}),
.dout(metadata_table)
@@ -172,8 +176,16 @@ module VX_cache_miss_resrv #(
`ifdef DBG_PRINT_CACHE_MSRQ
always @(posedge clk) begin
if (mrvq_push || mrvq_pop || increment_head || recover_state) begin
$write("%t: cache%0d:%0d msrq: push=%b pop=%b incr=%d recv=%d", $time, CACHE_ID, BANK_ID, mrvq_push, mrvq_pop, increment_head, recover_state);
if (miss_add || miss_resrv_schedule_st0 || miss_resrv_pop_st2) begin
if (miss_add)
if (is_msrq_st2)
$write("%t: cache%0d:%0d msrq-restore addr%0d=%0h ready=%b", $time, CACHE_ID, BANK_ID, restore_ptr, `LINE_TO_BYTE_ADDR(miss_add_addr, BANK_ID), init_ready_state_st2);
else
$write("%t: cache%0d:%0d msrq-push addr%0d=%0h ready=%b wid=%0d PC=%0h", $time, CACHE_ID, BANK_ID, tail_ptr, `LINE_TO_BYTE_ADDR(miss_add_addr, BANK_ID), init_ready_state_st2, debug_wid_st2, debug_pc_st2);
else if (miss_resrv_schedule_st0)
$write("%t: cache%0d:%0d msrq-schedule wid=%0d PC=%0h", $time, CACHE_ID, BANK_ID, debug_wid_st0, debug_pc_st0);
else if (miss_resrv_pop_st2)
$write("%t: cache%0d:%0d msrq-pop addr%0d wid=%0d PC=%0h", $time, CACHE_ID, BANK_ID, head_ptr, debug_wid_st2, debug_pc_st2);
for (integer j = 0; j < MRVQ_SIZE; j++) begin
if (valid_table[j]) begin
$write(" ");
@@ -181,6 +193,11 @@ module VX_cache_miss_resrv #(
if (~ready_table[j]) $write("!");
$write("addr%0d=%0h", j, `LINE_TO_BYTE_ADDR(addr_table[j], BANK_ID));
end
else if (schedule_ptr == $bits(schedule_ptr)'(j)) begin
$write(" *");
if (~ready_table[j]) $write("!");
$write("[addr%0d=%0h]", j, `LINE_TO_BYTE_ADDR(addr_table[j], BANK_ID));
end
end
$write("\n");
end

16
hw/rtl/cache/VX_snp_rsp_arb.v vendored
View File

@@ -17,25 +17,25 @@ module VX_snp_rsp_arb #(
input wire snp_rsp_ready
);
wire [`BANK_BITS-1:0] fsq_bank;
wire fsq_valid;
wire [`BANK_BITS-1:0] sel_bank;
wire sel_valid;
VX_fixed_arbiter #(
.N(NUM_BANKS)
) sel_ffsq (
) sel_arb (
.clk (clk),
.reset (reset),
.requests (per_bank_snp_rsp_valid),
.grant_index (fsq_bank),
.grant_valid (fsq_valid),
.grant_index (sel_bank),
.grant_valid (sel_valid),
`UNUSED_PIN (grant_onehot)
);
assign snp_rsp_valid = fsq_valid;
assign snp_rsp_tag = per_bank_snp_rsp_tag[fsq_bank];
assign snp_rsp_valid = sel_valid;
assign snp_rsp_tag = per_bank_snp_rsp_tag[sel_bank];
for (genvar i = 0; i < NUM_BANKS; i++) begin
assign per_bank_snp_rsp_ready[i] = snp_rsp_ready && (fsq_bank == `BANK_BITS'(i));
assign per_bank_snp_rsp_ready[i] = snp_rsp_ready && (sel_bank == `BANK_BITS'(i));
end
endmodule

