a2o/dev/pd/synth/readme.md

195 lines
6.9 KiB
Markdown

# Synthesis
```
yosys -s synth.yo &> yosys.txt
```
## Arrays
```
grep tri_1 verilog/work/*
verilog/work/fu_fpr.v: tri_144x78_2r4w fpr0(
verilog/work/fu_fpr.v: tri_144x78_2r4w fpr1(
verilog/work/iuq_ic_dir.v: tri_128x34_4w_1r1w idir(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array0(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array1(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array2(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array3(
verilog/work/mmq.v: tri_128x16_1r1w_1 lru_array0(
verilog/work/rv.v: tri_144x78_2r4w
verilog/work/xu_gpr.v: tri_144x78_2r4w gpr0(
verilog/work/xu_gpr.v: tri_144x78_2r4w gpr1(
grep tri_2 verilog/work/*
verilog/work/lq_data.v: tri_256x144_8w_1r1w #(.addressable_ports(256), .addressbus_width(8), .port_bitwidth(144), .bit_write_type(9), .ways(8)) tridcarr(
grep tri_3 verilog/work/*
verilog/work/lq_pfetch.v: tri_32x70_2w_1r1w rpt(
grep tri_5 verilog/work/*
verilog/work/iuq_ic_dir.v: tri_512x162_4w_0 idata(
grep tri_6 verilog/work/*
verilog/work/iuq_btb.v: tri_64x72_1r1w btb0(
verilog/work/lq_ctl.v: tri_64x34_8w_1r1w #(.addressable_ports(64), .addressbus_width(6), .port_bitwidth(WAYDATASIZE), .ways(8)) arr(
verilog/work/lq_ldq_relq.v: tri_64x144_1r1w rdat(
verilog/work/lq_lsq.v: tri_64x34_8w_1r1w #(.addressable_ports(64), .addressbus_width(6), .port_bitwidth(WAYDATASIZE), .ways(8)) arr(
verilog/work/xu_spr.v: tri_64x72_1r1w xu_spr_aspr(
grep tri_bht verilog/work/*
verilog/work/iuq.v: tri_bht_1024x8_1r1w bht0(
verilog/work/iuq.v: tri_bht_1024x8_1r1w bht1(
verilog/work/iuq.v: tri_bht_512x4_1r1w bht2(
grep tri_cam verilog/work/*
verilog/work/iuq_ic_ierat.v: tri_cam_16x143_1r1w1c ierat_cam(
verilog/work/lq_derat.v: tri_cam_32x143_1r1w1c derat_cam(
grep tri_iuq verilog/work/*
verilog/work/iuq_cpl.v: tri_iuq_cpl_arr #(.ADDRESSABLE_PORTS(64), .ADDRESSBUS_WIDTH(6), .PORT_BITWIDTH(entry_length), .LATCHED_READ(1'b1), .LATCHED_READ_DATA(1'b1), .LATCHED_WRITE(1'b1))
```
## By Unit
* XU (GPR, SPR)
verilog/work/xu_gpr.v: tri_144x78_2r4w gpr0(
verilog/work/xu_gpr.v: tri_144x78_2r4w gpr1(
verilog/work/xu_spr.v: tri_64x72_1r1w xu_spr_aspr(
* FU (FPR)
verilog/work/fu_fpr.v: tri_144x78_2r4w fpr0(
verilog/work/fu_fpr.v: tri_144x78_2r4w fpr1(
* RV (LQ)
verilog/work/rv.v: tri_144x78_2r4w
* IU (CPL, ERAT, DIR, DATA, BTB, BHT)
verilog/work/iuq_cpl.v: tri_iuq_cpl_arr #(.ADDRESSABLE_PORTS(64), .ADDRESSBUS_WIDTH(6), .PORT_BITWIDTH(entry_length), .LATCHED_READ(1'b1), .LATCHED_READ_DATA(1'b1), .LATCHED_WRITE(1'b1))
verilog/work/iuq_ic_ierat.v: tri_cam_16x143_1r1w1c ierat_cam(
verilog/work/iuq_ic_dir.v: tri_128x34_4w_1r1w idir(
verilog/work/iuq_ic_dir.v: tri_512x162_4w_0 idata(
verilog/work/iuq_btb.v: tri_64x72_1r1w btb0(
verilog/work/iuq.v: tri_bht_1024x8_1r1w bht0(
verilog/work/iuq.v: tri_bht_1024x8_1r1w bht1(
verilog/work/iuq.v: tri_bht_512x4_1r1w bht2(
* LQ (ERAT, DIR, DATA. PFETCH, RLDQ, STQ)
verilog/work/lq_derat.v: tri_cam_32x143_1r1w1c derat_cam(
verilog/work/lq_pfetch.v: tri_32x70_2w_1r1w rpt(
verilog/work/lq_data.v: tri_256x144_8w_1r1w #(.addressable_ports(256), .addressbus_width(8), .port_bitwidth(144), .bit_write_type(9), .ways(8)) tridcarr(
verilog/work/lq_ctl.v: tri_64x34_8w_1r1w #(.addressable_ports(64), .addressbus_width(6), .port_bitwidth(WAYDATASIZE), .ways(8)) arr(
verilog/work/lq_ldq_relq.v: tri_64x144_1r1w rdat(
