void
test8bit (void)
{
i1 = _mm_cmpistrm (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
k1 = _mm_cmpistri (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
k1 = _mm_cmpistra (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
k1 = _mm_cmpistrc (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
k1 = _mm_cmpistro (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
k1 = _mm_cmpistrs (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
k1 = _mm_cmpistrz (i2, i3, k4); /* { dg-error "the third argument must be an 8-bit immediate" } */
i1 = _mm_cmpestrm (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
k1 = _mm_cmpestri (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
k1 = _mm_cmpestra (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
k1 = _mm_cmpestrc (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
k1 = _mm_cmpestro (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
k1 = _mm_cmpestrs (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
k1 = _mm_cmpestrz (i2, k2, i3, k3, k4); /* { dg-error "the fifth argument must be an 8-bit immediate" } */
b1 = _mm256_blend_ps (b2, b3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
k1 = _cvtss_sh (f1, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm256_cvtps_ph (b2, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
b1 = _mm256_dp_ps (b2, b3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
e1 = _mm256_permute2f128_pd (e2, e3, k4);/* { dg-error "the last argument must be an 8-bit immediate" } */
b1 = _mm256_permute2f128_ps (b2, b3, k4);/* { dg-error "the last argument must be an 8-bit immediate" } */
l1 = _mm256_permute2f128_si256 (l2, l3, k4);/* { dg-error "the last argument must be an 8-bit immediate" } */
b1 = _mm256_permute_ps (b2, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_aeskeygenassist_si128 (i2, k4);/* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_blend_epi16 (i2, i3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_clmulepi64_si128 (i2, i3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_cvtps_ph (a1, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
d1 = _mm_dp_pd (d2, d3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
a1 = _mm_dp_ps (a2, a3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
a1 = _mm_insert_ps (a2, a3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_mpsadbw_epu8 (i2, i3, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
a1 = _mm_permute_ps (a2, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_slli_si128 (i2, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
i1 = _mm_srli_si128 (i2, k4); /* { dg-error "the last argument must be an 8-bit immediate" } */
}
static void
sse4_2_test (void)
{
union
{
__m128i x[NUM];
char c[NUM *16];
} src1, src2;
__m128i res, correct;
int correct_flags, l1, l2;
int flags, cf, zf, sf, of, af;
int i;
for (i = 0; i < NUM *16; i++)
{
src1.c[i] = rand ();
src2.c[i] = rand ();
}
for (i = 0; i < NUM; i++)
{
l1 = rand () % 18;
l2 = rand () % 18;
switch ((rand () % 4))
{
case 0:
res = _mm_cmpestrm (src1.x[i], l1, src2.x[i], l2, IMM_VAL0);
