mirror of
https://github.com/spoopyghosty0/frameport.git
synced 2026-10-24 22:00:31 +02:00
UnicronNLandClaude Opus 5.5
4f042f7225
Interpreter + JIT for ARM, Thumb-2, VFPv4 and NEON, ARM EABI syscall translation, signals, threads and self-modifying code. Runs static bionic programs on the Frame. Tests: A64 encoders vs llvm-mc, a differential instruction fuzzer and program tests against qemu-arm. Not shipped; the README lists the steps towards running 32-bit-only APKs. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
111 lines
8.8 KiB
C
111 lines
8.8 KiB
C
/* Advanced SIMD intrinsics: hashes the results of a wide set of operations on pseudo-random inputs. */
|
|
#include <arm_neon.h>
|
|
#include <stdint.h>
|
|
#include <stdio.h>
|
|
#include <string.h>
|
|
|
|
static uint32_t h = 2166136261u;
|
|
static void mixb(const void *p, size_t n) { const uint8_t *b = p; for (size_t k = 0; k < n; k++) h = (h ^ b[k]) * 16777619u; }
|
|
#define M(v) do { __typeof__(v) _t = (v); mixb(&_t, sizeof _t); } while (0)
|
|
|
|
static uint8_t buf[256] __attribute__((aligned(16)));
|
|
static uint32_t rs = 99;
|
|
static uint32_t rnd(void) { rs ^= rs << 13; rs ^= rs >> 17; rs ^= rs << 5; return rs; }
|
|
|
|
int main(void)
|
|
{
|
|
for (int round = 0; round < 200; round++) {
|
|
for (int i = 0; i < 256; i++) buf[i] = (uint8_t)rnd();
|
|
if (round % 3 == 0) for (int i = 0; i < 256; i += 7) buf[i] = 0x80; /* saturation corner cases */
|
|
int8x16_t a8 = vld1q_s8((int8_t *)buf), b8 = vld1q_s8((int8_t *)buf + 16);
|
|
uint8x16_t ua8 = vreinterpretq_u8_s8(a8), ub8 = vreinterpretq_u8_s8(b8);
|
|
int16x8_t a16 = vld1q_s16((int16_t *)(buf + 32)), b16 = vld1q_s16((int16_t *)(buf + 48));
|
|
int32x4_t a32 = vld1q_s32((int32_t *)(buf + 64)), b32 = vld1q_s32((int32_t *)(buf + 80));
|
|
int64x2_t a64 = vld1q_s64((int64_t *)(buf + 96)), b64 = vld1q_s64((int64_t *)(buf + 112));
|
|
uint16x8_t ua16 = vreinterpretq_u16_s16(a16), ub16 = vreinterpretq_u16_s16(b16);
|
|
uint32x4_t ua32 = vreinterpretq_u32_s32(a32), ub32 = vreinterpretq_u32_s32(b32);
|
|
M(vaddq_s8(a8, b8)); M(vsubq_s16(a16, b16)); M(vmulq_s32(a32, b32)); M(vaddq_s64(a64, b64));
|
|
M(vqaddq_s8(a8, b8)); M(vqaddq_u8(ua8, ub8)); M(vqsubq_s16(a16, b16)); M(vqsubq_u32(ua32, ub32));
|
|
M(vqaddq_s64(a64, b64)); M(vqsubq_u64(vreinterpretq_u64_s64(a64), vreinterpretq_u64_s64(b64)));
