# [Assignment2: RISC-V Toolchain](https://hackmd.io/@sysprog/2022-arch-homework2)

reference: [黃冠予](https://hackmd.io/@ZLQisilvQvOh2DclLmk1bg/SyjKI7sZi#Optimization) [leetcode1720](https://leetcode.com/problems/decode-xored-array/)

Motivation: When I trace his assemble code, I find some useless instruct, like blow(line1 ~ line5), and he use malloc to get memory space for store result in c but he use address 0 for store result in assenmble code, i think it is not safety.

some use less instruction

index_zero:
    addi a4, x0, 0  // a4 = 0       
print: 
    slli t0, a4, 2  // t0 = 0 << 2 = 0     
    lw a0, 0(t0)    // load address(t0)
    
li a4,0
sw t0,0(a4)           #result[0] = first

In Ripes he use address 0(.text section) for store result, and his program actually change .text section
Memory map show address 0x00000000 - 0x000000f8 is .text section
before execute

after execution for Q1

## original code

### c

int* decode(int* encoded, int encodedSize, int first, int* returnSize){
    int* result = (int*)malloc(sizeof(int)*(encodedSize+1));
    result[0]=first;
    for(int i=0;i<encodedSize;i++){
        result[i+1]=result[i] ^ encoded[i];
    }
    *returnSize = encodedSize+1;
    return result;
}

### assemble code

.data 
num1: .word 1,2,3
numSize1: .word 3
first1: .word 1
num2: .word 6,2,7,3
numSize2: .word 4
first2: .word 4
num3: .word 5,6,7,8,9
numSize3: .word 5
first3: .word 2
nextline: .string "\n"
space: .string " "
.text
main:
    la a1,num1            #get num1 address
    la a2,numSize1        #get numSize1 address
    la a3,first1          #get first1 address
    lw a2,0(a2)           #a2 = numSize1 = 3
    lw a3,0(a3)           #a3 = first1 = 1
    li a4,0               #int i = 0
    li a5,0               #int count = 0
    li a6,3               #input set =3
    jal ra,decode.L1      #goto decode.L1
    la a1,num2            #get num2 address
    la a2,numSize2        #get numSize2 address
    la a3,first2          #get first2 address
    lw a2,0(a2)           #a2 = numSize2 = 4
    lw a3,0(a3)           #a3 = first2 = 4
    jal ra,decode.L1      #goto decode.L1
    la a1,num3            #get num3 address
    la a2,numSize3        #get numSize3 address
    la a3,first3          #get first3 address
    lw a2,0(a2)           #a2 = numSize3 = 5
    lw a3,0(a3)           #a3 = first3 = 2
decode.L1:
    add t0,a3,x0         
    sw t0,0(a4)           #result[0] = first
decode.L2:                #loop
    beq a4,a2,index_zero  #if i=numSize then goto index_zero   
    slli t0,a4,2          #a4*4
    add t2,t0,a1          #get num1[i]
    lw t1,0(t2)           #get num1[i] 
    lw t2,0(t0)           #get result[i]
    xor t1,t1,t2          #result[i] ^ num[i]
    addi t2,t0,4          #a4*4,k = i + 1
    sw t1,0(t2)           #result[k] = result[i] ^ num[i]
    addi a4,a4,1          #i++  
    j decode.L2           #goto decode.L2
index_zero:
    addi a4,x0,0          #i=0
print: 
    slli t0,a4,2          #a4*4
    lw a0,0(t0)           #get result[i]
    li a7,1               #systemcall print
    ecall                 #execute
    addi a4,a4,1          #i++
    la a0,space           #print " "
    li a7,4
    ecall
    ble a4,a2,print       #if i<numSize then goto print
    la a0,nextline        #nextline
    li a7,4
    ecall
    addi a5,a5,1          #count++
    beq a5,a6,exit        #excute all dataset
    li a4,0               #int i = 0
    jr ra
exit:
    li a7,10
    ecall

## My_Improvements
### c
adding test_data and some code makeing it can execute.

#include <stdio.h>
#include <stdlib.h>

void output(int* result, int result_size) {
    for (int i = 0; i < result_size; i++) {
        printf("%d ", result[i]);
    }
    printf("\n");
    return;
}
int* decode(int* encoded, int encodedSize, int first, int* returnSize){
    int* result = (int*)malloc(sizeof(int)*(encodedSize+1));
    result[0]=first;
    for(int i=0;i<encodedSize;i++){
        result[i+1]=result[i] ^ encoded[i];
    }
    *returnSize = encodedSize+1;
    return result;
}
int main() 
{
    int num1[] = {1,2,3};
    int num2[] = {6,2,7,3};
    int num3[] = {5,6,7,8,9};
    int num1_size = 3;
    int num2_size = 4;
    int num3_size = 5;
    int first_1 = 1;
    int first_2 = 4;
    int first_3 = 2;
    int *result = NULL;
    int result_size = 0;
    result = decode(num1, num1_size, first_1, &result_size);
    output(result, result_size);
    result = decode(num2, num2_size, first_2, &result_size);
    output(result, result_size);
    result = decode(num3, num3_size, first_3, &result_size);
    output(result, result_size);
}

