Github_Repos/cvw
1
0
Fork 1
mirror of https://github.com/openhwgroup/cvw synced 2025-02-11 06:05:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

Removed spill support from icache.

Browse Source
This commit is contained in:
Ross Thompson 2022-01-03 21:03:02 -06:00
parent 8c7638688b
commit b045d84147
2 changed files with 9 additions and 227 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

51
wally-pipelined/src/cache/icache.sv vendored
View File

@ -77,18 +77,10 @@ module icache
logic ICacheMemWriteEnable; logic ICacheMemWriteEnable;
logic [`PA_BITS-1:0] FinalPCPF; logic [`PA_BITS-1:0] FinalPCPF;
// Output signals from cache memory // Output signals from cache memory
logic [31:0] ICacheMemReadData;
logic ICacheReadEn; logic ICacheReadEn;
logic [BLOCKLEN-1:0] ReadLineF; logic [BLOCKLEN-1:0] ReadLineF;
logic [15:0] SpillDataBlock0;
logic spill;
logic spillSave;
logic [1:1] SelAdr_q;
logic [`PA_BITS-1:0] PCPSpillF;
logic [1:0] SelAdr; logic [1:0] SelAdr;
logic [INDEXLEN-1:0] RAdr; logic [INDEXLEN-1:0] RAdr;
@ -109,16 +101,10 @@ module icache
logic [NUMWAYS-1:0] SRAMWayWriteEnable; logic [NUMWAYS-1:0] SRAMWayWriteEnable;
// on spill we want to get the first 2 bytes of the next cache block. mux2 #(INDEXLEN)
// the spill only occurs if the PCPF mod BlockByteLength == -2. Therefore we can
// simply add 2 to land on the next cache block.
assign PCPSpillF = PCPF + {{{PA_WIDTH}{1'b0}}, 2'b10};
mux3 #(INDEXLEN)
AdrSelMux(.d0(PCNextF[INDEXLEN+OFFSETLEN-1:OFFSETLEN]), AdrSelMux(.d0(PCNextF[INDEXLEN+OFFSETLEN-1:OFFSETLEN]),
.d1(PCF[INDEXLEN+OFFSETLEN-1:OFFSETLEN]), .d1(PCF[INDEXLEN+OFFSETLEN-1:OFFSETLEN]),
.d2(PCPSpillF[INDEXLEN+OFFSETLEN-1:OFFSETLEN]), .s(SelAdr[0]),
.s(SelAdr),
.y(RAdr)); .y(RAdr));
@ -168,37 +154,10 @@ module icache
assign ReadLineSetsF[BLOCKLEN/16-1] = {16'b0, ReadLineF[BLOCKLEN-1:BLOCKLEN-16]}; assign ReadLineSetsF[BLOCKLEN/16-1] = {16'b0, ReadLineF[BLOCKLEN-1:BLOCKLEN-16]};
endgenerate endgenerate
assign ICacheMemReadData = ReadLineSetsF[FinalPCPF[$clog2(BLOCKLEN / 32) + 1 : 1]]; assign FinalInstrRawF = ReadLineSetsF[FinalPCPF[$clog2(BLOCKLEN / 32) + 1 : 1]];
// spills require storing the first cache block so it can merged
// with the second
// can optimize size, for now just make it the size of the data
// leaving the cache memory.
flopenr #(16) SpillInstrReg(.clk(clk),
.en(spillSave),
.reset(reset),
.d(ICacheMemReadData[15:0]),
.q(SpillDataBlock0));
assign FinalInstrRawF = spill ? {ICacheMemReadData[15:0], SpillDataBlock0} : ICacheMemReadData;
// Detect if the instruction is compressed
//assign CompressedF = FinalInstrRawF[1:0] != 2'b11;
//assign spill = &PCF[$clog2(BLOCKLEN/32)+1:1];
assign spill = 0;
assign FinalPCPF = PCPF;
// this mux needs to be delayed 1 cycle as it occurs 1 pipeline stage later.
