Merged priv.S edits

This commit is contained in:
David Harris 2023-04-03 18:07:14 -07:00
commit 57ee9f3a5a
6 changed files with 119 additions and 23 deletions

38
src/cache/cache.sv vendored
View File

@ -96,8 +96,7 @@ module cache #(parameter LINELEN, NUMLINES, NUMWAYS, LOGBWPL, WORDLEN, MUXINTE
logic [LINELEN-1:0] ReadDataLine, ReadDataLineCache;
logic SelFetchBuffer;
logic CacheEn;
logic [CACHEWORDSPERLINE-1:0] MemPAdrDecoded;
logic [LINELEN/8-1:0] LineByteMask, DemuxedByteMask, FetchBufferByteSel;
logic [LINELEN/8-1:0] LineByteMask;
logic [$clog2(LINELEN/8) - $clog2(MUXINTERVAL/8) - 1:0] WordOffsetAddr;
genvar index;
@ -161,21 +160,30 @@ module cache #(parameter LINELEN, NUMLINES, NUMWAYS, LOGBWPL, WORDLEN, MUXINTE
/////////////////////////////////////////////////////////////////////////////////////////////
// Write Path
/////////////////////////////////////////////////////////////////////////////////////////////
if(!READ_ONLY_CACHE) begin:WriteSelLogic
logic [CACHEWORDSPERLINE-1:0] MemPAdrDecoded;
logic [LINELEN/8-1:0] DemuxedByteMask, FetchBufferByteSel;
// Adjust byte mask from word to cache line
onehotdecoder #(LOGCWPL) adrdec(.bin(PAdr[LOGCWPL+LOGLLENBYTES-1:LOGLLENBYTES]), .decoded(MemPAdrDecoded));
for(index = 0; index < 2**LOGCWPL; index++) begin
assign DemuxedByteMask[(index+1)*(WORDLEN/8)-1:index*(WORDLEN/8)] = MemPAdrDecoded[index] ? ByteMask : '0;
// Adjust byte mask from word to cache line
onehotdecoder #(LOGCWPL) adrdec(.bin(PAdr[LOGCWPL+LOGLLENBYTES-1:LOGLLENBYTES]), .decoded(MemPAdrDecoded));
for(index = 0; index < 2**LOGCWPL; index++) begin
assign DemuxedByteMask[(index+1)*(WORDLEN/8)-1:index*(WORDLEN/8)] = MemPAdrDecoded[index] ? ByteMask : '0;
end
assign FetchBufferByteSel = SetValid & ~SetDirty ? '1 : ~DemuxedByteMask; // If load miss set all muxes to 1.
// Merge write data into fetched cache line for store miss
for(index = 0; index < LINELEN/8; index++) begin
mux2 #(8) WriteDataMux(.d0(CacheWriteData[(8*index)%WORDLEN+7:(8*index)%WORDLEN]),
.d1(FetchBuffer[8*index+7:8*index]), .s(FetchBufferByteSel[index]), .y(LineWriteData[8*index+7:8*index]));
end
assign LineByteMask = SetValid ? '1 : SetDirty ? DemuxedByteMask : '0;
end
assign FetchBufferByteSel = SetValid & ~SetDirty ? '1 : ~DemuxedByteMask; // If load miss set all muxes to 1.
assign LineByteMask = SetValid ? '1 : SetDirty ? DemuxedByteMask : '0;
// Merge write data into fetched cache line for store miss
for(index = 0; index < LINELEN/8; index++) begin
mux2 #(8) WriteDataMux(.d0(CacheWriteData[(8*index)%WORDLEN+7:(8*index)%WORDLEN]),
.d1(FetchBuffer[8*index+7:8*index]), .s(FetchBufferByteSel[index]), .y(LineWriteData[8*index+7:8*index]));
end
else
begin:WriteSelLogic
// No need for this mux if the cache does not handle writes.
assign LineWriteData = FetchBuffer;
assign LineByteMask = '1;
end
/////////////////////////////////////////////////////////////////////////////////////////////
// Flush logic
/////////////////////////////////////////////////////////////////////////////////////////////

View File

@ -98,7 +98,9 @@ module cacheLRU
assign LRUUpdate[t1] = LRUUpdate[s] & WayEncoded[r];
end
mux2 #(1) LRUMuxes[NUMWAYS-2:0](CurrLRU, ~WayExpanded, LRUUpdate, NextLRU);
// The root node of the LRU tree will always be selected in LRUUpdate. No mux needed.
assign NextLRU[NUMWAYS-2] = ~WayExpanded[NUMWAYS-2];
mux2 #(1) LRUMuxes[NUMWAYS-3:0](CurrLRU[NUMWAYS-3:0], ~WayExpanded[NUMWAYS-3:0], LRUUpdate[NUMWAYS-3:0], NextLRU[NUMWAYS-3:0]);
// Compute next victim way.
for(s = NUMWAYS-2; s >= NUMWAYS/2; s--) begin
@ -128,8 +130,8 @@ module cacheLRU
always_ff @(posedge clk) begin
if (reset) for (int set = 0; set < NUMLINES; set++) LRUMemory[set] <= '0;
if(CacheEn) begin
if((InvalidateCache | FlushCache) & ~FlushStage) for (int set = 0; set < NUMLINES; set++) LRUMemory[set] <= '0;
else if (LRUWriteEn & ~FlushStage) begin
// if((InvalidateCache | FlushCache) & ~FlushStage) for (int set = 0; set < NUMLINES; set++) LRUMemory[set] <= '0;
if (LRUWriteEn & ~FlushStage) begin
LRUMemory[PAdr] <= NextLRU;
end
if(LRUWriteEn & ~FlushStage & (PAdr == CacheSet))

