Skip to content

Commit

Permalink
Merge branch 'rtl_cache'
Browse files Browse the repository at this point in the history
  • Loading branch information
@tinebp
tinebp committed Nov 14, 2024
2 parents 6dbbc62 + dfc7b61 commit 5844de8
Show file tree
Hide file tree
Showing 81 changed files with 3,029 additions and 2,325 deletions.
19 changes: 13 additions & 6 deletions ci/regression.sh.in
View file
Original file line number Diff line number Diff line change
Expand Up @@ -163,8 +163,9 @@ cache()
CONFIGS="-DL1_LINE_SIZE=$XSIZE -DLMEM_DISABLE" ./ci/blackbox.sh --driver=simx --app=sgemmx

# test cache ways
CONFIGS="-DICACHE_NUM_WAYS=8 -DDCACHE_NUM_WAYS=8" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
CONFIGS="-DICACHE_NUM_WAYS=8 -DDCACHE_NUM_WAYS=8" ./ci/blackbox.sh --driver=simx --app=sgemmx
CONFIGS="-DICACHE_NUM_WAYS=1 -DDCACHE_NUM_WAYS=1" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
CONFIGS="-DICACHE_NUM_WAYS=4 -DDCACHE_NUM_WAYS=8" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
CONFIGS="-DICACHE_NUM_WAYS=4 -DDCACHE_NUM_WAYS=8" ./ci/blackbox.sh --driver=simx --app=sgemmx

# test cache banking
CONFIGS="-DLMEM_NUM_BANKS=4 -DDCACHE_NUM_BANKS=1" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
Expand All @@ -174,11 +175,17 @@ cache()
CONFIGS="-DDCACHE_NUM_BANKS=2" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
CONFIGS="-DDCACHE_NUM_BANKS=2" ./ci/blackbox.sh --driver=simx --app=sgemmx

# replacement policy
CONFIGS="-DDCACHE_REPL_POLICY=0" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
CONFIGS="-DDCACHE_REPL_POLICY=1" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx
CONFIGS="-DDCACHE_REPL_POLICY=2" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx

# test writeback
CONFIGS="-DDCACHE_WRITEBACK=1" ./ci/blackbox.sh --driver=rtlsim --app=mstress
CONFIGS="-DDCACHE_WRITEBACK=1" ./ci/blackbox.sh --driver=simx --app=mstress
CONFIGS="-DSOCKET_SIZE=1 -DDCACHE_WRITEBACK=1 -DL2_WRITEBACK=1 -DL3_WRITEBACK=1" ./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --l2cache --l3cache --app=mstress
CONFIGS="-DSOCKET_SIZE=1 -DDCACHE_WRITEBACK=1 -DL2_WRITEBACK=1 -DL3_WRITEBACK=1" ./ci/blackbox.sh --driver=simx --cores=2 --clusters=2 --l2cache --l3cache --app=mstress
CONFIGS="-DDCACHE_WRITEBACK=1 -DDCACHE_DIRTYBYTES=0 -DDCACHE_NUM_WAYS=4" ./ci/blackbox.sh --driver=rtlsim --app=mstress
CONFIGS="-DDCACHE_WRITEBACK=1 -DDCACHE_DIRTYBYTES=1 -DDCACHE_NUM_WAYS=4" ./ci/blackbox.sh --driver=rtlsim --app=mstress
CONFIGS="-DDCACHE_WRITEBACK=1 -DDCACHE_NUM_WAYS=4" ./ci/blackbox.sh --driver=simx --app=mstress
CONFIGS="-DSOCKET_SIZE=1 -DL2_WRITEBACK=1 -DL3_WRITEBACK=1" ./ci/blackbox.sh --driver=rtlsim --cores=2 --clusters=2 --l2cache --l3cache --app=mstress
CONFIGS="-DSOCKET_SIZE=1 -DL2_WRITEBACK=1 -DL3_WRITEBACK=1" ./ci/blackbox.sh --driver=simx --cores=2 --clusters=2 --l2cache --l3cache --app=mstress