// *** read enable may not be necessary.
flopenr #(1) SelAdrReg(.clk(clk),
.reset(reset),
.en(ICacheReadEn),
.d(SelAdr[1]),
.q(SelAdr_q[1]));
assign FinalPCPF = SelAdr_q[1] ? PCPSpillF : PCPF;
// *** CHANGE ME // *** CHANGE ME
// if not cacheable the offset bits needs to be sent to the EBU. // if not cacheable the offset bits needs to be sent to the EBU.
@ -228,8 +187,6 @@ module icache
.ICacheFetchLine, .ICacheFetchLine,
.CacheableF, .CacheableF,
.hit, .hit,
.spill,
.spillSave,
.SelAdr, .SelAdr,
.LRUWriteEn); .LRUWriteEn);

185
wally-pipelined/src/cache/icachefsm.sv vendored
View File

@ -43,7 +43,6 @@ module icachefsm
// icache internal inputs // icache internal inputs
input logic hit, input logic hit,
input logic spill,
// icache internal outputs // icache internal outputs
output logic ICacheReadEn, output logic ICacheReadEn,
@ -57,54 +56,19 @@ module icachefsm
output logic ICacheFetchLine, output logic ICacheFetchLine,
// icache internal outputs // icache internal outputs
output logic spillSave,
output logic [1:0] SelAdr, output logic [1:0] SelAdr,
output logic LRUWriteEn output logic LRUWriteEn
); );
// FSM states // FSM states
typedef enum {STATE_READY, typedef enum {STATE_READY,
STATE_HIT_SPILL, // spill, block 0 hit
STATE_HIT_SPILL_MISS_FETCH_WDV, // block 1 miss, issue read to AHB and wait data.
STATE_HIT_SPILL_MISS_FETCH_DONE, // write data into SRAM/LUT
STATE_HIT_SPILL_MERGE, // Read block 0 of CPU access, should be able to optimize into STATE_HIT_SPILL.
// a challenge is the spill signal gets us out of the ready state and moves us to
// 1 of the 2 spill branches. However the original fsm design had us return to
// the ready state when the spill + hits/misses were fully resolved. The problem
// is the spill signal is based on PCPF so when we return to READY to check if the
// cache has a hit it still expresses spill. We can fix in 1 of two ways.
// 1. we can add 1 extra state at the end of each spill branch to returns the instruction
// to the CPU advancing the CPU and icache to the next instruction.
// 2. We can assert a signal which is delayed 1 cycle to suppress the spill when we get
// to the READY state.
// The first first option is more robust and increases the number of states by 2. The
// second option is seams like it should work, but I worry there is a hidden interaction
// between CPU stalling and that register.
// Picking option 1.
STATE_HIT_SPILL_FINAL, // this state replicates STATE_READY's replay of the
// spill access but does nto consider spill. It also does not do another operation.
STATE_MISS_FETCH_WDV, // aligned miss, issue read to AHB and wait for data. STATE_MISS_FETCH_WDV, // aligned miss, issue read to AHB and wait for data.
STATE_MISS_FETCH_DONE, // write data into SRAM/LUT STATE_MISS_FETCH_DONE, // write data into SRAM/LUT
STATE_MISS_READ, // read block 1 from SRAM/LUT STATE_MISS_READ, // read block 1 from SRAM/LUT
STATE_MISS_READ_DELAY, // read block 1 from SRAM/LUT STATE_MISS_READ_DELAY, // read block 1 from SRAM/LUT
STATE_MISS_SPILL_FETCH_WDV, // spill, miss on block 0, issue read to AHB and wait STATE_CPU_BUSY
STATE_MISS_SPILL_FETCH_DONE, // write data into SRAM/LUT
STATE_MISS_SPILL_READ1, // read block 0 from SRAM/LUT
STATE_MISS_SPILL_2, // return to ready if hit or do second block update.
STATE_MISS_SPILL_2_START, // return to ready if hit or do second block update.
STATE_MISS_SPILL_MISS_FETCH_WDV, // miss on block 1, issue read to AHB and wait
STATE_MISS_SPILL_MISS_FETCH_DONE, // write data to SRAM/LUT
STATE_MISS_SPILL_MERGE, // read block 0 of CPU access,
STATE_MISS_SPILL_FINAL, // this state replicates STATE_READY's replay of the
// spill access but does nto consider spill. It also does not do another operation.
