DC LAB8 & SDC约束 & 四种时序路径分析

1.启动DC

dc_shell-t -64bit -topo

2.读入设计

read_ddc EXCEPTIONS.ddc
current_design EXCEPTIONS
link

---

3. 查看所有违例的约束报告

3.1 report_constraint -all_violators (alias rc)

---

3.2 view report_constraint -all_violators -verbose -significant_digits 4 (打印详细报告)

---

4.查看时序报告 report_timing -significant_digits 4

• 组合逻辑延时是7.7833-2=5.7833
• 对于adr_i/coff到comb logic到dout，(in2out,应该定义虚拟时钟vclk) ,将clk-->comb-->clk, clk-->comb-->vclk, vclk-->comb-->clk,这三条设置false
• 上图是adr_i/coff到comb logic到dout时序路径，有vio（clk–>input delay 2ns -->coff --> comb logic 5.7833ns --> dout） 这条路径应设置false path

---

5. 约束组合逻辑(adr_i/coff到comb logic到dout)

• To accomplish this,the combinatorial paths will be constrained using a new virtual clock which you will call vclk.
• The sequential paths are already constrained using the 5ns design clock clk,and input and output delays of 2ns and 4ns,respectively.
• The following schematic illustrates this concept (note that the two input ports, adr_ i and coeff,are collapsed into one port here to simplify the schematic):

5.1 创建虚拟时钟

create_clock -name vclk -period 6
set in_ports [get_ports "coeff* adr_i*"]
set_input_delay 0 -clock vclk -add_delay $in_ports
set_output_delay 0 -clock vclk -add_delay [all_outputs]

5.1.1 set_input_delay 用法

dc_shell-topo> set_input_delay -help
Usage: set_input_delay    # set input_delay
        [-clock clock_name]    (relative clock)
        [-clock_fall]          (delay is relative to falling edge of clock)
        [-level_sensitive]     (delay is from level-sensitive latch)
        [-rise]                (specifies rising delay)
        [-fall]                (specifies falling delay)
        [-max]                 (specifies maximum delay)
        [-min]                 (specifies minimum delay)
        [-add_delay]           (don't remove existing input delay)
        [-network_latency_included]
                               (specifies that clock_network_latency is included)
        [-source_latency_included]
                               (specifies that clock_source_latency is included)
        [-reference_pin pin_list]
                               (list of pins)
        delay_value            (path delay)
        port_pin_list          (list of ports and/or pins)

---

5.2 创建虚拟时钟vclk后再report_timing

view report_timing -group vclk -significant_digits 4

• 组合逻辑的路径应该只有vclk约束,设置clk-->comb-->clk, clk-->comb-->vclk, vclk-->comb-->clk为false
• The launch clock is vclk and the capture clock is vclk . This is not correct!
• The combinatorial paths should be constrained only by vclk . The two clocks should not interact.

---

5.2.1 组合逻辑约束,clk与vclk logically_exclusive（设置clk-->comb-->vclk, vclk-->comb-->clk为false）

• 对于clk-->comb-->clk只有手动设置false了 （5.3.1）

#两种时钟，四种组合，约束后只能从clk到clk或者vclk到vclk
#
set_clock_group -name false_grp1 -logically_exclusive -group clk -group vclk


# Altenatively:
#
# set_false_path -from [get_clocks clk] -to [get_clocks vclk]
# set_false_path -from [get_clocks vclk] -to [get_clocks clk]

---

5.2.2 设置clk与vclk llogically_exclusive后再去report_timing -group vclk -significant_digits 4

• 组合逻辑延时 5.7833ns

---

5.3 查看clk的timing， report_timing -group clk -significant_digits 4

• 这里不会显示vclk-->clk,因为前面(5.2.1)设置了logically exclusive
• 只会显示clk-->clk

• 上图是addi/coff到comb到dout的时序路径，有vio（clk–>input delay 2ns -->coff --> comb logic 5.7833ns --> dout） 这条路径设置false path ，
• 这条路径用虚拟时钟vclk检查
• 前面5.2.1虽然设置了logic exclusive（clk-->comb-->vclk, vclk-->comb-->clk为false），但是clk—>comb–>clk还是得手动去 false

---

5.3.1 组合逻辑约束，设置clk--->comb-->clk为false path

set_false_path -from [get_clocks clk] \
	-through $in_ports \
	-through [all_outputs] -to [get_clocks clk]

---

5.3.2 设置clk--->comb-->clk为false path 后再report_timing -from $in_ports -to [all_outputs]

• 由于设置clk-->comb-->clk, clk-->comb-->vclk, vclk-->comb-->clk为false，故只会显示vclk–>vcomb–>clk这条时序路径

---

6. 查看clk group timing ，report_timing -group clk

• 从下图可以看出reg2reg 中间组合逻辑延时为9.1589ns，Tcq=0.4383
• 组合逻辑介于一个周期和两个周期之间，所以设置multicycle path（setup设置N，hold则对应N-1）

两个寄存器间的Tcq+Tcomb = 9.5943，单周期检查太严格，故设置多周期检查，setup在第二个周期在检查

6.1 设置mul_result_reg*/D （reg2reg）多周期检查setup set_multicycle_path

##若setup设置N，hold 对应设置为N-1
set_multicycle_path 2 -setup -to mul_result_reg*/D

6.1.1 设置完setup，多周期再检查reg2reg（coeff_q_reg_14_ to mul_result_reg_31_）的setup

• 由于只设置了setup，所以这里检查setup的 comb max delay是没有问题的
• 但是检查hold，就过于严苛了

6.2 检查hold（肯定violation，多周期设置没设置hold，只设置了setup，hold现在在5ns检查太严格了）

---

6.2.1 设置 mul_result_reg*/D （reg2reg）多周期检查hold

set_multicycle_path 1 -hold -t o mul_result_reg*/D

6.2.2 设置hold多周期后再检查

report_timing -significant_digits 4 -to mul_result_reg*/D -delay min

7.report_constraint -all_violators

• 所有timing vio都已经清除完毕了

8.四种时序路径总结

8.1 in2reg

report_timing -group clk

[-add_delay]           (don't remove existing input delay)

---

8.2 reg2reg

• 这里设置了set_multicycle_path 2 -setup -to mul_result_reg*/D

---

8.3 reg2out

---

8.4 in2out

8.4.1 in2out(adr_i[0]-->dout[6]) demo

report_timing -from adr_* -to dout[6]
report_timing -from adr_* -to dout*

---

8.4.2 in2out(coeff[5]-->dout[6]) demo

report_timing -from coeff* -to dout[6]
report_timing -from coeff[0] -to dout[6]

9.打印所有port、cells、pins

get_ports

---

---

get_cells

get_pins