### handwritten assemble code
I use buffer for store result, and try to keep same execution step.

Improvement:
1. Replace counter with arry_size + arry[0] ,so I don't need counter++ in loop

int arry[3] = [0,0,0];
int arry_size = 3

int i = 0;
while (i<0) {
    do something
    i++;
} 

*end = arry + arry_size
while (arry != end) {
    do something
}

2. Change branch condition so i don't use j label

#before, using beq and j instruction
label:
    beq a0, a1, exit
    do something
    j label
    
exit:
    do something

#after, only use bne instruction
label:
    do something
    bne a0, a1, label
    
exit:
    do something

execution step : load q1 -> decode.L1 -> decode.L2 -> index_zero -> print -> return -> load q2......

.data
buffer: .word 0,0,0,0,0,0,0
num1: .word 1,2,3
numSize1: .word 3
first1: .word 1
num2: .word 6,2,7,3
numSize2: .word 4
first2: .word 4
num3: .word 5,6,7,8,9
numSize3: .word 5
first3: .word 2
nextline: .string "\n"
space: .string " "
.text
main:
    la a1, num1            #get num1 address
    la a2, numSize1        #get numSize1 address
    la a3, first1          #get first1 address
    lw a2, 0(a2)           #a2 = numSize1 = 3
    lw a3, 0(a3)           #a3 = first1 = 1
    jal ra, decode.L1      #goto decode.L1
    la a1,num2             #get num2 address
    la a2,numSize2         #get numSize2 address
    la a3,first2           #get first2 address
    lw a2,0(a2)            #a2 = numSize2 = 4
    lw a3,0(a3)            #a3 = first2 = 4
    jal ra,decode.L1       #goto decode.L1
    la a1,num3             #get num3 address
    la a2,numSize3         #get numSize3 address
    la a3,first3           #get first3 address
    lw a2,0(a2)            #a2 = numSize3 = 5
    lw a3,0(a3)            #a3 = first3 = 2
    jal ra,decode.L1       #goto decode.L1
    li a7,10
    ecall
decode.L1:
    slli a2, a2, 2         #a2 =  bufferSize * 4(size of int)
    la a0, buffer          #get address buffer
    sw a3,0(a0)            #result[0] = first
    add t0, a1, a2         #a2 = num end = num[0] + 4 * numsize
decode.L2:                 #loop
    lw t1,0(a1)            #get num1[i]
    addi a0, a0, 4         #result++
    xor a3, a3, t1         #a3 = result[i] ^ num[i]
    addi a1, a1, 4         #num++
    sw a3,0(a0)            #result[i + 1] = result[i] ^ num[i]
    bne t0, a1, decode.L2  #if num end == num[i] then goto index_zero
index_zero:
    la a1, buffer          #a1 = address of buffer[0]
    addi a2, a2, 4         #bufferSize = numSize + 1
    add t0, a1, a2         #t0 = buffer end = buffer[0] + 4 * buffersize
print:
    lw a0,0(a1)            #get result[i]
    addi a1, a1, 4         #buffer++
    li a7,1                #systemcall print
    ecall                  #execute
    la a0, space           #print " "
    li a7, 4
    ecall
    bne a1,t0,print        #if a1 not equal buffer end goto print
    la a0, nextline        #nextline
    li a7,4
    ecall
    j ra

output

## compare original code and My_Improvement
### performance on Ripes
original code
My_Improvement
i got fewer execution cycle and fewer Instruction Retired

### TO DO
- [ ] Making original and handwritten assemble code can implement on rv32emu
## result with GNU Toolchain
### O0
#### compile

$ iscv-none-elf-gcc -march=rv32i -mabi=ilp32 -O0 -o temp.elf temp.c
#### objdump
the assnmble code is too long to show, so I only show about main, decode and output function

$ riscv-none-elf-objdump -d temp.elf

00010184 <output>:
   10184:	fd010113          	addi	sp,sp,-48
   10188:	02112623          	sw	ra,44(sp)
   1018c:	02812423          	sw	s0,40(sp)
   10190:	03010413          	addi	s0,sp,48
   10194:	fca42e23          	sw	a0,-36(s0)
   10198:	fcb42c23          	sw	a1,-40(s0)
   1019c:	fe042623          	sw	zero,-20(s0)
   101a0:	0340006f          	j	101d4 <output+0x50>
   101a4:	fec42783          	lw	a5,-20(s0)
   101a8:	00279793          	slli	a5,a5,0x2
   101ac:	fdc42703          	lw	a4,-36(s0)
   101b0:	00f707b3          	add	a5,a4,a5
   101b4:	0007a783          	lw	a5,0(a5)
   101b8:	00078593          	mv	a1,a5
   101bc:	000227b7          	lui	a5,0x22
   101c0:	ad078513          	addi	a0,a5,-1328 # 21ad0 <__clzsi2+0x8c>
   101c4:	435000ef          	jal	ra,10df8 <printf>
   101c8:	fec42783          	lw	a5,-20(s0)
   101cc:	00178793          	addi	a5,a5,1
   101d0:	fef42623          	sw	a5,-20(s0)
   101d4:	fec42703          	lw	a4,-20(s0)
   101d8:	fd842783          	lw	a5,-40(s0)
   101dc:	fcf744e3          	blt	a4,a5,101a4 <output+0x20>
   101e0:	00a00513          	li	a0,10
   101e4:	469000ef          	jal	ra,10e4c <putchar>
   101e8:	00000013          	nop
   101ec:	02c12083          	lw	ra,44(sp)
   101f0:	02812403          	lw	s0,40(sp)
   101f4:	03010113          	addi	sp,sp,48
   101f8:	00008067          	ret