STATE_CPU_BUSY,
STATE_CPU_BUSY_SPILL
} statetype; } statetype;
(* mark_debug = "true" *) statetype CurrState, NextState; (* mark_debug = "true" *) statetype CurrState, NextState;
@ -119,7 +83,6 @@ module icachefsm
always_comb begin always_comb begin
//IfuBusFetch = 1'b0; //IfuBusFetch = 1'b0;
ICacheMemWriteEnable = 1'b0; ICacheMemWriteEnable = 1'b0;
spillSave = 1'b0;
SelAdr = 2'b00; SelAdr = 2'b00;
ICacheReadEn = 1'b0; ICacheReadEn = 1'b0;
ICacheStallF = 1'b1; ICacheStallF = 1'b1;
@ -137,7 +100,7 @@ module icachefsm
SelAdr = 2'b01; SelAdr = 2'b01;
ICacheStallF = 1'b0; ICacheStallF = 1'b0;
end end
else if (CacheableF & hit & ~spill) begin else if (CacheableF & hit) begin
ICacheStallF = 1'b0; ICacheStallF = 1'b0;
LRUWriteEn = 1'b1; LRUWriteEn = 1'b1;
if(CPUBusy) begin if(CPUBusy) begin
@ -146,17 +109,9 @@ module icachefsm
end else begin end else begin
NextState = STATE_READY; NextState = STATE_READY;
end end
end else if (CacheableF & hit & spill) begin end else if (CacheableF & ~hit) begin
spillSave = 1'b1;
SelAdr = 2'b10;
LRUWriteEn = 1'b1;
NextState = STATE_HIT_SPILL;
end else if (CacheableF & ~hit & ~spill) begin
SelAdr = 2'b01; /// *********( SelAdr = 2'b01; /// *********(
NextState = STATE_MISS_FETCH_WDV; NextState = STATE_MISS_FETCH_WDV;
end else if (CacheableF & ~hit & spill) begin
SelAdr = 2'b01;
NextState = STATE_MISS_SPILL_FETCH_WDV;
end else begin end else begin
if(CPUBusy) begin if(CPUBusy) begin
NextState = STATE_CPU_BUSY; NextState = STATE_CPU_BUSY;
@ -166,49 +121,6 @@ module icachefsm
end end
end end
end end
// branch 1, hit spill and 2, miss spill hit
STATE_HIT_SPILL: begin
SelAdr = 2'b10;
ICacheReadEn = 1'b1;
if (hit) begin
NextState = STATE_HIT_SPILL_FINAL;
end else begin
NextState = STATE_HIT_SPILL_MISS_FETCH_WDV;
end
end
STATE_HIT_SPILL_MISS_FETCH_WDV: begin
SelAdr = 2'b10;
//IfuBusFetch = 1'b1;
if (ICacheBusAck) begin
NextState = STATE_HIT_SPILL_MISS_FETCH_DONE;
end else begin
NextState = STATE_HIT_SPILL_MISS_FETCH_WDV;
end
end
STATE_HIT_SPILL_MISS_FETCH_DONE: begin
SelAdr = 2'b10;
ICacheMemWriteEnable = 1'b1;
NextState = STATE_HIT_SPILL_MERGE;
end
STATE_HIT_SPILL_MERGE: begin
SelAdr = 2'b10;
ICacheReadEn = 1'b1;
NextState = STATE_HIT_SPILL_FINAL;
end
STATE_HIT_SPILL_FINAL: begin
ICacheReadEn = 1'b1;
SelAdr = 2'b00;
ICacheStallF = 1'b0;
LRUWriteEn = 1'b1;
if(CPUBusy) begin
NextState = STATE_CPU_BUSY_SPILL;
SelAdr = 2'b10;
end else begin
NextState = STATE_READY;
end
end
// branch 3 miss no spill // branch 3 miss no spill
STATE_MISS_FETCH_WDV: begin STATE_MISS_FETCH_WDV: begin
SelAdr = 2'b01; SelAdr = 2'b01;
@ -242,80 +154,6 @@ module icachefsm
NextState = STATE_READY; NextState = STATE_READY;
end end
end end
// branch 4 miss spill hit, and 5 miss spill miss
STATE_MISS_SPILL_FETCH_WDV: begin
SelAdr = 2'b01;
//IfuBusFetch = 1'b1;
if (ICacheBusAck) begin
NextState = STATE_MISS_SPILL_FETCH_DONE;
end else begin
NextState = STATE_MISS_SPILL_FETCH_WDV;
end
end
STATE_MISS_SPILL_FETCH_DONE: begin
SelAdr = 2'b01;
ICacheMemWriteEnable = 1'b1;
NextState = STATE_MISS_SPILL_READ1;
end
STATE_MISS_SPILL_READ1: begin // always be a hit as we just wrote that cache block.
