static void get4x4var_sse2(const uint8_t *src, int src_stride,
const uint8_t *ref, int ref_stride,
unsigned int *sse, int *sum) {
const __m128i zero = _mm_setzero_si128();
const __m128i src0 = _mm_unpacklo_epi8(READ64(src, src_stride, 0), zero);
const __m128i src1 = _mm_unpacklo_epi8(READ64(src, src_stride, 2), zero);
const __m128i ref0 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 0), zero);
const __m128i ref1 = _mm_unpacklo_epi8(READ64(ref, ref_stride, 2), zero);
const __m128i diff0 = _mm_sub_epi16(src0, ref0);
const __m128i diff1 = _mm_sub_epi16(src1, ref1);
// sum
__m128i vsum = _mm_add_epi16(diff0, diff1);
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 4));
vsum = _mm_add_epi16(vsum, _mm_srli_si128(vsum, 2));
*sum = (int16_t)_mm_extract_epi16(vsum, 0);
// sse
vsum =
_mm_add_epi32(_mm_madd_epi16(diff0, diff0), _mm_madd_epi16(diff1, diff1));
vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 8));
vsum = _mm_add_epi32(vsum, _mm_srli_si128(vsum, 4));
*sse = _mm_cvtsi128_si32(vsum);
}
/**
* Fast scan test. This test is meant to find gross errors caused by read/write
* level failing on a single rank or dimm. The idea is to scan through all of
* memory in large steps. The large steps hit each rank multiple times, but not
* every byte. If the whole rank has errors, his should find it quickly. This test
* is suitable for an alive test during early boot.
*
* @param area Starting physical address
* @param max_address
* Ending physical address, exclusive
* @param bursts Burst to run
*
* @return Number of errors
*/
int __bdk_dram_test_fast_scan(uint64_t area, uint64_t max_address, int bursts)
{
int failures = 0;
const uint64_t step = 0x10008; /* The 8 is so we walk through cache lines too */
const uint64_t pattern1 = 0xaaaaaaaaaaaaaaaa;
const uint64_t pattern2 = 0x5555555555555555;
/* Walk through the region incrementing our offset by STEP */
uint64_t a = area;
while (a + 16 <= max_address)
{
WRITE64(a, pattern1);
WRITE64(a+8, pattern2);
__bdk_dram_flush_to_mem_range(a, a + 16);
a += step;
}
/* Read back, checking the writes */
a = area;
while (a + 16 <= max_address)
{
/* Prefetch 3 ahead for better performance */
uint64_t pre = a + step * 2;
if (pre + 16 < max_address)
BDK_PREFETCH(pre, 0);
/* Check pattern 1 */
uint64_t data1 = READ64(a);
if (bdk_unlikely(data1 != pattern1))
{
failures++;
__bdk_dram_report_error(a, data1, pattern1, 0, -1);
}
/* Check pattern 2 */
uint64_t data2 = READ64(a+8);
if (bdk_unlikely(data2 != pattern2))
{
failures++;
__bdk_dram_report_error(a+8, data2, pattern2, 0, -1);
}
a += step;
}
return failures;
}
char *
mono_interp_dis_mintop(const guint16 *base, const guint16 *ip)
{
GString *str = g_string_new ("");
guint32 token;
int target;
g_string_append_printf (str, "IL_%04x: %-10s", ip - base, mono_interp_opname [*ip]);
switch (mono_interp_opargtype [*ip]) {
case MintOpNoArgs:
break;
case MintOpUShortInt:
g_string_append_printf (str, " %u", * (guint16 *)(ip + 1));
break;
case MintOpTwoShorts:
g_string_append_printf (str, " %u,%u", * (guint16 *)(ip + 1), * (guint16 *)(ip + 2));
break;
case MintOpShortAndInt:
g_string_append_printf (str, " %u,%u", * (guint16 *)(ip + 1), (guint32)READ32(ip + 2));
break;
case MintOpShortInt:
g_string_append_printf (str, " %d", * (short *)(ip + 1));
break;
case MintOpClassToken:
case MintOpMethodToken:
case MintOpFieldToken:
token = * (guint16 *)(ip + 1);
g_string_append_printf (str, " %u", token);
break;
case MintOpInt:
g_string_append_printf (str, " %d", (gint32)READ32 (ip + 1));
break;
case MintOpLongInt:
g_string_append_printf (str, " %lld", (gint64)READ64 (ip + 1));
break;
case MintOpFloat: {
gint32 tmp = READ32 (ip + 1);
g_string_append_printf (str, " %g", * (float *)&tmp);
break;
}
case MintOpDouble: {
gint64 tmp = READ64 (ip + 1);
g_string_append_printf (str, " %g", * (double *)&tmp);
break;
}
case MintOpShortBranch:
target = ip + * (short *)(ip + 1) - base;
g_string_append_printf (str, " IL_%04x", target);
break;
case MintOpBranch:
target = ip + (gint32)READ32 (ip + 1) - base;
g_string_append_printf (str, " IL_%04x", target);
break;
case MintOpSwitch: {
const guint16 *p = ip + 1;
int sval = (gint32)READ32 (p);
int i;
p += 2;
g_string_append_printf (str, "(");
for (i = 0; i < sval; ++i) {
int offset;
if (i > 0)
g_string_append_printf (str, ", ");
offset = (gint32)READ32 (p);
g_string_append_printf (str, "IL_%04x", p + offset - base);
p += 2;
}
g_string_append_printf (str, ")");
break;
}
default:
g_string_append_printf (str, "unknown arg type\n");
}
return g_string_free (str, FALSE);
}