121
hw/rtl/cache/VX_tag_data_access.v vendored
View File

@@ -31,15 +31,10 @@ module VX_tag_data_access #(
`IGNORE_WARNINGS_END
`endif
input wire stall,
input wire is_snp_st1,
input wire snp_invalidate_st1,
input wire stall_bank_pipe,
input wire force_request_miss_st1,
input wire[`LINE_SELECT_BITS-1:0] readaddr_st1,
input wire[`LINE_ADDR_WIDTH-1:0] writeaddr_st1,
input wire[`LINE_ADDR_WIDTH-1:0] addr_st1,
input wire valid_req_st1,
input wire writefill_st1,
@@ -52,17 +47,15 @@ module VX_tag_data_access #(
input wire[`UP(`WORD_SELECT_WIDTH)-1:0] wordsel_st1,
`IGNORE_WARNINGS_END
input wire force_miss_st1,
output wire[`WORD_WIDTH-1:0] readword_st1,
output wire[`BANK_LINE_WIDTH-1:0] readdata_st1,
output wire[`TAG_SELECT_BITS-1:0] readtag_st1,
output wire miss_st1,
output wire dirty_st1,
output wire[BANK_LINE_SIZE-1:0] dirtyb_st1,
output wire fill_saw_dirty_st1,
output wire snp_to_mrvq_st1,
output wire mrvq_init_ready_state_st1
output wire[BANK_LINE_SIZE-1:0] dirtyb_st1
);
`UNUSED_VAR (stall)
wire qual_read_valid_st1;
wire qual_read_dirty_st1;
@@ -78,15 +71,11 @@ module VX_tag_data_access #(
wire[`BANK_LINE_WORDS-1:0][WORD_SIZE-1:0] use_write_enable;
wire[`BANK_LINE_WIDTH-1:0] use_write_data;
wire fill_sent;
wire invalidate_line;
wire use_invalidate;
wire tags_match;
wire real_writefill = valid_req_st1 && writefill_st1
&& ((~use_read_valid_st1) || (use_read_valid_st1 && ~tags_match));
wire[`TAG_SELECT_BITS-1:0] writetag_st1 = writeaddr_st1[`TAG_LINE_ADDR_RNG];
wire[`LINE_SELECT_BITS-1:0] writeladdr_st1 = writeaddr_st1[`LINE_SELECT_BITS-1:0];
wire[`TAG_SELECT_BITS-1:0] addrtag_st1 = addr_st1[`TAG_LINE_ADDR_RNG];
wire[`LINE_SELECT_BITS-1:0] addrline_st1 = addr_st1[`LINE_SELECT_BITS-1:0];
VX_tag_data_store #(
.CACHE_SIZE (CACHE_SIZE),
@@ -96,27 +85,25 @@ module VX_tag_data_access #(
) tag_data_store (
.clk (clk),
.reset (reset),
.stall_bank_pipe(stall_bank_pipe),
.read_addr (readaddr_st1),
.read_addr (addrline_st1),
.read_valid (qual_read_valid_st1),
.read_dirty (qual_read_dirty_st1),
.read_dirtyb (qual_read_dirtyb_st1),
.read_tag (qual_read_tag_st1),
.read_data (qual_read_data_st1),
.invalidate (invalidate_line),
.invalidate (use_invalidate),
.write_enable(use_write_enable),
.write_fill (real_writefill),
.write_addr (writeladdr_st1),
.tag_index (writetag_st1),
.write_data (use_write_data),
.fill_sent (fill_sent)
.write_fill (writefill_st1),
.write_addr (addrline_st1),
.tag_index (addrtag_st1),
.write_data (use_write_data)
);
assign use_read_valid_st1 = qual_read_valid_st1 || !DRAM_ENABLE; // If shared memory, always valid
assign use_read_dirty_st1 = qual_read_dirty_st1 && DRAM_ENABLE && WRITE_ENABLE; // Dirty only applies in Dcache
assign use_read_tag_st1 = DRAM_ENABLE ? qual_read_tag_st1 : writetag_st1; // Tag is always the same in SM
assign use_read_tag_st1 = DRAM_ENABLE ? qual_read_tag_st1 : addrtag_st1; // Tag is always the same in SM
assign use_read_dirtyb_st1= qual_read_dirtyb_st1;
assign use_read_data_st1 = qual_read_data_st1;