|
|
M(vhaddq_s8(a8, b8)); M(vrhaddq_u16(ua16, ub16)); M(vhsubq_s32(a32, b32));
|
|
M(vabdq_s8(a8, b8)); M(vabaq_u16(ua16, ub16, ua16)); M(vmaxq_s16(a16, b16)); M(vminq_u32(ua32, ub32));
|
|
M(vpadd_s16(vget_low_s16(a16), vget_high_s16(b16))); M(vpmax_u8(vget_low_u8(ua8), vget_high_u8(ub8)));
|
|
M(vpmin_s32(vget_low_s32(a32), vget_high_s32(b32)));
|
|
M(vmlaq_s16(a16, b16, a16)); M(vmlsq_u32(ua32, ub32, ua32)); M(vmulq_p8(vreinterpretq_p8_s8(a8), vreinterpretq_p8_s8(b8)));
|
|
M(vqdmulhq_s16(a16, b16)); M(vqrdmulhq_s32(a32, b32)); M(vqdmulh_n_s16(vget_low_s16(a16), (int16_t)rnd()));
|
|
M(vshlq_s8(a8, vshrq_n_s8(b8, 4))); M(vshlq_u16(ua16, vshrq_n_s16(b16, 11))); M(vqshlq_s32(a32, vshrq_n_s32(b32, 26)));
|
|
M(vrshlq_s64(a64, vshrq_n_s64(b64, 58))); M(vqrshlq_u8(ua8, vshrq_n_s8(b8, 5)));
|
|
M(vshrq_n_u8(ua8, 3)); M(vshrq_n_s16(a16, 15)); M(vsraq_n_s32(a32, b32, 7)); M(vrshrq_n_u64(vreinterpretq_u64_s64(a64), 63));
|
|
M(vrsraq_n_s8(a8, b8, 8)); M(vshlq_n_s16(a16, 9)); M(vqshlq_n_u32(ua32, 3)); M(vqshluq_n_s8(a8, 2));
|
|
M(vsliq_n_u16(ua16, ub16, 5)); M(vsriq_n_u32(ua32, ub32, 31)); M(vsli_n_u64(vget_low_u64(vreinterpretq_u64_s64(a64)), vget_high_u64(vreinterpretq_u64_s64(b64)), 40));
|
|
M(vshrn_n_s32(a32, 9)); M(vrshrn_n_u16(ua16, 3)); M(vqshrn_n_s16(a16, 2)); M(vqrshrun_n_s32(a32, 13)); M(vqshrun_n_s64(a64, 30));
|
|
M(vshll_n_s8(vget_low_s8(a8), 3)); M(vmovl_u16(vget_high_u16(ua16))); M(vshll_n_u8(vget_low_u8(ua8), 8));
|
|
M(vmovn_s64(a64)); M(vqmovn_s32(a32)); M(vqmovn_u16(ua16)); M(vqmovun_s16(a16));
|
|
M(vaddl_s8(vget_low_s8(a8), vget_high_s8(b8))); M(vsubw_u16(ua32, vget_low_u16(ub16))); M(vaddhn_s32(a32, b32));
|
|
M(vraddhn_u64(vreinterpretq_u64_s64(a64), vreinterpretq_u64_s64(b64))); M(vrsubhn_s16(a16, b16));
|
|
M(vabdl_u8(vget_low_u8(ua8), vget_high_u8(ub8))); M(vabal_s16(a32, vget_low_s16(a16), vget_low_s16(b16)));
|
|
M(vmull_s16(vget_low_s16(a16), vget_high_s16(b16))); M(vmlal_u32(vreinterpretq_u64_s64(a64), vget_low_u32(ua32), vget_high_u32(ub32)));
|
|
M(vmlsl_s8(a16, vget_low_s8(a8), vget_low_s8(b8))); M(vqdmull_s32(vget_low_s32(a32), vget_high_s32(b32)));
|
|
M(vqdmlal_s16(a32, vget_low_s16(a16), vget_high_s16(b16))); M(vqdmlsl_s32(a64, vget_low_s32(a32), vget_low_s32(b32)));
|
|