# cache clustering
CONFIGS="-DSOCKET_SIZE=4 -DNUM_DCACHES=4 -DNUM_ICACHES=2" ./ci/blackbox.sh --driver=rtlsim --app=sgemmx --cores=4 --warps=1 --threads=2
Expand Down
8 changes: 4 additions & 4 deletions configure
View file
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,7 @@ copy_files() {
filename_no_ext="${filename%.in}"
dest_file="$dest_dir/$filename_no_ext"
mkdir -p "$dest_dir"
sed "s|@VORTEX_HOME@|$SCRIPT_DIR|g; s|@XLEN@|$XLEN|g; s|@TOOLDIR@|$TOOLDIR|g; s|@OSVERSION@|$OSVERSION|g; s|@INSTALLDIR@|$PREFIX|g; s|@CURRENTDIR@|$CURRENT_DIR|g" "$file" > "$dest_file"
sed "s|@VORTEX_HOME@|$SOURCE_DIR|g; s|@XLEN@|$XLEN|g; s|@TOOLDIR@|$TOOLDIR|g; s|@OSVERSION@|$OSVERSION|g; s|@INSTALLDIR@|$PREFIX|g; s|@BUILDDIR@|$CURRENT_DIR|g" "$file" > "$dest_file"
# apply permissions to bash scripts
read -r firstline < "$dest_file"
if [[ "$firstline" =~ ^#!.*bash ]]; then
Expand Down Expand Up @@ -169,8 +169,8 @@ fi
SUBDIRS=("." "!ci" "!perf" "hw*" "kernel*" "runtime*" "sim*" "tests*")

# Get the directory of the script
SCRIPT_DIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"
SOURCE_DIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"

THIRD_PARTY_DIR=$SCRIPT_DIR/third_party
THIRD_PARTY_DIR=$SOURCE_DIR/third_party

copy_files "$SCRIPT_DIR" "$CURRENT_DIR"
copy_files "$SOURCE_DIR" "$CURRENT_DIR"
12 changes: 7 additions & 5 deletions hw/rtl/VX_cluster.sv
View file
Original file line number Diff line number Diff line change
Expand Up @@ -67,7 +67,7 @@ module VX_cluster import VX_gpu_pkg::*; #(
);

VX_gbar_unit #(
.INSTANCE_ID ($sformatf("gbar%0d", CLUSTER_ID))
.INSTANCE_ID (`SFORMATF(("gbar%0d", CLUSTER_ID)))
) gbar_unit (
.clk (clk),
.reset (reset),
Expand All @@ -84,7 +84,7 @@ module VX_cluster import VX_gpu_pkg::*; #(
`RESET_RELAY (l2_reset, reset);

VX_cache_wrap #(
.INSTANCE_ID ($sformatf("%s-l2cache", INSTANCE_ID)),
.INSTANCE_ID (`SFORMATF(("%s-l2cache", INSTANCE_ID))),
.CACHE_SIZE (`L2_CACHE_SIZE),
.LINE_SIZE (`L2_LINE_SIZE),
.NUM_BANKS (`L2_NUM_BANKS),
Expand All @@ -98,8 +98,10 @@ module VX_cluster import VX_gpu_pkg::*; #(
.TAG_WIDTH (L2_TAG_WIDTH),
.WRITE_ENABLE (1),
.WRITEBACK (`L2_WRITEBACK),
.DIRTY_BYTES (`L2_WRITEBACK),
.DIRTY_BYTES (`L2_DIRTYBYTES),
.REPL_POLICY (`L2_REPL_POLICY),
.UUID_WIDTH (`UUID_WIDTH),
.FLAGS_WIDTH (`MEM_REQ_FLAGS_WIDTH),
.CORE_OUT_BUF (3),
.MEM_OUT_BUF (3),
.NC_ENABLE (1),
Expand Down Expand Up @@ -129,7 +131,7 @@ module VX_cluster import VX_gpu_pkg::*; #(

VX_socket #(
.SOCKET_ID ((CLUSTER_ID * `NUM_SOCKETS) + socket_id),
.INSTANCE_ID ($sformatf("%s-socket%0d", INSTANCE_ID, socket_id))
.INSTANCE_ID (`SFORMATF(("%s-socket%0d", INSTANCE_ID, socket_id)))
) socket (
`SCOPE_IO_BIND (scope_socket+socket_id)

Expand All @@ -152,6 +154,6 @@ module VX_cluster import VX_gpu_pkg::*; #(
);
end

`BUFFER_EX(busy, (| per_socket_busy), 1'b1, (`NUM_SOCKETS > 1));
`BUFFER_EX(busy, (| per_socket_busy), 1'b1, 1, (`NUM_SOCKETS > 1));

endmodule
61 changes: 44 additions & 17 deletions hw/rtl/VX_config.vh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -170,6 +170,10 @@
`define L3_LINE_SIZE `MEM_BLOCK_SIZE
`endif

