OpceanAI's picture
Model card, paper, figures, logs and scripts
2049fc8 verified
Raw History Blame Contribute Delete
16.7 kB
.text
.globl vk_8_init_scaled
vk_8_init_scaled:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
lw t3, 48(t2)
lw t4, 52(t2)
lw t5, 56(t2)
lw t6, 60(t2)
1:
li a5, 0
li a6, 0
li a7, 0
li t0, 0
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a5, t3, t4
.insn r 0x0B, 5, 0, a5, t5, t6
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a6, t3, t4
.insn r 0x0B, 5, 0, a6, t5, t6
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, a7, t3, t4
.insn r 0x0B, 5, 0, a7, t5, t6
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
.insn r 0x0B, 5, 0, t0, t3, t4
.insn r 0x0B, 5, 0, t0, t5, t6
lw t1, 0(a3)
mul a5, a5, t1
mulh a5, a5, a4
sw a5, 0(a0)
lw t1, 4(a3)
mul a6, a6, t1
mulh a6, a6, a4
sw a6, 4(a0)
lw t1, 8(a3)
mul a7, a7, t1
mulh a7, a7, a4
sw a7, 8(a0)
lw t1, 12(a3)
mul t0, t0, t1
mulh t0, t0, a4
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 32
addi a3, a3, 16
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_8_init_raw
vk_8_init_raw:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
lw t3, 48(t2)
lw t4, 52(t2)
lw t5, 56(t2)
lw t6, 60(t2)
1:
li a5, 0
li a6, 0
li a7, 0
li t0, 0
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a5, t3, t4
.insn r 0x0B, 5, 0, a5, t5, t6
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a6, t3, t4
.insn r 0x0B, 5, 0, a6, t5, t6
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, a7, t3, t4
.insn r 0x0B, 5, 0, a7, t5, t6
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
.insn r 0x0B, 5, 0, t0, t3, t4
.insn r 0x0B, 5, 0, t0, t5, t6
sw a5, 0(a0)
sw a6, 4(a0)
sw a7, 8(a0)
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 32
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_8_acc_scaled
vk_8_acc_scaled:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
lw t3, 48(t2)
lw t4, 52(t2)
lw t5, 56(t2)
lw t6, 60(t2)
1:
lw a5, 0(a0)
lw a6, 4(a0)
lw a7, 8(a0)
lw t0, 12(a0)
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a5, t3, t4
.insn r 0x0B, 5, 0, a5, t5, t6
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a6, t3, t4
.insn r 0x0B, 5, 0, a6, t5, t6
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, a7, t3, t4
.insn r 0x0B, 5, 0, a7, t5, t6
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
.insn r 0x0B, 5, 0, t0, t3, t4
.insn r 0x0B, 5, 0, t0, t5, t6
lw t1, 0(a3)
mul a5, a5, t1
mulh a5, a5, a4
sw a5, 0(a0)
lw t1, 4(a3)
mul a6, a6, t1
mulh a6, a6, a4
sw a6, 4(a0)
lw t1, 8(a3)
mul a7, a7, t1
mulh a7, a7, a4
sw a7, 8(a0)
lw t1, 12(a3)
mul t0, t0, t1
mulh t0, t0, a4
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 32
addi a3, a3, 16
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_8_acc_raw
vk_8_acc_raw:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
lw t3, 48(t2)
lw t4, 52(t2)
lw t5, 56(t2)
lw t6, 60(t2)
1:
lw a5, 0(a0)
lw a6, 4(a0)
lw a7, 8(a0)
lw t0, 12(a0)
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a5, t3, t4
.insn r 0x0B, 5, 0, a5, t5, t6
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a6, t3, t4
.insn r 0x0B, 5, 0, a6, t5, t6
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, a7, t3, t4
.insn r 0x0B, 5, 0, a7, t5, t6
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