000101fc <decode>:
   101fc:	fd010113          	addi	sp,sp,-48
   10200:	02112623          	sw	ra,44(sp)
   10204:	02812423          	sw	s0,40(sp)
   10208:	03010413          	addi	s0,sp,48
   1020c:	fca42e23          	sw	a0,-36(s0)
   10210:	fcb42c23          	sw	a1,-40(s0)
   10214:	fcc42a23          	sw	a2,-44(s0)
   10218:	fcd42823          	sw	a3,-48(s0)
   1021c:	fd842783          	lw	a5,-40(s0)
   10220:	00178793          	addi	a5,a5,1
   10224:	00279793          	slli	a5,a5,0x2
   10228:	00078513          	mv	a0,a5
   1022c:	308000ef          	jal	ra,10534 <malloc>
   10230:	00050793          	mv	a5,a0
   10234:	fef42423          	sw	a5,-24(s0)
   10238:	fe842783          	lw	a5,-24(s0)
   1023c:	fd442703          	lw	a4,-44(s0)
   10240:	00e7a023          	sw	a4,0(a5)
   10244:	fe042623          	sw	zero,-20(s0)
   10248:	0540006f          	j	1029c <decode+0xa0>
   1024c:	fec42783          	lw	a5,-20(s0)
   10250:	00279793          	slli	a5,a5,0x2
   10254:	fe842703          	lw	a4,-24(s0)
   10258:	00f707b3          	add	a5,a4,a5
   1025c:	0007a683          	lw	a3,0(a5)
   10260:	fec42783          	lw	a5,-20(s0)
   10264:	00279793          	slli	a5,a5,0x2
   10268:	fdc42703          	lw	a4,-36(s0)
   1026c:	00f707b3          	add	a5,a4,a5
   10270:	0007a703          	lw	a4,0(a5)
   10274:	fec42783          	lw	a5,-20(s0)
   10278:	00178793          	addi	a5,a5,1
   1027c:	00279793          	slli	a5,a5,0x2
   10280:	fe842603          	lw	a2,-24(s0)
   10284:	00f607b3          	add	a5,a2,a5
   10288:	00e6c733          	xor	a4,a3,a4
   1028c:	00e7a023          	sw	a4,0(a5)
   10290:	fec42783          	lw	a5,-20(s0)
   10294:	00178793          	addi	a5,a5,1
   10298:	fef42623          	sw	a5,-20(s0)
   1029c:	fec42703          	lw	a4,-20(s0)
   102a0:	fd842783          	lw	a5,-40(s0)
   102a4:	faf744e3          	blt	a4,a5,1024c <decode+0x50>
   102a8:	fd842783          	lw	a5,-40(s0)
   102ac:	00178713          	addi	a4,a5,1
   102b0:	fd042783          	lw	a5,-48(s0)
   102b4:	00e7a023          	sw	a4,0(a5)
   102b8:	fe842783          	lw	a5,-24(s0)
   102bc:	00078513          	mv	a0,a5
   102c0:	02c12083          	lw	ra,44(sp)
   102c4:	02812403          	lw	s0,40(sp)
   102c8:	03010113          	addi	sp,sp,48
   102cc:	00008067          	ret