SelAdr = 2'b01; // there is a 1 cycle delay after setting the address before the date arrives.
ICacheReadEn = 1'b1;
LRUWriteEn = 1'b1;
NextState = STATE_MISS_SPILL_2;
end
STATE_MISS_SPILL_2: begin
SelAdr = 2'b10;
spillSave = 1'b1; /// *** Could pipeline these to make it clearer in the fsm.
ICacheReadEn = 1'b1;
NextState = STATE_MISS_SPILL_2_START;
end
STATE_MISS_SPILL_2_START: begin
if (~hit) begin
NextState = STATE_MISS_SPILL_MISS_FETCH_WDV;
end else begin
ICacheReadEn = 1'b1;
SelAdr = 2'b00;
ICacheStallF = 1'b0;
LRUWriteEn = 1'b1;
if(CPUBusy) begin
NextState = STATE_CPU_BUSY_SPILL;
SelAdr = 2'b10;
end else begin
NextState = STATE_READY;
end
end
end
STATE_MISS_SPILL_MISS_FETCH_WDV: begin
SelAdr = 2'b10;
//IfuBusFetch = 1'b1;
if (ICacheBusAck) begin
NextState = STATE_MISS_SPILL_MISS_FETCH_DONE;
end else begin
NextState = STATE_MISS_SPILL_MISS_FETCH_WDV;
end
end
STATE_MISS_SPILL_MISS_FETCH_DONE: begin
SelAdr = 2'b10;
ICacheMemWriteEnable = 1'b1;
NextState = STATE_MISS_SPILL_MERGE;
end
STATE_MISS_SPILL_MERGE: begin
SelAdr = 2'b10;
ICacheReadEn = 1'b1;
NextState = STATE_MISS_SPILL_FINAL;
end
STATE_MISS_SPILL_FINAL: begin
ICacheReadEn = 1'b1;
SelAdr = 2'b00;
ICacheStallF = 1'b0;
LRUWriteEn = 1'b1;
if(CPUBusy) begin
NextState = STATE_CPU_BUSY_SPILL;
SelAdr = 2'b10;
end else begin
NextState = STATE_READY;
end
end
STATE_CPU_BUSY: begin STATE_CPU_BUSY: begin
ICacheStallF = 1'b0; ICacheStallF = 1'b0;
if(CPUBusy) begin if(CPUBusy) begin
@ -326,17 +164,6 @@ module icachefsm
NextState = STATE_READY; NextState = STATE_READY;
end end
end end
STATE_CPU_BUSY_SPILL: begin
ICacheStallF = 1'b0;
ICacheReadEn = 1'b1;
if(CPUBusy) begin
NextState = STATE_CPU_BUSY_SPILL;
SelAdr = 2'b10;
end
else begin
NextState = STATE_READY;
end
end
default: begin default: begin
SelAdr = 2'b01; SelAdr = 2'b01;
NextState = STATE_READY; NextState = STATE_READY;
@ -345,9 +172,7 @@ module icachefsm
endcase endcase
end end
assign ICacheFetchLine = (CurrState == STATE_HIT_SPILL_MISS_FETCH_WDV) | assign ICacheFetchLine = CurrState == STATE_MISS_FETCH_WDV;
(CurrState == STATE_MISS_FETCH_WDV) |
(CurrState == STATE_MISS_SPILL_FETCH_WDV) |
(CurrState == STATE_MISS_SPILL_MISS_FETCH_WDV);
endmodule endmodule