`ifndef MEMORY_BANKS
`define MEMORY_BANKS 2
`endif

`ifdef XLEN_64

`ifndef STACK_BASE_ADDR
Expand Down Expand Up @@ -566,7 +570,12 @@

// Number of Associative Ways
`ifndef ICACHE_NUM_WAYS
`define ICACHE_NUM_WAYS 1
`define ICACHE_NUM_WAYS 4
`endif

// Replacement Policy
`ifndef ICACHE_REPL_POLICY
`define ICACHE_REPL_POLICY 1
`endif

// Dcache Configurable Knobs //////////////////////////////////////////////////
Expand Down Expand Up @@ -615,19 +624,29 @@

// Memory Response Queue Size
`ifndef DCACHE_MRSQ_SIZE
`define DCACHE_MRSQ_SIZE 0
`define DCACHE_MRSQ_SIZE 4
`endif

// Number of Associative Ways
`ifndef DCACHE_NUM_WAYS
`define DCACHE_NUM_WAYS 1
`define DCACHE_NUM_WAYS 4
`endif

// Enable Cache Writeback
`ifndef DCACHE_WRITEBACK
`define DCACHE_WRITEBACK 0
`endif

// Enable Cache Dirty bytes
`ifndef DCACHE_DIRTYBYTES
`define DCACHE_DIRTYBYTES `DCACHE_WRITEBACK
`endif

// Replacement Policy
`ifndef DCACHE_REPL_POLICY
`define DCACHE_REPL_POLICY 1
`endif

// LMEM Configurable Knobs ////////////////////////////////////////////////////

`ifndef LMEM_DISABLE
Expand All @@ -650,12 +669,8 @@

// Cache Size
`ifndef L2_CACHE_SIZE
`ifdef ALTERA_S10
`define L2_CACHE_SIZE 2097152
`else
`define L2_CACHE_SIZE 1048576
`endif
`endif

// Number of Banks
`ifndef L2_NUM_BANKS
Expand All @@ -679,28 +694,34 @@

// Memory Response Queue Size
`ifndef L2_MRSQ_SIZE
`define L2_MRSQ_SIZE 0
`define L2_MRSQ_SIZE 4
`endif

// Number of Associative Ways
`ifndef L2_NUM_WAYS
`define L2_NUM_WAYS 2
`define L2_NUM_WAYS 8
`endif

// Enable Cache Writeback
`ifndef L2_WRITEBACK
`define L2_WRITEBACK 0
`endif

// Enable Cache Dirty bytes
`ifndef L2_DIRTYBYTES
`define L2_DIRTYBYTES `L2_WRITEBACK
`endif

// Replacement Policy
`ifndef L2_REPL_POLICY
`define L2_REPL_POLICY 1
`endif

// L3cache Configurable Knobs /////////////////////////////////////////////////

// Cache Size
`ifndef L3_CACHE_SIZE
`ifdef ALTERA_S10
`define L3_CACHE_SIZE 2097152
`else
`define L3_CACHE_SIZE 1048576
`endif
`endif

// Number of Banks
Expand All @@ -725,21 +746,27 @@

// Memory Response Queue Size
`ifndef L3_MRSQ_SIZE
`define L3_MRSQ_SIZE 0
`define L3_MRSQ_SIZE 4
`endif

// Number of Associative Ways
`ifndef L3_NUM_WAYS
`define L3_NUM_WAYS 4
`define L3_NUM_WAYS 8
`endif

// Enable Cache Writeback
`ifndef L3_WRITEBACK
`define L3_WRITEBACK 0
`endif

`ifndef MEMORY_BANKS
`define MEMORY_BANKS 2
// Enable Cache Dirty bytes
`ifndef L3_DIRTYBYTES
`define L3_DIRTYBYTES `L3_WRITEBACK
`endif

// Replacement Policy
`ifndef L3_REPL_POLICY
`define L3_REPL_POLICY 1
`endif

// Number of Memory Ports from LLC
Expand Down
6 changes: 3 additions & 3 deletions hw/rtl/VX_define.vh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -335,10 +335,10 @@
.data_out (dst) \
)