000102d0 <main>:
   102d0:	fa010113          	addi	sp,sp,-96
   102d4:	04112e23          	sw	ra,92(sp)
   102d8:	04812c23          	sw	s0,88(sp)
   102dc:	06010413          	addi	s0,sp,96
   102e0:	00100793          	li	a5,1
   102e4:	fcf42423          	sw	a5,-56(s0)
   102e8:	00200793          	li	a5,2
   102ec:	fcf42623          	sw	a5,-52(s0)
   102f0:	00300793          	li	a5,3
   102f4:	fcf42823          	sw	a5,-48(s0)
   102f8:	000227b7          	lui	a5,0x22
   102fc:	ad478793          	addi	a5,a5,-1324 # 21ad4 <__clzsi2+0x90>
   10300:	0007a603          	lw	a2,0(a5)
   10304:	0047a683          	lw	a3,4(a5)
   10308:	0087a703          	lw	a4,8(a5)
   1030c:	00c7a783          	lw	a5,12(a5)
   10310:	fac42c23          	sw	a2,-72(s0)
   10314:	fad42e23          	sw	a3,-68(s0)
   10318:	fce42023          	sw	a4,-64(s0)
   1031c:	fcf42223          	sw	a5,-60(s0)
   10320:	000227b7          	lui	a5,0x22
   10324:	ae478793          	addi	a5,a5,-1308 # 21ae4 <__clzsi2+0xa0>
   10328:	0007a583          	lw	a1,0(a5)
   1032c:	0047a603          	lw	a2,4(a5)
   10330:	0087a683          	lw	a3,8(a5)
   10334:	00c7a703          	lw	a4,12(a5)
   10338:	0107a783          	lw	a5,16(a5)
   1033c:	fab42223          	sw	a1,-92(s0)
   10340:	fac42423          	sw	a2,-88(s0)
   10344:	fad42623          	sw	a3,-84(s0)
   10348:	fae42823          	sw	a4,-80(s0)
   1034c:	faf42a23          	sw	a5,-76(s0)
   10350:	00300793          	li	a5,3
   10354:	fef42623          	sw	a5,-20(s0)
   10358:	00400793          	li	a5,4
   1035c:	fef42423          	sw	a5,-24(s0)
   10360:	00500793          	li	a5,5
   10364:	fef42223          	sw	a5,-28(s0)
   10368:	00100793          	li	a5,1
   1036c:	fef42023          	sw	a5,-32(s0)
   10370:	00400793          	li	a5,4
   10374:	fcf42e23          	sw	a5,-36(s0)
   10378:	00200793          	li	a5,2
   1037c:	fcf42c23          	sw	a5,-40(s0)
   10380:	fc042a23          	sw	zero,-44(s0)
   10384:	fa042023          	sw	zero,-96(s0)
   10388:	fa040713          	addi	a4,s0,-96
   1038c:	fc840793          	addi	a5,s0,-56
   10390:	00070693          	mv	a3,a4
   10394:	fe042603          	lw	a2,-32(s0)
   10398:	fec42583          	lw	a1,-20(s0)
   1039c:	00078513          	mv	a0,a5
   103a0:	e5dff0ef          	jal	ra,101fc <decode>
   103a4:	fca42a23          	sw	a0,-44(s0)
   103a8:	fa042783          	lw	a5,-96(s0)
   103ac:	00078593          	mv	a1,a5
   103b0:	fd442503          	lw	a0,-44(s0)
   103b4:	dd1ff0ef          	jal	ra,10184 <output>
   103b8:	fa040713          	addi	a4,s0,-96
   103bc:	fb840793          	addi	a5,s0,-72
   103c0:	00070693          	mv	a3,a4
   103c4:	fdc42603          	lw	a2,-36(s0)
   103c8:	fe842583          	lw	a1,-24(s0)
   103cc:	00078513          	mv	a0,a5
   103d0:	e2dff0ef          	jal	ra,101fc <decode>
   103d4:	fca42a23          	sw	a0,-44(s0)
   103d8:	fa042783          	lw	a5,-96(s0)
   103dc:	00078593          	mv	a1,a5
   103e0:	fd442503          	lw	a0,-44(s0)
   103e4:	da1ff0ef          	jal	ra,10184 <output>
   103e8:	fa040713          	addi	a4,s0,-96
   103ec:	fa440793          	addi	a5,s0,-92
   103f0:	00070693          	mv	a3,a4
   103f4:	fd842603          	lw	a2,-40(s0)
   103f8:	fe442583          	lw	a1,-28(s0)
   103fc:	00078513          	mv	a0,a5
   10400:	dfdff0ef          	jal	ra,101fc <decode>
   10404:	fca42a23          	sw	a0,-44(s0)
   10408:	fa042783          	lw	a5,-96(s0)
   1040c:	00078593          	mv	a1,a5
   10410:	fd442503          	lw	a0,-44(s0)
   10414:	d71ff0ef          	jal	ra,10184 <output>
   10418:	00000793          	li	a5,0
   1041c:	00078513          	mv	a0,a5
   10420:	05c12083          	lw	ra,92(sp)
   10424:	05812403          	lw	s0,88(sp)
   10428:	06010113          	addi	sp,sp,96
   1042c:	00008067          	ret

observation

main function

> satck use : 96 (bytes)
> using register a0 ~ a5, s0, ra
saving all variable(like nums1[0], nums1[1], first_1......) in stack
using s0 for stack address location
using a4, a5 for get address or load value from stack
using a0, a1, a2, a3 for pass value to deoce function
using a0, a1 for pass value to output function
passing num1~3 with address of stack rather than using address at data section

deoce function
> satck use : 48 (bytes)
using register: a0 ~ a5, s0, ra, s0, sp
using a5 for store address from malloc
output function
> satck use : 48 (bytes)
using register: a0 ~ a5, s0, ra, s0, sp