@@ -131,67 +118,69 @@ module VX_tag_data_access #(
end
end
wire [`BANK_LINE_WORDS-1:0][WORD_SIZE-1:0] we;
wire [`BANK_LINE_WORDS-1:0][WORD_SIZE-1:0] write_enable;
wire [`BANK_LINE_WIDTH-1:0] data_write;
wire should_write = mem_rw_st1
&& valid_req_st1
&& use_read_valid_st1
&& ~miss_st1
&& ~is_snp_st1
&& ~real_writefill;
wire normal_write = valid_req_st1
&& !writefill_st1
&& !is_snp_st1
&& !miss_st1
&& !force_miss_st1
&& mem_rw_st1
&& use_read_valid_st1;
wire fill_write = valid_req_st1 && writefill_st1 && !force_miss_st1;
for (genvar i = 0; i < `BANK_LINE_WORDS; i++) begin
wire normal_write = ((`WORD_SELECT_WIDTH == 0) || (wordsel_st1 == `UP(`WORD_SELECT_WIDTH)'(i)))
&& should_write;
wire normal_write_w = ((`WORD_SELECT_WIDTH == 0) || (wordsel_st1 == `UP(`WORD_SELECT_WIDTH)'(i)))
&& normal_write;
assign we[i] = real_writefill ? {WORD_SIZE{1'b1}} :
normal_write ? mem_byteen_st1 :
assign write_enable[i] = fill_write ? {WORD_SIZE{1'b1}} :
normal_write_w ? mem_byteen_st1 :
{WORD_SIZE{1'b0}};
assign data_write[i * `WORD_WIDTH +: `WORD_WIDTH] = real_writefill ? writedata_st1[i * `WORD_WIDTH +: `WORD_WIDTH] : writeword_st1;
assign data_write[i * `WORD_WIDTH +: `WORD_WIDTH] = writefill_st1 ? writedata_st1[i * `WORD_WIDTH +: `WORD_WIDTH] : writeword_st1;
end
assign use_write_enable = (writefill_st1 && ~real_writefill) ? 0 : we;
assign use_write_data = data_write;
// use "case equality" to handle uninitialized tag when block entry is not valid
assign tags_match = (writetag_st1 === use_read_tag_st1);
assign tags_match = (addrtag_st1 === use_read_tag_st1);
wire snoop_hit_no_pending = valid_req_st1 && is_snp_st1 && use_read_valid_st1 && tags_match && (use_read_dirty_st1 || snp_invalidate_st1) && ~force_request_miss_st1;
wire req_invalid = valid_req_st1 && ~is_snp_st1 && ~use_read_valid_st1 && ~writefill_st1;
wire req_miss = valid_req_st1 && ~is_snp_st1 && use_read_valid_st1 && ~writefill_st1 && ~tags_match;
wire real_miss = req_invalid || req_miss;
wire force_core_miss = (force_request_miss_st1 && ~is_snp_st1 && ~writefill_st1 && valid_req_st1 && ~real_miss);
assign snp_to_mrvq_st1 = valid_req_st1 && is_snp_st1 && force_request_miss_st1;
assign use_write_enable = write_enable;
assign use_write_data = data_write;
assign use_invalidate = valid_req_st1 && is_snp_st1 && use_read_valid_st1 && tags_match
&& (use_read_dirty_st1 || snp_invalidate_st1) // block is dirty or need to force invalidation
&& !force_miss_st1;
// The second term is basically saying always make an entry ready if there's already antoher entry waiting, even if you yourself see a miss
assign mrvq_init_ready_state_st1 = snp_to_mrvq_st1
|| (force_request_miss_st1 && ~is_snp_st1 && ~writefill_st1 && valid_req_st1);
wire core_req_miss = valid_req_st1 && !is_snp_st1 && !writefill_st1 // is core request
&& (!use_read_valid_st1 || !tags_match); // block missing or has wrong tag
assign miss_st1 = real_miss || snoop_hit_no_pending || force_core_miss;