M(vmull_p8(vreinterpret_p8_s8(vget_low_s8(a8)), vreinterpret_p8_s8(vget_high_s8(b8))));
|
|
M(vmulq_lane_s16(a16, vget_low_s16(b16), 3)); M(vmlaq_lane_s32(a32, b32, vget_high_s32(a32), 1));
|
|
M(vmull_lane_u16(vget_low_u16(ua16), vget_high_u16(ub16), 2)); M(vqdmulhq_lane_s32(a32, vget_low_s32(b32), 0));
|
|
M(vqrdmulh_lane_s16(vget_low_s16(a16), vget_high_s16(b16), 1)); M(vqdmull_lane_s16(vget_low_s16(a16), vget_low_s16(b16), 2));
|
|
M(vandq_s8(a8, b8)); M(vorrq_u16(ua16, ub16)); M(veorq_s32(a32, b32)); M(vbicq_s64(a64, b64)); M(vornq_u8(ua8, ub8));
|
|
M(vbslq_u8(ua8, ub8, vreinterpretq_u8_s16(a16))); M(vmvnq_u8(ua8));
|
|
M(vceqq_s8(a8, b8)); M(vcgtq_s16(a16, b16)); M(vcgeq_u32(ua32, ub32)); M(vcltq_s8(a8, b8)); M(vtstq_u16(ua16, ub16));
|
|
M(vceqq_s8(a8, vdupq_n_s8(0))); M(vcgtq_s32(a32, vdupq_n_s32(0)));
|
|
M(vabsq_s8(a8)); M(vnegq_s16(a16)); M(vqabsq_s32(a32)); M(vqnegq_s8(a8)); M(vclzq_u16(ua16)); M(vclsq_s32(a32)); M(vcntq_u8(ua8));
|
|
M(vpaddlq_s8(a8)); M(vpadalq_u16(ua32, ub16)); M(vpaddl_u32(vget_low_u32(ua32)));
|
|
M(vrev64q_s8(a8)); M(vrev32q_u16(ua16)); M(vrev16q_u8(ua8)); M(vrev64q_s32(a32));
|
|
M(vextq_s8(a8, b8, 5)); M(vextq_s16(a16, b16, 7)); M(vext_s32(vget_low_s32(a32), vget_high_s32(b32), 1));
|
|
M(vzipq_s8(a8, b8)); M(vuzpq_u16(ua16, ub16)); M(vtrnq_s32(a32, b32)); M(vzip_u32(vget_low_u32(ua32), vget_high_u32(ub32)));
|
|
M(vdupq_lane_s8(vget_low_s8(a8), 5)); M(vdupq_n_u16((uint16_t)rnd())); M(vdup_lane_s32(vget_high_s32(a32), 1));
|
|
uint8x8x2_t t2 = {{vget_low_u8(ua8), vget_high_u8(ub8)}};
|
|
M(vtbl2_u8(t2, vand_u8(vget_low_u8(ub8), vdup_n_u8(0x1f))));
|
|
uint8x8x4_t t4 = {{vget_low_u8(ua8), vget_high_u8(ua8), vget_low_u8(ub8), vget_high_u8(ub8)}};
|
|
M(vtbx4_u8(vget_low_u8(ua8), t4, vget_high_u8(ub8)));
|
|
M(vmovq_n_s32(-7)); M(vmov_n_u8(200)); M(vmvnq_s16(vdupq_n_s16(0x7f)));
|
|
M(vgetq_lane_s16(a16, 5)); M(vsetq_lane_u32(77, ua32, 3)); M(vgetq_lane_s64(a64, 1));
|
|
M(vrecpeq_u32(ua32)); M(vrsqrteq_u32(ua32));
|
|
/* floating point */
|
|
float fa[4], fb[4];
|
|
for (int i = 0; i < 4; i++) {
|
|
fa[i] = (float)((int32_t)rnd()) / 65536.0f;
|
|
fb[i] = (float)((int32_t)rnd()) / 1e6f;
|
|
}
|
|
if (round % 5 == 0) { fa[1] = 0.0f; fb[2] = -0.0f; fa[3] = 1e-40f; fb[0] = 3.4e38f; }
|
|
float32x4_t f1 = vld1q_f32(fa), f2 = vld1q_f32(fb);
|
|