verilog/work/lq_lsq.v: tri_64x34_8w_1r1w #(.addressable_ports(64), .addressbus_width(6), .port_bitwidth(WAYDATASIZE), .ways(8)) arr(
* MMU (TLB)
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array0(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array1(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array2(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array3(
verilog/work/mmq.v: tri_128x16_1r1w_1 lru_array0(
## By Type
### Normal
* tri_144x78_2r4w
verilog/work/xu_gpr.v: tri_144x78_2r4w gpr0(
verilog/work/xu_gpr.v: tri_144x78_2r4w gpr1(
verilog/work/fu_fpr.v: tri_144x78_2r4w fpr0(
verilog/work/fu_fpr.v: tri_144x78_2r4w fpr1(
verilog/work/rv.v: tri_144x78_2r4w
* tri_64x72_1r1w
verilog/work/xu_spr.v: tri_64x72_1r1w xu_spr_aspr(
verilog/work/iuq_btb.v: tri_64x72_1r1w btb0(
* tri_512x162_4w_0
verilog/work/iuq_ic_dir.v: tri_512x162_4w_0 idata(
* tri_32x70_2w_1r1w
verilog/work/lq_pfetch.v: tri_32x70_2w_1r1w rpt(
* tri_256x144_8w_1r1w
verilog/work/lq_data.v: tri_256x144_8w_1r1w #(.addressable_ports(256), .addressbus_width(8), .port_bitwidth(144), .bit_write_type(9), .ways(8)) tridcarr(
* tri_64x34_8w_1r1w
verilog/work/lq_ctl.v: tri_64x34_8w_1r1w #(.addressable_ports(64), .addressbus_width(6), .port_bitwidth(WAYDATASIZE), .ways(8)) arr(
verilog/work/lq_lsq.v: tri_64x34_8w_1r1w #(.addressable_ports(64), .addressbus_width(6), .port_bitwidth(WAYDATASIZE), .ways(8)) arr(
* tri_64x144_1r1w
verilog/work/lq_ldq_relq.v: tri_64x144_1r1w rdat(
* tri_128x168_1w_0
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array0(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array1(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array2(
verilog/work/mmq.v: tri_128x168_1w_0 tlb_array3(
* tri_128x16_1r1w_1
verilog/work/mmq.v: tri_128x16_1r1w_1 lru_array0(
### Complex
#### Branch History
* tri_bht_1024x8_1r1w
verilog/work/iuq.v: tri_bht_1024x8_1r1w bht0(
verilog/work/iuq.v: tri_bht_1024x8_1r1w bht1(
* inner array:
tri_512x16_1r1w_1 bht0(
* tri_bht_512x4_1r1w
verilog/work/iuq.v: tri_bht_512x4_1r1w bht2(
* inner array:
tri_512x16_1r1w_1 bht0(
#### Completion
* tri_iuq_cpl_arr
verilog/work/iuq_cpl.v: tri_iuq_cpl_arr #(.ADDRESSABLE_PORTS(64), .ADDRESSBUS_WIDTH(6), .PORT_BITWIDTH(entry_length), .LATCHED_READ(1'b1), .LATCHED_READ_DATA(1'b1), .LATCHED_WRITE(1'b1))
* inner arrays (143)
RAM64X1D #(.INIT(64'h0000000000000000)) RAM64X1D0(
RAM64X1D #(.INIT(64'h0000000000000000)) RAM64X1D1(
* =2x64x143
#### ERATs (CAM)
* tri_cam_16x143_1r1w1c
verilog/work/iuq_ic_ierat.v: tri_cam_16x143_1r1w1c ierat_cam(
* tri_cam_32x143_1r1w1c
verilog/work/lq_derat.v: tri_cam_32x143_1r1w1c derat_cam(
## Summary
* the difficult arrays are the 2r4w (gpr, fpr, rv) and the cams; everything else is 1r1w
* cpl array is 1r1w (written in iu6, read in cp0), arranged even/odd for i0/i1; CPL_Q_DEPTH=32 means 32 even + 32 odd(?)
* some of these are directly changed with gen parameters; others may be do-able with some combo or parameters/spr settings/simple logic changes
* GPR rename pool size
* FPR rename pool size
* completion queue depth
* IERAT size
* IC size
* IC ways
* BTB size
* BHT size
* DERAT size
* DC size
* DC ways
* TLB size
* TLB ways
* e.g. no xlate, small caches
* IERAT, DERAT replaced with single-entry always-hit (no CAMs)
* IC, DC 1W small data
* no TLB
* alter BTB, BHT, rename, completion as necessary
* mmu logic (and fpu if not needed) could be dropped