assign miss_st1 = core_req_miss;
assign dirty_st1 = valid_req_st1 && use_read_valid_st1 && use_read_dirty_st1;
assign dirtyb_st1 = use_read_dirtyb_st1;
assign readdata_st1 = use_read_data_st1;
assign readtag_st1 = use_read_tag_st1;
assign fill_sent = miss_st1;
assign fill_saw_dirty_st1 = real_writefill && dirty_st1;
assign invalidate_line = snoop_hit_no_pending;
`ifdef DBG_PRINT_CACHE_BANK
always @(*) begin
if (valid_req_st1 && writefill_st1) begin
if (!(!use_read_valid_st1 || !tags_match)) begin
$display("%t: warning: redundant fill - addr=%0h", $time, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID));
end
end
end
`ifdef DBG_PRINT_CACHE_DATA
always @(posedge clk) begin
if (valid_req_st1) begin
if ((| use_write_enable)) begin
if (writefill_st1) begin
$display("%t: cache%0d:%0d data-fill: wid=%0d, PC=%0h, tag=%0h, rd=%0d, dirty=%b, blk_addr=%0d, tag_id=%0h, data=%0h", $time, CACHE_ID, BANK_ID, debug_wid_st1, debug_pc_st1, debug_tagid_st1, debug_rd_st1, dirty_st1, writeladdr_st1, writetag_st1, use_write_data);
end else begin
$display("%t: cache%0d:%0d data-write: wid=%0d, PC=%0h, tag=%0h, rd=%0d, dirty=%b, blk_addr=%0d, tag_id=%0h, wsel=%0d, data=%0h", $time, CACHE_ID, BANK_ID, debug_wid_st1, debug_pc_st1, debug_tagid_st1, debug_rd_st1, dirty_st1, writeladdr_st1, writetag_st1, wordsel_st1, writeword_st1);
end
end else
if (miss_st1) begin
$display("%t: cache%0d:%0d data-miss: wid=%0d, PC=%0h, tag=%0h, rd=%0d, dirty=%b", $time, CACHE_ID, BANK_ID, debug_wid_st1, debug_pc_st1, debug_tagid_st1, debug_rd_st1, dirty_st1);
$display("%t: cache%0d:%0d data-miss: addr=%0h, wid=%0d, PC=%0h, valid=%b, tagmatch=%b, blk_addr=%0d, tag_id=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID), debug_wid_st1, debug_pc_st1, use_read_dirty_st1, tags_match, addrline_st1, addrtag_st1);
end else if ((| use_write_enable)) begin
if (writefill_st1) begin
$display("%t: cache%0d:%0d data-fill: addr=%0h, dirty=%b, blk_addr=%0d, tag_id=%0h, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID), dirty_st1, addrline_st1, addrtag_st1, use_write_data);
end else begin
$display("%t: cache%0d:%0d data-read: wid=%0d, PC=%0h, tag=%0h, rd=%0d, dirty=%b, blk_addr=%0d, tag_id=%0h, wsel=%0d, data=%0h", $time, CACHE_ID, BANK_ID, debug_wid_st1, debug_pc_st1, debug_tagid_st1, debug_rd_st1, dirty_st1, readaddr_st1, qual_read_tag_st1, wordsel_st1, qual_read_data_st1);
$display("%t: cache%0d:%0d data-write: addr=%0h, wid=%0d, PC=%0h, dirty=%b, blk_addr=%0d, tag_id=%0h, wsel=%0d, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID), debug_wid_st1, debug_pc_st1, dirty_st1, addrline_st1, addrtag_st1, wordsel_st1, writeword_st1);
end
end else begin
$display("%t: cache%0d:%0d data-read: addr=%0h, wid=%0d, PC=%0h, dirty=%b, blk_addr=%0d, tag_id=%0h, wsel=%0d, data=%0h", $time, CACHE_ID, BANK_ID, `LINE_TO_BYTE_ADDR(addr_st1, BANK_ID), debug_wid_st1, debug_pc_st1, dirty_st1, addrline_st1, qual_read_tag_st1, wordsel_st1, qual_read_data_st1);
end
end
end

