| 1 |
/* { dg-do compile } */ |
| 2 |
/* { dg-options "-march=rv64gcv -mabi=lp64d -O3 -fno-schedule-insns -fno-schedule-insns2" } */ |
| 3 |
|
| 4 |
#include "riscv_vector.h" |
| 5 |
|
| 6 |
void matrix_transpose_in_register(uint32_t* outputMat, uint32_t* inputMat) { |
| 7 |
vuint32m1_t matRegIn0 = __riscv_vle32_v_u32m1(inputMat, 4); |
| 8 |
vuint32m1_t matRegIn1 = __riscv_vle32_v_u32m1(inputMat + 4, 4); |
| 9 |
vuint32m1_t matRegIn2 = __riscv_vle32_v_u32m1(inputMat + 8, 4); |
| 10 |
vuint32m1_t matRegIn3 = __riscv_vle32_v_u32m1(inputMat + 12, 4); |
| 11 |
|
| 12 |
vbool32_t oddMask = __riscv_vreinterpret_v_u32m1_b32(__riscv_vmv_v_x_u32m1(0xaaaa, 1)); |
| 13 |
|
| 14 |
vuint32m1_t smallTransposeMat0 = __riscv_vslideup_vx_u32m1_tumu(oddMask, matRegIn0, matRegIn1, 1, 4); |
| 15 |
vuint32m1_t smallTransposeMat2 = __riscv_vslideup_vx_u32m1_tumu(oddMask, matRegIn2, matRegIn3, 1, 4); |
| 16 |
|
| 17 |
vbool32_t evenMask = __riscv_vreinterpret_v_u32m1_b32(__riscv_vmv_v_x_u32m1(0x5555, 1)); |
| 18 |
|
| 19 |
vuint32m1_t smallTransposeMat1 = __riscv_vslidedown_vx_u32m1_tumu(evenMask, matRegIn1, matRegIn0, 1, 4); |
| 20 |
vuint32m1_t smallTransposeMat3 = __riscv_vslidedown_vx_u32m1_tumu(evenMask, matRegIn3, matRegIn2, 1, 4); |
| 21 |
|
| 22 |
vuint32m1_t outMat0 = __riscv_vslideup_vx_u32m1_tu(smallTransposeMat0, smallTransposeMat2, 2, 4); |
| 23 |
vuint32m1_t outMat1 = __riscv_vslideup_vx_u32m1_tu(smallTransposeMat1, smallTransposeMat3, 2, 4); |
| 24 |
|
| 25 |
vuint32m1_t outMat2 = __riscv_vslidedown_vx_u32m1_tu(smallTransposeMat2, smallTransposeMat0, 2, 2); |
| 26 |
vuint32m1_t outMat3 = __riscv_vslidedown_vx_u32m1_tu(smallTransposeMat3, smallTransposeMat1, 2, 2); |
| 27 |
__riscv_vse32_v_u32m1(outputMat, outMat0, 4); |
| 28 |
__riscv_vse32_v_u32m1(outputMat + 4, outMat1, 4); |
| 29 |
__riscv_vse32_v_u32m1(outputMat + 8, outMat2, 4); |
| 30 |
__riscv_vse32_v_u32m1(outputMat + 12, outMat3, 4); |
| 31 |
} |
| 32 |
|
| 33 |
/* { dg-final { scan-assembler-times {vsetivli\s+zero,\s*4,\s*e32,\s*m1,\s*t[au],\s*m[au]} 2 } } */ |
| 34 |
/* { dg-final { scan-assembler-times {vsetivli\s+zero,\s*2,\s*e32,\s*m1,\s*t[au],\s*m[au]} 1 } } */ |
| 35 |
/* { dg-final { scan-assembler-times {vsetivli} 3 } } */ |
| 36 |
/* { dg-final { scan-assembler-not {vsetvli} } } */ |