rv32emu --stats

$ $HOME/rv32emu/build/rv32emu --stats temp.elf

1 0 2 1 
4 2 0 7 4 
2 7 1 6 14 7 
inferior exit code 0
CSR cycle count: 10767

readelf

$ riscv-none-elf-readelf -h temp.elf

ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00 
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              EXEC (Executable file)
  Machine:                           RISC-V
  Version:                           0x1
  Entry point address:               0x100dc
  Start of program headers:          52 (bytes into file)
  Start of section headers:          95236 (bytes into file)
  Flags:                             0x0
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         3
  Size of section headers:           40 (bytes)
  Number of section headers:         15
  Section header string table index: 14

elf size

$ riscv-none-elf-size temp.elf 
   text	   data	    bss	    dec	    hex	filename
  75776	   2816	    812	  79404	  1362c	temp.elf

### Ofast
#### compile

riscv-none-elf-gcc -march=rv32i -mabi=ilp32 -Ofast -o temp_fast.elf temp.c

#### objdump

$ riscv-none-elf-objdump -d temp_fast.elf

000100c4 <main>:
   100c4:	000227b7          	lui	a5,0x22
   100c8:	a0478793          	addi	a5,a5,-1532 # 21a04 <__clzsi2+0x90>
   100cc:	fb010113          	addi	sp,sp,-80
   100d0:	0007af83          	lw	t6,0(a5)
   100d4:	0047af03          	lw	t5,4(a5)
   100d8:	0087ae83          	lw	t4,8(a5)
   100dc:	00c7ae03          	lw	t3,12(a5)
   100e0:	0107a303          	lw	t1,16(a5)
   100e4:	0147a883          	lw	a7,20(a5)
   100e8:	0187a803          	lw	a6,24(a5)
   100ec:	01c7a703          	lw	a4,28(a5)
   100f0:	00100293          	li	t0,1
   100f4:	0207a783          	lw	a5,32(a5)
   100f8:	00512823          	sw	t0,16(sp)
   100fc:	00200293          	li	t0,2
   10100:	00c10693          	addi	a3,sp,12
   10104:	00100613          	li	a2,1
   10108:	00512a23          	sw	t0,20(sp)
   1010c:	00300593          	li	a1,3
   10110:	00300293          	li	t0,3
   10114:	01010513          	addi	a0,sp,16
   10118:	04112623          	sw	ra,76(sp)
   1011c:	00512c23          	sw	t0,24(sp)
   10120:	01f12e23          	sw	t6,28(sp)
   10124:	03e12023          	sw	t5,32(sp)
   10128:	03d12223          	sw	t4,36(sp)
   1012c:	03c12423          	sw	t3,40(sp)
   10130:	02612623          	sw	t1,44(sp)
   10134:	03112823          	sw	a7,48(sp)
   10138:	03012a23          	sw	a6,52(sp)
   1013c:	02e12c23          	sw	a4,56(sp)
   10140:	02f12e23          	sw	a5,60(sp)
   10144:	00012623          	sw	zero,12(sp)
   10148:	180000ef          	jal	ra,102c8 <decode>
   1014c:	00c12583          	lw	a1,12(sp)
   10150:	10c000ef          	jal	ra,1025c <output>
   10154:	00c10693          	addi	a3,sp,12
   10158:	00400613          	li	a2,4
   1015c:	00400593          	li	a1,4
   10160:	01c10513          	addi	a0,sp,28
   10164:	164000ef          	jal	ra,102c8 <decode>
   10168:	00c12583          	lw	a1,12(sp)
   1016c:	0f0000ef          	jal	ra,1025c <output>
   10170:	00c10693          	addi	a3,sp,12
   10174:	00200613          	li	a2,2
   10178:	00500593          	li	a1,5
   1017c:	02c10513          	addi	a0,sp,44
   10180:	148000ef          	jal	ra,102c8 <decode>
   10184:	00c12583          	lw	a1,12(sp)
   10188:	0d4000ef          	jal	ra,1025c <output>
   1018c:	04c12083          	lw	ra,76(sp)
   10190:	00000513          	li	a0,0
   10194:	05010113          	addi	sp,sp,80
   10198:	00008067          	ret
   
0001025c <output>:
   1025c:	06b05263          	blez	a1,102c0 <output+0x64>
   10260:	fe010113          	addi	sp,sp,-32
   10264:	00812c23          	sw	s0,24(sp)
   10268:	00912a23          	sw	s1,20(sp)
   1026c:	01212823          	sw	s2,16(sp)
   10270:	01312623          	sw	s3,12(sp)
   10274:	00112e23          	sw	ra,28(sp)
   10278:	00058913          	mv	s2,a1
   1027c:	00050413          	mv	s0,a0
   10280:	00000493          	li	s1,0
   10284:	000229b7          	lui	s3,0x22
   10288:	00042583          	lw	a1,0(s0)
   1028c:	a0098513          	addi	a0,s3,-1536 # 21a00 <__clzsi2+0x8c>
   10290:	00148493          	addi	s1,s1,1
   10294:	295000ef          	jal	ra,10d28 <printf>
   10298:	00440413          	addi	s0,s0,4
   1029c:	fe9916e3          	bne	s2,s1,10288 <output+0x2c>
   102a0:	01812403          	lw	s0,24(sp)
   102a4:	01c12083          	lw	ra,28(sp)
   102a8:	01412483          	lw	s1,20(sp)
   102ac:	01012903          	lw	s2,16(sp)
   102b0:	00c12983          	lw	s3,12(sp)
   102b4:	00a00513          	li	a0,10
   102b8:	02010113          	addi	sp,sp,32
   102bc:	2c10006f          	j	10d7c <putchar>
   102c0:	00a00513          	li	a0,10
   102c4:	2b90006f          	j	10d7c <putchar>