cf = _mm_cmpestrc (src1.x[i], l1, src2.x[i], l2, IMM_VAL0);
zf = _mm_cmpestrz (src1.x[i], l1, src2.x[i], l2, IMM_VAL0);
sf = _mm_cmpestrs (src1.x[i], l1, src2.x[i], l2, IMM_VAL0);
of = _mm_cmpestro (src1.x[i], l1, src2.x[i], l2, IMM_VAL0);
af = _mm_cmpestra (src1.x[i], l1, src2.x[i], l2, IMM_VAL0);
correct = cmp_em (&src1.x[i], l1, &src2.x[i], l2, IMM_VAL0,
&correct_flags);
break;
case 1:
res = _mm_cmpestrm (src1.x[i], l1, src2.x[i], l2, IMM_VAL1);
cf = _mm_cmpestrc (src1.x[i], l1, src2.x[i], l2, IMM_VAL1);
zf = _mm_cmpestrz (src1.x[i], l1, src2.x[i], l2, IMM_VAL1);
sf = _mm_cmpestrs (src1.x[i], l1, src2.x[i], l2, IMM_VAL1);
of = _mm_cmpestro (src1.x[i], l1, src2.x[i], l2, IMM_VAL1);
af = _mm_cmpestra (src1.x[i], l1, src2.x[i], l2, IMM_VAL1);
correct = cmp_em (&src1.x[i], l1, &src2.x[i], l2, IMM_VAL1,
&correct_flags);
break;
case 2:
res = _mm_cmpestrm (src1.x[i], l1, src2.x[i], l2, IMM_VAL2);
cf = _mm_cmpestrc (src1.x[i], l1, src2.x[i], l2, IMM_VAL2);
zf = _mm_cmpestrz (src1.x[i], l1, src2.x[i], l2, IMM_VAL2);
sf = _mm_cmpestrs (src1.x[i], l1, src2.x[i], l2, IMM_VAL2);
of = _mm_cmpestro (src1.x[i], l1, src2.x[i], l2, IMM_VAL2);
af = _mm_cmpestra (src1.x[i], l1, src2.x[i], l2, IMM_VAL2);
correct = cmp_em (&src1.x[i], l1, &src2.x[i], l2, IMM_VAL2,
&correct_flags);
break;
default:
res = _mm_cmpestrm (src1.x[i], l1, src2.x[i], l2, IMM_VAL3);
cf = _mm_cmpestrc (src1.x[i], l1, src2.x[i], l2, IMM_VAL3);
zf = _mm_cmpestrz (src1.x[i], l1, src2.x[i], l2, IMM_VAL3);
sf = _mm_cmpestrs (src1.x[i], l1, src2.x[i], l2, IMM_VAL3);
of = _mm_cmpestro (src1.x[i], l1, src2.x[i], l2, IMM_VAL3);
af = _mm_cmpestra (src1.x[i], l1, src2.x[i], l2, IMM_VAL3);
correct = cmp_em (&src1.x[i], l1, &src2.x[i], l2, IMM_VAL3,
&correct_flags);
break;
}
if (memcmp (&correct, &res, sizeof (res)))
abort ();
flags = 0;
if (cf)
flags |= CFLAG;
if (zf)
flags |= ZFLAG;
if (sf)
flags |= SFLAG;
if (of)
flags |= OFLAG;
if (flags != correct_flags
|| (af && (cf || zf))
|| (!af && !(cf || zf)))
abort ();
}
}
int test_mm_cmpestrc(__m128i A, int LA, __m128i B, int LB) {
// CHECK-LABEL: test_mm_cmpestrc
// CHECK: call i32 @llvm.x86.sse42.pcmpestric128(<16 x i8> %{{.*}}, i32 %{{.*}}, <16 x i8> %{{.*}}, i32 %{{.*}}, i8 7)
return _mm_cmpestrc(A, LA, B, LB, 7);
}
int test_mm_cmpestrc(__m128i A, int LA, __m128i B, int LB) {
// CHECK-LABEL: test_mm_cmpestrc
// CHECK: @llvm.x86.sse42.pcmpestric128
return _mm_cmpestrc(A, LA, B, LB, 7);
}
}bool validate_utf8_sse(const char *src, size_t len) {
const char *end = src + len;
while (src + 16 < end) {
__m128i chunk = _mm_loadu_si128((const __m128i *)(src));
int asciiMask = _mm_movemask_epi8(chunk);
if (!asciiMask) {
src += 16;
continue;
}
__m128i chunk_signed = _mm_add_epi8(chunk, _mm_set1_epi8(0x80));
__m128i cond2 =
_mm_cmplt_epi8(_mm_set1_epi8(0xc2 - 1 - 0x80), chunk_signed);
__m128i state = _mm_set1_epi8((char)(0x0 | 0x80));
state = _mm_blendv_epi8(state, _mm_set1_epi8((char)(0x2 | 0xc0)), cond2);
__m128i cond3 =
_mm_cmplt_epi8(_mm_set1_epi8(0xe0 - 1 - 0x80), chunk_signed);
state = _mm_blendv_epi8(state, _mm_set1_epi8((char)(0x3 | 0xe0)), cond3);