59
hw/rtl/cache/VX_tag_data_store.v vendored
View File

@@ -12,7 +12,6 @@ module VX_tag_data_store #(
) (
input wire clk,
input wire reset,
input wire stall_bank_pipe,
input wire[`LINE_SELECT_BITS-1:0] read_addr,
output wire read_valid,
@@ -26,19 +25,13 @@ module VX_tag_data_store #(
input wire write_fill,
input wire[`LINE_SELECT_BITS-1:0] write_addr,
input wire[`TAG_SELECT_BITS-1:0] tag_index,
input wire[`BANK_LINE_WIDTH-1:0] write_data,
input wire fill_sent
input wire[`BANK_LINE_WIDTH-1:0] write_data
);
reg [`TAG_SELECT_BITS-1:0] tag [`BANK_LINE_COUNT-1:0];
reg [`BANK_LINE_WORDS-1:0][WORD_SIZE-1:0] dirtyb[`BANK_LINE_COUNT-1:0];
reg [`BANK_LINE_COUNT-1:0] dirty;
reg [`BANK_LINE_COUNT-1:0] valid;
assign read_valid = valid[read_addr];
assign read_dirty = dirty[read_addr];
assign read_dirtyb = dirtyb [read_addr];
assign read_tag = tag [read_addr];
wire do_write = (| write_enable);
@@ -48,30 +41,40 @@ module VX_tag_data_store #(
valid[i] <= 0;
dirty[i] <= 0;
end
end else if (!stall_bank_pipe) begin
if (do_write) begin
valid[write_addr] <= 1;
tag [write_addr] <= tag_index;
if (write_fill) begin
dirty[write_addr] <= 0;
dirtyb[write_addr] <= 0;
end else begin
dirty[write_addr] <= 1;
dirtyb[write_addr] <= dirtyb[write_addr] | write_enable;
end
end else if (fill_sent) begin
dirty[write_addr] <= 0;
dirtyb[write_addr] <= 0;
end
if (invalidate) begin
if (do_write) begin
assert(!invalidate);
dirty[write_addr] <= !write_fill;
valid[write_addr] <= 1;
end else if (invalidate) begin
valid[write_addr] <= 0;
end
end
end
wire [(`BANK_LINE_WORDS * WORD_SIZE)-1:0] ram_wren;
assign ram_wren = write_enable & {(`BANK_LINE_WORDS * WORD_SIZE){!stall_bank_pipe}};
reg [`BANK_LINE_WORDS-1:0][WORD_SIZE-1:0] dirtyb[`BANK_LINE_COUNT-1:0];
always @(posedge clk) begin
if (do_write) begin
dirtyb[write_addr] <= write_fill ? 0 : (dirtyb[write_addr] | write_enable);
end
end
assign read_dirtyb = dirtyb [read_addr];
VX_dp_ram #(
.DATAW(`TAG_SELECT_BITS),
.SIZE(`BANK_LINE_COUNT),
.BYTEENW(1),
.BUFFERED(0),
.RWCHECK(1)
) tags (
.clk(clk),
.waddr(write_addr),
.raddr(read_addr),
.wren(do_write),
.rden(1'b1),
.din(tag_index),
.dout(read_tag)
);
VX_dp_ram #(
.DATAW(`BANK_LINE_WORDS * WORD_SIZE * 8),
@@ -79,11 +82,11 @@ module VX_tag_data_store #(
.BYTEENW(`BANK_LINE_WORDS * WORD_SIZE),
.BUFFERED(0),
.RWCHECK(1)
) dp_ram (
) data (
.clk(clk),
.waddr(write_addr),
.raddr(read_addr),
.wren(ram_wren),
.wren(write_enable),
.rden(1'b1),
.din(write_data),
.dout(read_data)

2
hw/rtl/libs/VX_generic_queue.v
View File

@@ -3,7 +3,7 @@
module VX_generic_queue #(
parameter DATAW = 1,
parameter SIZE = 2,
parameter BUFFERED = 1,
parameter BUFFERED = 0,
parameter ADDRW = $clog2(SIZE),
parameter SIZEW = $clog2(SIZE+1)
) (