000102c8 <decode>:
   102c8:	fe010113          	addi	sp,sp,-32
   102cc:	01512223          	sw	s5,4(sp)
   102d0:	00158a93          	addi	s5,a1,1
   102d4:	01212823          	sw	s2,16(sp)
   102d8:	002a9913          	slli	s2,s5,0x2
   102dc:	00812c23          	sw	s0,24(sp)
   102e0:	00050413          	mv	s0,a0
   102e4:	00090513          	mv	a0,s2
   102e8:	00912a23          	sw	s1,20(sp)
   102ec:	01312623          	sw	s3,12(sp)
   102f0:	01412423          	sw	s4,8(sp)
   102f4:	00060493          	mv	s1,a2
   102f8:	00112e23          	sw	ra,28(sp)
   102fc:	00058993          	mv	s3,a1
   10300:	00068a13          	mv	s4,a3
   10304:	160000ef          	jal	ra,10464 <malloc>
   10308:	00952023          	sw	s1,0(a0)
   1030c:	03305663          	blez	s3,10338 <decode+0x70>
   10310:	ffc90593          	addi	a1,s2,-4
   10314:	00040793          	mv	a5,s0
   10318:	00450713          	addi	a4,a0,4
   1031c:	00b405b3          	add	a1,s0,a1
   10320:	0007a603          	lw	a2,0(a5)
   10324:	00470713          	addi	a4,a4,4
   10328:	00478793          	addi	a5,a5,4
   1032c:	00c4c4b3          	xor	s1,s1,a2
   10330:	fe972e23          	sw	s1,-4(a4)
   10334:	fef596e3          	bne	a1,a5,10320 <decode+0x58>
   10338:	01c12083          	lw	ra,28(sp)
   1033c:	01812403          	lw	s0,24(sp)
   10340:	015a2023          	sw	s5,0(s4)
   10344:	01412483          	lw	s1,20(sp)
   10348:	01012903          	lw	s2,16(sp)
   1034c:	00c12983          	lw	s3,12(sp)
   10350:	00812a03          	lw	s4,8(sp)
   10354:	00412a83          	lw	s5,4(sp)
   10358:	02010113          	addi	sp,sp,32
   1035c:	00008067          	ret

#### observation

main function
> satck use : 80 (bytes)
saving all variable(like nums1[0], nums1[1], first_1......) in stack
using register t0 ~t6, a0 ~ a6, s0, ra
only read data address in once, because in data section nums2 and nums3 is continuous
saving num1, num2, num3 in stack is continuous
using sp for stack address location

deoce function
> satck use : 32 (bytes)
using register: a0 ~ a5, s0 ~ s4, ra, s0, sp
using address != address_end for loop

output function
> satck use : 32 (bytes)
using register: a0 ~ a5, s1 ~ s3, ra, sp

using more register replace stack

#### rv32emu --stats

$ $HOME/rv32emu/build/rv32emu --stats temp_fast.elf

1 0 2 1 
4 2 0 7 4 
2 7 1 6 14 7 
inferior exit code 0
CSR cycle count: 10403

#### readelf

$ riscv-none-elf-readelf -h temp_fast.elf

ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00 
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              EXEC (Executable file)
  Machine:                           RISC-V
  Version:                           0x1
  Entry point address:               0x101b4
  Start of program headers:          52 (bytes into file)
  Start of section headers:          95252 (bytes into file)
  Flags:                             0x0
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         3
  Size of section headers:           40 (bytes)
  Number of section headers:         15
  Section header string table index: 14

#### lf size

$ riscv-none-elf-size temp_fast.elf 

   text	   data	    bss	    dec	    hex	filename
  75568	   2816	    812	  79196	  1355c	temp_fast.elf