__m128i mask3 = _mm_slli_si128(cond3, 1);
__m128i cond4 =
_mm_cmplt_epi8(_mm_set1_epi8(0xf0 - 1 - 0x80), chunk_signed);
// Fall back to the scalar processing
if (_mm_movemask_epi8(cond4)) {
break;
}
__m128i count = _mm_and_si128(state, _mm_set1_epi8(0x7));
__m128i count_sub1 = _mm_subs_epu8(count, _mm_set1_epi8(0x1));
__m128i counts = _mm_add_epi8(count, _mm_slli_si128(count_sub1, 1));
__m128i shifts = count_sub1;
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 1));
counts = _mm_add_epi8(
counts, _mm_slli_si128(_mm_subs_epu8(counts, _mm_set1_epi8(0x2)), 2));
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 2));
if (asciiMask ^ _mm_movemask_epi8(_mm_cmpgt_epi8(counts, _mm_set1_epi8(0))))
return false; // error
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 4));
if (_mm_movemask_epi8(_mm_cmpgt_epi8(
_mm_sub_epi8(_mm_slli_si128(counts, 1), counts), _mm_set1_epi8(1))))
return false; // error
shifts = _mm_add_epi8(shifts, _mm_slli_si128(shifts, 8));
__m128i mask = _mm_and_si128(state, _mm_set1_epi8(0xf8));
shifts =
_mm_and_si128(shifts, _mm_cmplt_epi8(counts, _mm_set1_epi8(2))); // <=1
chunk =
_mm_andnot_si128(mask, chunk); // from now on, we only have usefull bits
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 1),
_mm_srli_si128(_mm_slli_epi16(shifts, 7), 1));
__m128i chunk_right = _mm_slli_si128(chunk, 1);
__m128i chunk_low = _mm_blendv_epi8(
chunk,
_mm_or_si128(chunk, _mm_and_si128(_mm_slli_epi16(chunk_right, 6),
_mm_set1_epi8(0xc0))),
_mm_cmpeq_epi8(counts, _mm_set1_epi8(1)));
__m128i chunk_high =
_mm_and_si128(chunk, _mm_cmpeq_epi8(counts, _mm_set1_epi8(2)));
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 2),
_mm_srli_si128(_mm_slli_epi16(shifts, 6), 2));
chunk_high = _mm_srli_epi32(chunk_high, 2);
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 4),
_mm_srli_si128(_mm_slli_epi16(shifts, 5), 4));
chunk_high = _mm_or_si128(
chunk_high, _mm_and_si128(_mm_and_si128(_mm_slli_epi32(chunk_right, 4),
_mm_set1_epi8(0xf0)),
mask3));
int c = _mm_extract_epi16(counts, 7);
int source_advance = !(c & 0x0200) ? 16 : !(c & 0x02) ? 15 : 14;
__m128i high_bits = _mm_and_si128(chunk_high, _mm_set1_epi8(0xf8));
if (!_mm_testz_si128(
mask3,
_mm_or_si128(_mm_cmpeq_epi8(high_bits, _mm_set1_epi8(0x00)),
_mm_cmpeq_epi8(high_bits, _mm_set1_epi8(0xd8)))))
return false;
shifts = _mm_blendv_epi8(shifts, _mm_srli_si128(shifts, 8),
_mm_srli_si128(_mm_slli_epi16(shifts, 4), 8));
chunk_high = _mm_slli_si128(chunk_high, 1);
__m128i shuf =
_mm_add_epi8(shifts, _mm_set_epi8(15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5,
4, 3, 2, 1, 0));
chunk_low = _mm_shuffle_epi8(chunk_low, shuf);
chunk_high = _mm_shuffle_epi8(chunk_high, shuf);
__m128i utf16_low = _mm_unpacklo_epi8(chunk_low, chunk_high);
__m128i utf16_high = _mm_unpackhi_epi8(chunk_low, chunk_high);
if (_mm_cmpestrc(_mm_cvtsi64_si128(0xfdeffdd0fffffffe), 4, utf16_high, 8,
_SIDD_UWORD_OPS | _SIDD_CMP_RANGES) |
_mm_cmpestrc(_mm_cvtsi64_si128(0xfdeffdd0fffffffe), 4, utf16_low, 8,
_SIDD_UWORD_OPS | _SIDD_CMP_RANGES)) {
return false;
}
src += source_advance;
}
return validate_utf8(src, end - src);
}