`define BUFFER_EX(dst, src, ena, latency) \
`define BUFFER_EX(dst, src, ena, RSTW, latency) \
VX_pipe_register #( \
.DATAW ($bits(dst)), \
.RESETW ($bits(dst)), \
.RESETW (RSTW), \
.DEPTH (latency) \
) __``dst``__ ( \
.clk (clk), \
Expand All @@ -348,7 +348,7 @@
.data_out (dst) \
)

`define BUFFER(dst, src) `BUFFER_EX(dst, src, 1'b1, 1)
`define BUFFER(dst, src) `BUFFER_EX(dst, src, 1'b1, 0, 1)

`define POP_COUNT_EX(out, in, model) \
VX_popcount #( \
Expand Down
42 changes: 18 additions & 24 deletions hw/rtl/VX_platform.vh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -37,16 +37,13 @@ endgenerate
`define ASSERT(cond, msg) \
assert(cond) else $error msg

`define RUNTIME_ASSERT(cond, msg) \
always @(posedge clk) begin \
assert(cond) else $error msg; \
`define RUNTIME_ASSERT(cond, msg) \
always @(posedge clk) begin \
if (!reset) begin \
`ASSERT(cond, msg); \
end \
end

`define __SCOPE
`define __SCOPE_X
`define __SCOPE_ON
`define __SCOPE_OFF

`ifndef TRACING_ALL
`define TRACING_ON /* verilator tracing_on */
`define TRACING_OFF /* verilator tracing_off */
Expand Down Expand Up @@ -128,6 +125,8 @@ endgenerate
end
`endif

`define SFORMATF(x) $sformatf x

`else // SYNTHESIS

`define STATIC_ASSERT(cond, msg)
Expand All @@ -137,6 +136,7 @@ endgenerate

`define DEBUG_BLOCK(x)
`define TRACE(level, args)
`define SFORMATF(x) ""

`define TRACING_ON
`define TRACING_OFF
Expand All @@ -153,45 +153,39 @@ endgenerate
`define UNUSED_PIN(x) . x ()
`define UNUSED_ARG(x) x

`define __SCOPE (* mark_debug="true" *)

`define __SCOPE_X

`define __SCOPE_ON \
`undef __SCOPE_X \
`define __SCOPE_X `__SCOPE

`define __SCOPE_OFF \
`undef __SCOPE_X \
`define __SCOPE_X

`endif

///////////////////////////////////////////////////////////////////////////////

`ifdef QUARTUS
`define MAX_FANOUT 8
`define IF_DATA_SIZE(x) $bits(x.data)
`define MAX_LUTRAM 1024
`define USE_BLOCK_BRAM (* ramstyle = "block" *)
`define USE_FAST_BRAM (* ramstyle = "MLAB, no_rw_check" *)
`define NO_RW_RAM_CHECK (* altera_attribute = "-name add_pass_through_logic_to_inferred_rams off" *)
`define DISABLE_BRAM (* ramstyle = "logic" *)
`define PRESERVE_NET (* preserve *)
`define BLACKBOX_CELL (* black_box *)
`define STRING string
`elsif VIVADO
`define MAX_FANOUT 8
`define IF_DATA_SIZE(x) $bits(x.data)
`define MAX_LUTRAM 1024
`define USE_BLOCK_BRAM (* ram_style = "block" *)
`define USE_FAST_BRAM (* ram_style = "distributed" *)
`define NO_RW_RAM_CHECK (* rw_addr_collision = "no" *)
`define DISABLE_BRAM (* ram_style = "registers" *)
`define PRESERVE_NET (* keep = "true" *)
`define BLACKBOX_CELL (* black_box *)
`define STRING
`else
`define MAX_FANOUT 8
`define IF_DATA_SIZE(x) x.DATA_WIDTH
`define MAX_LUTRAM 1024
`define USE_BLOCK_BRAM
`define USE_FAST_BRAM
`define NO_RW_RAM_CHECK
`define DISABLE_BRAM
`define PRESERVE_NET
`define BLACKBOX_CELL
`define STRING string
`endif

Expand All @@ -217,7 +211,7 @@ endgenerate

`define CLAMP(x, lo, hi) (((x) > (hi)) ? (hi) : (((x) < (lo)) ? (lo) : (x)))

`define UP(x) (((x) != 0) ? (x) : 1)
`define UP(x) (((x) > 0) ? (x) : 1)

`define CDIV(n,d) ((n + d - 1) / (d))

Expand Down
Loading

0 comments on commit 5844de8

Please sign in to comment.