### Os
#### compile

riscv-none-elf-gcc -march=rv32i -mabi=ilp32 -Os -o temp_store.elf temp.c

#### objdump

$ riscv-none-elf-objdump -d temp_store.elf

000100c4 <main>:
   100c4:	fb010113          	addi	sp,sp,-80
   100c8:	000225b7          	lui	a1,0x22
   100cc:	04812423          	sw	s0,72(sp)
   100d0:	00c00613          	li	a2,12
   100d4:	9cc58413          	addi	s0,a1,-1588 # 219cc <__clzsi2+0x8c>
   100d8:	01010513          	addi	a0,sp,16
   100dc:	9cc58593          	addi	a1,a1,-1588
   100e0:	04112623          	sw	ra,76(sp)
   100e4:	2e5000ef          	jal	ra,10bc8 <memcpy>
   100e8:	00c40593          	addi	a1,s0,12
   100ec:	01000613          	li	a2,16
   100f0:	01c10513          	addi	a0,sp,28
   100f4:	2d5000ef          	jal	ra,10bc8 <memcpy>
   100f8:	01c40593          	addi	a1,s0,28
   100fc:	01400613          	li	a2,20
   10100:	02c10513          	addi	a0,sp,44
   10104:	2c5000ef          	jal	ra,10bc8 <memcpy>
   10108:	00c10693          	addi	a3,sp,12
   1010c:	00100613          	li	a2,1
   10110:	00300593          	li	a1,3
   10114:	01010513          	addi	a0,sp,16
   10118:	00012623          	sw	zero,12(sp)
   1011c:	180000ef          	jal	ra,1029c <decode>
   10120:	00c12583          	lw	a1,12(sp)
   10124:	110000ef          	jal	ra,10234 <output>
   10128:	00c10693          	addi	a3,sp,12
   1012c:	00400613          	li	a2,4
   10130:	00400593          	li	a1,4
   10134:	01c10513          	addi	a0,sp,28
   10138:	164000ef          	jal	ra,1029c <decode>
   1013c:	00c12583          	lw	a1,12(sp)
   10140:	0f4000ef          	jal	ra,10234 <output>
   10144:	00c10693          	addi	a3,sp,12
   10148:	00200613          	li	a2,2
   1014c:	00500593          	li	a1,5
   10150:	02c10513          	addi	a0,sp,44
   10154:	148000ef          	jal	ra,1029c <decode>
   10158:	00c12583          	lw	a1,12(sp)
   1015c:	0d8000ef          	jal	ra,10234 <output>
   10160:	04c12083          	lw	ra,76(sp)
   10164:	04812403          	lw	s0,72(sp)
   10168:	00000513          	li	a0,0
   1016c:	05010113          	addi	sp,sp,80
   10170:	00008067          	ret

0010234 <output>:
   10234:	fe010113          	addi	sp,sp,-32
   10238:	00812c23          	sw	s0,24(sp)
   1023c:	00912a23          	sw	s1,20(sp)
   10240:	01212823          	sw	s2,16(sp)
   10244:	01312623          	sw	s3,12(sp)
   10248:	00112e23          	sw	ra,28(sp)
   1024c:	00050913          	mv	s2,a0
   10250:	00058493          	mv	s1,a1
   10254:	00000413          	li	s0,0
   10258:	000229b7          	lui	s3,0x22
   1025c:	02944263          	blt	s0,s1,10280 <output+0x4c>
   10260:	01812403          	lw	s0,24(sp)
   10264:	01c12083          	lw	ra,28(sp)
   10268:	01412483          	lw	s1,20(sp)
   1026c:	01012903          	lw	s2,16(sp)
   10270:	00c12983          	lw	s3,12(sp)
   10274:	00a00513          	li	a0,10
   10278:	02010113          	addi	sp,sp,32
   1027c:	4710006f          	j	10eec <putchar>
   10280:	00241793          	slli	a5,s0,0x2
   10284:	00f907b3          	add	a5,s2,a5
   10288:	0007a583          	lw	a1,0(a5)
   1028c:	9c898513          	addi	a0,s3,-1592 # 219c8 <__clzsi2+0x88>
   10290:	00140413          	addi	s0,s0,1
   10294:	405000ef          	jal	ra,10e98 <printf>
   10298:	fc5ff06f          	j	1025c <output+0x28>

0001029c <decode>:
   1029c:	fe010113          	addi	sp,sp,-32
   102a0:	01412423          	sw	s4,8(sp)
   102a4:	00158a13          	addi	s4,a1,1
   102a8:	01212823          	sw	s2,16(sp)
   102ac:	00050913          	mv	s2,a0
   102b0:	002a1513          	slli	a0,s4,0x2
   102b4:	00812c23          	sw	s0,24(sp)
   102b8:	00912a23          	sw	s1,20(sp)
   102bc:	01312623          	sw	s3,12(sp)
   102c0:	00112e23          	sw	ra,28(sp)
   102c4:	00060993          	mv	s3,a2
   102c8:	00058413          	mv	s0,a1
   102cc:	00068493          	mv	s1,a3
   102d0:	160000ef          	jal	ra,10430 <malloc>
   102d4:	01352023          	sw	s3,0(a0)
   102d8:	00050713          	mv	a4,a0
   102dc:	00000793          	li	a5,0
   102e0:	00470713          	addi	a4,a4,4
   102e4:	0287c463          	blt	a5,s0,1030c <decode+0x70>
   102e8:	01c12083          	lw	ra,28(sp)
   102ec:	01812403          	lw	s0,24(sp)
   102f0:	0144a023          	sw	s4,0(s1)
   102f4:	01012903          	lw	s2,16(sp)
   102f8:	01412483          	lw	s1,20(sp)
   102fc:	00c12983          	lw	s3,12(sp)
   10300:	00812a03          	lw	s4,8(sp)
   10304:	02010113          	addi	sp,sp,32
   10308:	00008067          	ret
   1030c:	00279693          	slli	a3,a5,0x2
   10310:	00d906b3          	add	a3,s2,a3
   10314:	0006a683          	lw	a3,0(a3)
   10318:	ffc72603          	lw	a2,-4(a4)
   1031c:	00178793          	addi	a5,a5,1
   10320:	00c6c6b3          	xor	a3,a3,a2
   10324:	00d72023          	sw	a3,0(a4)
   10328:	fb9ff06f          	j	102e0 <decode+0x44>