View File

@ -135,10 +135,16 @@ module decompress (
IllegalCompInstrD = 1;
InstrD = {16'b0, instr16}; // preserve instruction for mtval on trap
end
// coverage off
// are excluding this branch from coverage because in rv64gc XLEN is always 64 and thus greater than 32 bits
// This branch will only be taken if instr16[12:10] == 3'b111 and 'XLEN !> 32, because all other
// possible values for instr16[12:10] are covered by branches above. XLEN !> 32
// will never occur in rv64gc so this branch can not be covered
else begin // illegal instruction
IllegalCompInstrD = 1;
InstrD = {16'b0, instr16}; // preserve instruction for mtval on trap
end
// coverage on
5'b01101: InstrD = {immCJ, 5'b00000, 7'b1101111}; // c.j
5'b01110: InstrD = {immCB[11:5], 5'b00000, rs1p, 3'b000, immCB[4:0], 7'b1100011}; // c.beqz
5'b01111: InstrD = {immCB[11:5], 5'b00000, rs1p, 3'b001, immCB[4:0], 7'b1100011}; // c.bnez

View File

@ -57,6 +57,23 @@ main:
fcvt.l.q a0, ft3
fcvt.lu.q a0, ft3
// Tests verfying that half and quad floating point convertion instructions are not supported by rv64gc
# fcvt.h.d ft3, ft0 // Somehow this instruction is taking the route on line 124
// idea: enable the Q extension for this to work properly? A: Q and halfs not supported in rv64gc
# fcvt.h.w ft3, a0
# fcvt.w.h a0, ft0
# fcvt.q.w ft3, a0
# fcvt.w.q a0, ft0
# fcvt.q.d ft3, ft0
.word 0x38007553 // Testing the all False case for 119 - funct7 under, op = 101 0011
.word 0x40000053 // Line 145 All False Test case - illegal instruction?
.word 0xd0400053 // Line 156 All False Test case - illegal instruction?
.word 0xc0400053 // Line 162 All False Test case - illegal instruction?
.word 0xd2400053 // Line 168 All False Test case - illegal instruction?
.word 0xc2400053 // Line 174 All False Test case - illegal instruction?
# Test illegal instructions are detected
.word 0x00000007 // illegal floating-point load (bad Funct3)
.word 0x00000027 // illegal floating-point store (bad Funct3)

View File

@ -32,9 +32,23 @@ main:
csrs mstatus, t0
# calling compressed floating point load double instruction
//.halfword 0x2000 // CL type compressed floating-point ld-->funct3,imm,rs1',imm,rd',op
//.hword 0x2000 // CL type compressed floating-point ld-->funct3,imm,rs1',imm,rd',op
// binary version 0000 0000 0000 0000 0010 0000 0000 0000
mv s0, sp
c.fld fs0, 0(s0)
c.fsd fs0, 0(s0)
// c.fldsp fs0, 0
.hword 0x2002
// c.fsdsp fs0, 0
.hword 0xA002
//# Illegal compressed instruction with op = 01, instr[15:10] = 100111, and 0's everywhere else
//.hword 0x9C01
# Line Illegal compressed instruction
.hword 0x9C41
j done

View File

@ -36,11 +36,6 @@ main:
addi t0, zero, 0
csrr t0, stimecmp
# CSR coverage
csrw scause, zero
csrw stval, zero
csrw scounteren, zero
csrw satp, zero
# satp write with mstatus.TVM = 1
bseti t0, zero, 20
@ -62,4 +57,58 @@ main:
# Test write to STVAL, SCAUSE, SEPC, and STIMECMP CSRs
li t0, 0
csrw stval, t0
csrw scause, t0
csrw sepc, t0
csrw stimecmp, t0
csrw scounteren, zero
csrw satp, zero
# Switch to machine mode
li a0, 3
ecall
# Testing the HPMCOUNTERM performance counter: writing
# Base address is 2816 (MHPMCOUNTERBASE)
# There are 32 HPMCOUNTER registers
csrw 2816, t0
csrw 2817, t0
csrw 2818, t0
csrw 2819, t0
csrw 2820, t0
csrw 2821, t0
csrw 2822, t0
csrw 2823, t0
csrw 2824, t0
csrw 2825, t0
csrw 2826, t0
csrw 2827, t0
csrw 2828, t0
csrw 2829, t0
csrw 2830, t0
csrw 2831, t0
csrw 2832, t0
csrw 2833, t0
csrw 2834, t0
csrw 2835, t0
csrw 2836, t0
csrw 2837, t0
csrw 2838, t0
csrw 2839, t0
csrw 2840, t0
csrw 2841, t0
csrw 2842, t0
csrw 2843, t0
csrw 2844, t0
csrw 2845, t0
csrw 2846, t0
csrw 2847, t0
# Testing the HPMCOUNTERM performance counter: reading
csrr t0, 2817
j done