M(vaddq_f32(f1, f2)); M(vsubq_f32(f1, f2)); M(vmulq_f32(f1, f2)); M(vmlaq_f32(f1, f2, f1)); M(vmlsq_f32(f2, f1, f2));
|
|
M(vfmaq_f32(f1, f2, f1)); M(vfmsq_f32(f2, f1, f1));
|
|
M(vabdq_f32(f1, f2)); M(vmaxq_f32(f1, f2)); M(vminq_f32(f1, f2)); M(vpadd_f32(vget_low_f32(f1), vget_high_f32(f2)));
|
|
M(vpmax_f32(vget_low_f32(f1), vget_high_f32(f2))); M(vabsq_f32(f1)); M(vnegq_f32(f2));
|
|
M(vceqq_f32(f1, f2)); M(vcgeq_f32(f1, f2)); M(vcgtq_f32(f1, f2)); M(vcageq_f32(f1, f2)); M(vcagtq_f32(f1, f2));
|
|
M(vrecpeq_f32(f1)); M(vrecpsq_f32(f1, f2)); M(vrsqrteq_f32(vabsq_f32(f1))); M(vrsqrtsq_f32(f1, f2));
|
|
M(vcvtq_s32_f32(f1)); M(vcvtq_u32_f32(f2)); M(vcvtq_f32_s32(a32)); M(vcvtq_f32_u32(ua32));
|
|
M(vcvtq_n_s32_f32(f1, 8)); M(vcvtq_n_u32_f32(f2, 31)); M(vcvtq_n_f32_s32(a32, 16)); M(vcvtq_n_f32_u32(ua32, 1));
|
|
M(vmulq_lane_f32(f1, vget_low_f32(f2), 1)); M(vmlaq_lane_f32(f1, f2, vget_high_f32(f1), 0)); M(vmulq_n_f32(f1, 0.25f));
|
|
M(vcvt_f16_f32(f1)); M(vcvt_f32_f16(vcvt_f16_f32(f2)));
|
|
/* loads and stores */
|
|
uint8_t out[128] __attribute__((aligned(16)));
|
|
memset(out, 0, sizeof out);
|
|
int8x16x2_t l2 = vld2q_s8((int8_t *)buf + 1); vst2q_s8((int8_t *)out, l2); M(l2);
|
|
int16x4x3_t l3 = vld3_s16((int16_t *)(buf + 2)); vst3_s16((int16_t *)(out + 32), l3); M(l3);
|
|
int32x4x4_t l4 = vld4q_s32((int32_t *)(buf + 4)); vst4q_s32((int32_t *)(out + 64), l4); M(l4);
|
|
uint8x8x3_t l3b = vld3_u8(buf + 3); vst3_u8(out + 5, l3b);
|
|
int16x8_t ln = vld1q_lane_s16((int16_t *)(buf + 6), a16, 6); M(ln);
|
|
int32x2x2_t lane2 = vld2_lane_s32((int32_t *)(buf + 8), ((int32x2x2_t){{vget_low_s32(a32), vget_high_s32(b32)}}), 1); M(lane2);
|
|
uint16x4x4_t lane4 = vld4_lane_u16((uint16_t *)(buf + 10), ((uint16x4x4_t){{vget_low_u16(ua16), vget_high_u16(ua16), vget_low_u16(ub16), vget_high_u16(ub16)}}), 3); M(lane4);
|
|
vst4_lane_u16((uint16_t *)(out + 100), lane4, 2);
|
|
int8x8_t dupl = vld1_dup_s8((int8_t *)buf + 11); M(dupl);
|
|
int32x4_t dupq = vld1q_dup_s32((int32_t *)(buf + 12)); M(dupq);
|
|
int16x4x2_t dup2 = vld2_dup_s16((int16_t *)(buf + 14)); M(dup2);
|
|
uint32x2x3_t dup3 = vld3_dup_u32((uint32_t *)(buf + 16)); M(dup3);
|
|
uint8x8x4_t dup4 = vld4_dup_u8(buf + 18); M(dup4);
|
|
vst1q_lane_s64((int64_t *)(out + 120), a64, 1);
|
|
mixb(out, sizeof out);
|
|
}
|
|
printf("neon %08x\n", h);
|
|
return 0;
|
|
}
|