#### observation

main function
> satck use : 80 (bytes)
saving all variable(like nums1[0], nums1[1], first_1......) in stack
using register: a0 ~ a3, s0, ra, sp
using memcpy for store num in stack
only read data address in once, because in data section nums2 and nums3 is continuous
saving num1, num2, num3 in stack is continuous
using sp for stack address location

deoce function
> satck use : 32 (bytes)
using register: a0 ~ a3, s0 ~ s3, ra, sp
using address != address_end for loop

output function
> satck use : 32 (bytes)
using register: a0 ~ a5, s1 ~ s3, ra, sp

using more fuction replace using register
#### rv32emu --stats

$ $HOME/rv32emu/build/rv32emu --stats temp_store.elf 
 
1 0 2 1 
4 2 0 7 4 
2 7 1 6 14 7 
inferior exit code 0
CSR cycle count: 10528

readelf

$ riscv-none-elf-readelf -h temp_store.elf

ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00 
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              EXEC (Executable file)
  Machine:                           RISC-V
  Version:                           0x1
  Entry point address:               0x1018c
  Start of program headers:          52 (bytes into file)
  Start of section headers:          95252 (bytes into file)
  Flags:                             0x0
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         3
  Size of section headers:           40 (bytes)
  Number of section headers:         15
  Section header string table index: 14

elf-size

$ riscv-none-elf-size temp_store.elf

   text	   data	    bss	    dec	    hex	filename
  75532	   2816	    812	  79160	  13538	temp_store.elf

### question
#### question:
when I compare O0, Os, Ofast, i find "num1" are not store in data section when use O0 and Ofast, they use "li" instruction for get value of "num1", but using "lui" and "lw" instruction for get value of "num2" and "num3".

I using below command for getting assemble code, I don't know why compile using to way to store array in O0, Ofast.

$ riscv-none-elf-gcc -march=rv32i -mabi=ilp32 -Os -S
[c code](https://hackmd.io/7JFyoJ59R7m7aFopLthAKA?view#c1)

    int num1[] = {1,2,3};
    int num2[] = {6,2,7,3};
    int num3[] = {5,6,7,8,9};

in O0

# temp.c:23:     int num1[] = {1,2,3};
	li	a5,1		# tmp77,
	sw	a5,-56(s0)	# tmp77, num1[0]
	li	a5,2		# tmp78,
	sw	a5,-52(s0)	# tmp78, num1[1]
	li	a5,3		# tmp79,
	sw	a5,-48(s0)	# tmp79, num1[2]
# temp.c:24:     int num2[] = {6,2,7,3};
	lui	a5,%hi(.LC0)	# tmp80,
	addi	a5,a5,%lo(.LC0)	# tmp81, tmp80,
	lw	a2,0(a5)		# tmp82,
	lw	a3,4(a5)		# tmp83,
	lw	a4,8(a5)		# tmp84,
	lw	a5,12(a5)		# tmp85,
	sw	a2,-72(s0)	# tmp82, num2
	sw	a3,-68(s0)	# tmp83, num2
	sw	a4,-64(s0)	# tmp84, num2
	sw	a5,-60(s0)	# tmp85, num2

.LC0:
	.word	6
	.word	2
	.word	7
	.word	3
	.align	2
.LC1:
	.word	5
	.word	6
	.word	7
	.word	8
	.word	9
	.text
	.align	2
	.globl	main
	.type	main, @function

in Ofast

.LC0:
	.word	6
	.word	2
	.word	7
	.word	3
.LC1:
	.word	5
	.word	6
	.word	7
	.word	8
	.word	9
	.ident	"GCC: (xPack GNU RISC-V Embedded GCC x86_64) 12.2.0"

in Os

.LC0:
	.word	1
	.word	2
	.word	3
.LC1:
	.word	6
	.word	2
	.word	7
	.word	3
.LC2:
	.word	5
	.word	6
	.word	7
	.word	8
	.word	9
	.ident	"GCC: (xPack GNU RISC-V Embedded GCC x86_64) 12.2.0"