.insn r 0x0B, 5, 0, t0, t3, t4
.insn r 0x0B, 5, 0, t0, t5, t6
sw a5, 0(a0)
sw a6, 4(a0)
sw a7, 8(a0)
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 32
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_6_init_scaled
vk_6_init_scaled:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
1:
li a5, 0
li a6, 0
li a7, 0
li t0, 0
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
lw t1, 0(a3)
mul a5, a5, t1
mulh a5, a5, a4
sw a5, 0(a0)
lw t1, 4(a3)
mul a6, a6, t1
mulh a6, a6, a4
sw a6, 4(a0)
lw t1, 8(a3)
mul a7, a7, t1
mulh a7, a7, a4
sw a7, 8(a0)
lw t1, 12(a3)
mul t0, t0, t1
mulh t0, t0, a4
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 24
addi a3, a3, 16
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_6_init_raw
vk_6_init_raw:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
1:
li a5, 0
li a6, 0
li a7, 0
li t0, 0
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
sw a5, 0(a0)
sw a6, 4(a0)
sw a7, 8(a0)
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 24
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_6_acc_scaled
vk_6_acc_scaled:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
1:
lw a5, 0(a0)
lw a6, 4(a0)
lw a7, 8(a0)
lw t0, 12(a0)
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
lw t1, 0(a3)
mul a5, a5, t1
mulh a5, a5, a4
sw a5, 0(a0)
lw t1, 4(a3)
mul a6, a6, t1
mulh a6, a6, a4
sw a6, 4(a0)
lw t1, 8(a3)
mul a7, a7, t1
mulh a7, a7, a4
sw a7, 8(a0)
lw t1, 12(a3)
mul t0, t0, t1
mulh t0, t0, a4
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 24
addi a3, a3, 16
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret
.globl vk_6_acc_raw
vk_6_acc_raw:
addi sp, sp, -48
sw s0, 0(sp)
sw s1, 4(sp)
sw s2, 8(sp)
sw s3, 12(sp)
sw s4, 16(sp)
sw s5, 20(sp)
sw s6, 24(sp)
sw s7, 28(sp)
sw s8, 32(sp)
sw s9, 36(sp)
sw s10, 40(sp)
sw s11, 44(sp)
lui t2, %hi(vk_x)
lw t2, %lo(vk_x)(t2)
lw s0, 0(t2)
lw s1, 4(t2)
lw s2, 8(t2)
lw s3, 12(t2)
lw s4, 16(t2)
lw s5, 20(t2)
lw s6, 24(t2)
lw s7, 28(t2)
lw s8, 32(t2)
lw s9, 36(t2)
lw s10, 40(t2)
lw s11, 44(t2)
1:
lw a5, 0(a0)
lw a6, 4(a0)
lw a7, 8(a0)
lw t0, 12(a0)
srli t1, a1, 15
.insn r 0x0B, 6, 1, zero, t1, zero
slli t2, a1, 17
srli t2, t2, 17
.insn r 0x0B, 6, 0, zero, t2, zero
.insn r 0x0B, 5, 0, a5, s0, s1
.insn r 0x0B, 5, 0, a5, s2, s3
.insn r 0x0B, 5, 0, a5, s4, s5
.insn r 0x0B, 5, 0, a5, s6, s7
.insn r 0x0B, 5, 0, a5, s8, s9
.insn r 0x0B, 5, 0, a5, s10, s11
.insn r 0x0B, 5, 0, a6, s0, s1
.insn r 0x0B, 5, 0, a6, s2, s3
.insn r 0x0B, 5, 0, a6, s4, s5
.insn r 0x0B, 5, 0, a6, s6, s7
.insn r 0x0B, 5, 0, a6, s8, s9
.insn r 0x0B, 5, 0, a6, s10, s11
.insn r 0x0B, 5, 0, a7, s0, s1
.insn r 0x0B, 5, 0, a7, s2, s3
.insn r 0x0B, 5, 0, a7, s4, s5
.insn r 0x0B, 5, 0, a7, s6, s7
.insn r 0x0B, 5, 0, a7, s8, s9
.insn r 0x0B, 5, 0, a7, s10, s11
.insn r 0x0B, 5, 0, t0, s0, s1
.insn r 0x0B, 5, 0, t0, s2, s3
.insn r 0x0B, 5, 0, t0, s4, s5
.insn r 0x0B, 5, 0, t0, s6, s7
.insn r 0x0B, 5, 0, t0, s8, s9
.insn r 0x0B, 5, 0, t0, s10, s11
sw a5, 0(a0)
sw a6, 4(a0)
sw a7, 8(a0)
sw t0, 12(a0)
addi a0, a0, 16
addi a1, a1, 24
addi a2, a2, -4
bnez a2, 1b
lw s0, 0(sp)
lw s1, 4(sp)
lw s2, 8(sp)
lw s3, 12(sp)
lw s4, 16(sp)
lw s5, 20(sp)
lw s6, 24(sp)
lw s7, 28(sp)
lw s8, 32(sp)
lw s9, 36(sp)
lw s10, 40(sp)
lw s11, 44(sp)
addi sp, sp, 48
ret