github-mirrors/vortex
1
0
Fork 0
mirror of https://github.com/vortexgpgpu/vortex.git synced 2025-04-24 05:47:35 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Make ALU div/mul pipelines longer and support logic element multiplication mode for better long pipeline performance

Browse source
This commit is contained in:
wgulian3 2020-02-22 20:16:13 -05:00
parent a099cb25cf
commit ca61801199
2 changed files with 43 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

5
rtl/VX_alu.v
View file

@ -14,8 +14,8 @@ module VX_alu(
output reg out_alu_stall
);
localparam div_pipeline_len = 10;
localparam mul_pipeline_len = 3;
localparam div_pipeline_len = 20;
localparam mul_pipeline_len = 8;
wire[31:0] unsigned_div_result;
wire[31:0] unsigned_rem_result;
@ -62,6 +62,7 @@ module VX_alu(
.WIDTHB(64),
.WIDTHP(64),
.SPEED("HIGHEST"),
.FORCE_LE("YES"),
.PIPELINE(mul_pipeline_len)
) multiplier (
.clock(clk),

59
rtl/compat/VX_mult.v
View file

@ -5,7 +5,8 @@ module VX_mult
parameter WIDTHP=1,
parameter REP="UNSIGNED",
parameter SPEED="MIXED", // "MIXED" or "HIGHEST"
parameter PIPELINE=0
parameter PIPELINE=0,
parameter FORCE_LE="NO"
)
(
input clock, aclr, clken,
@ -30,21 +31,41 @@ module VX_mult
localparam lpm_speed=SPEED == "HIGHEST" ? 10:5;
lpm_mult#(
.LPM_WIDTHA(WIDTHA),
.LPM_WIDTHB(WIDTHB),
.LPM_WIDTHP(WIDTHP),
.LPM_REPRESENTATION(REP),
.LPM_PIPELINE(PIPELINE),
.MAXIMIZE_SPEED(lpm_speed)
) quartus_mult(
.clock(clock),
.aclr(aclr),
.clken(clken),
.dataa(dataa),
.datab(datab),
.result(result)
);
if (FORCE_LE == "YES") begin
lpm_mult#(
.LPM_WIDTHA(WIDTHA),
.LPM_WIDTHB(WIDTHB),
.LPM_WIDTHP(WIDTHP),
.LPM_REPRESENTATION(REP),
.LPM_PIPELINE(PIPELINE),
.DSP_BLOCK_BALANCING("LOGIC ELEMENTS"),
.MAXIMIZE_SPEED(lpm_speed)
) quartus_mult(
.clock(clock),
.aclr(aclr),
.clken(clken),
.dataa(dataa),
.datab(datab),
.result(result)
);
end
else begin
lpm_mult#(
.LPM_WIDTHA(WIDTHA),
.LPM_WIDTHB(WIDTHB),
.LPM_WIDTHP(WIDTHP),
.LPM_REPRESENTATION(REP),
.LPM_PIPELINE(PIPELINE),
.MAXIMIZE_SPEED(lpm_speed)
) quartus_mult(
.clock(clock),
.aclr(aclr),
.clken(clken),
.dataa(dataa),
.datab(datab),
.result(result)
);
end
end
else begin
@ -92,13 +113,13 @@ module VX_mult
/* * * * * * * * * * * * * * * * * * * * * * */
if (REP == "SIGNED") begin
assign result = $signed($signed(dataa_pipe_end) * $signed(datab_pipe_end));
assign result = $signed($signed(dataa_pipe_end)*$signed(datab_pipe_end));
end
else begin
assign result = dataa_pipe_end * datab_pipe_end;
assign result = dataa_pipe_end*datab_pipe_end;
end
end
endgenerate
endmodule : VX_mult
endmodule: VX_mult