5
hw/rtl/libs/VX_priority_encoder.v
View File

@@ -8,21 +8,18 @@ module VX_priority_encoder #(
output wire valid_out
);
reg [`LOG2UP(N)-1:0] data_out_r;
reg valid_out_r;
always @(*) begin
data_out_r = 0;
valid_out_r = 0;
for (integer i = 0; i < N; i++) begin
if (data_in[i]) begin
data_out_r = `LOG2UP(N)'(i);
valid_out_r = 1;
break;
end
end
end
assign data_out = data_out_r;
assign valid_out = valid_out_r;
assign valid_out = (| data_in);
endmodule

5
hw/simulate/Makefile
View File

@@ -10,6 +10,7 @@ DBG_PRINT_FLAGS += -DDBG_PRINT_CORE_DCACHE
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_BANK
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_SNP
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_MSRQ
DBG_PRINT_FLAGS += -DDBG_PRINT_CACHE_DATA
DBG_PRINT_FLAGS += -DDBG_PRINT_DRAM
DBG_PRINT_FLAGS += -DDBG_PRINT_PIPELINE
DBG_PRINT_FLAGS += -DDBG_PRINT_OPAE
@@ -44,7 +45,7 @@ gen-s:
verilator $(VF) -DNDEBUG $(SINGLECORE) -CFLAGS '$(CF) -DNDEBUG $(SINGLECORE)'
gen-sd:
verilator $(VF) -O0 $(SINGLECORE) -CFLAGS '$(CF) -O0 -g $(DBG) $(SINGLECORE)' --trace-fst --trace-threads 1 $(DBG)
verilator $(VF) -O0 $(SINGLECORE) -CFLAGS '$(CF) -O0 -g $(DBG) $(SINGLECORE)' --trace --trace-structs --trace-threads 1 $(DBG)
gen-st:
verilator $(VF) -DNDEBUG $(SINGLECORE) -CFLAGS '$(CF) -DNDEBUG -O2 $(SINGLECORE)' --threads $(THREADS)
@@ -53,7 +54,7 @@ gen-m:
verilator $(VF) -DNDEBUG $(MULTICORE) -CFLAGS '$(CF) -DNDEBUG $(MULTICORE)'
gen-md:
verilator $(VF) $(MULTICORE) -CFLAGS '$(CF) -O0 -g $(DBG) $(MULTICORE)' --trace-fst --trace-threads 1 $(DBG)
verilator $(VF) $(MULTICORE) -CFLAGS '$(CF) -O0 -g $(DBG) $(MULTICORE)' --trace --trace-structs --trace-threads 1 $(DBG)
gen-mt:
verilator $(VF) -DNDEBUG $(MULTICORE) -CFLAGS '$(CF) -DNDEBUG -O2 $(MULTICORE)' --threads $(THREADS)

4
hw/simulate/simulator.cpp
View File

@@ -30,9 +30,9 @@ Simulator::Simulator() {
#ifdef VCD_OUTPUT
Verilated::traceEverOn(true);
trace_ = new VerilatedFstC();
trace_ = new VerilatedVcdC();
vortex_->trace(trace_, 99);
trace_->open("trace.fst");
trace_->open("trace.vcd");
#endif
// reset the device

4
hw/simulate/simulator.h
View File

@@ -5,7 +5,7 @@
#include "verilated.h"
#ifdef VCD_OUTPUT
#include <verilated_fst_c.h>
#include <verilated_vcd_c.h>
#endif
#include <VX_config.h>
@@ -76,6 +76,6 @@ private:
RAM *ram_;
VVortex *vortex_;
#ifdef VCD_OUTPUT
VerilatedFstC *trace_;
VerilatedVcdC *trace_;
#endif
};

1
hw/unit_tests/cache/Makefile vendored
View File

@@ -8,6 +8,7 @@ DBG_PRINT_FLAGS = -DDBG_PRINT_CORE_ICACHE \
-DDBG_PRINT_CACHE_BANK \
-DDBG_PRINT_CACHE_SNP \
-DDBG_PRINT_CACHE_MSRQ \
-DDBG_PRINT_CACHE_DATA \
-DDBG_PRINT_DRAM \
-DDBG_PRINT_OPAE