aboutsummaryrefslogtreecommitdiff
path: root/libvpx/vpx_dsp/mips/variance_mmi.c
diff options
context:
space:
mode:
Diffstat (limited to 'libvpx/vpx_dsp/mips/variance_mmi.c')
-rw-r--r--libvpx/vpx_dsp/mips/variance_mmi.c248
1 files changed, 157 insertions, 91 deletions
diff --git a/libvpx/vpx_dsp/mips/variance_mmi.c b/libvpx/vpx_dsp/mips/variance_mmi.c
index c1780c33a..c2adcfa01 100644
--- a/libvpx/vpx_dsp/mips/variance_mmi.c
+++ b/libvpx/vpx_dsp/mips/variance_mmi.c
@@ -150,7 +150,7 @@ static const uint8_t bilinear_filters[8][2] = {
"psrlh %[ftmp2], %[ftmp2], %[ftmp6] \n\t" \
\
/* store: temp2[0] ~ temp2[3] */ \
- "and %[ftmp2], %[ftmp2], %[mask] \n\t" \
+ "pand %[ftmp2], %[ftmp2], %[mask] \n\t" \
"packushb %[ftmp2], %[ftmp2], %[ftmp0] \n\t" \
"gssdrc1 %[ftmp2], 0x00(%[temp2_ptr]) \n\t"
@@ -163,7 +163,7 @@ static const uint8_t bilinear_filters[8][2] = {
"psrlh %[ftmp4], %[ftmp4], %[ftmp6] \n\t" \
\
/* store: temp2[0] ~ temp2[3] */ \
- "and %[ftmp4], %[ftmp4], %[mask] \n\t" \
+ "pand %[ftmp4], %[ftmp4], %[mask] \n\t" \
"packushb %[ftmp4], %[ftmp4], %[ftmp0] \n\t" \
"gssdrc1 %[ftmp4], 0x00(%[temp2_ptr]) \n\t"
@@ -225,8 +225,8 @@ static const uint8_t bilinear_filters[8][2] = {
"psrlh %[ftmp3], %[ftmp3], %[ftmp14] \n\t" \
\
/* store: temp2[0] ~ temp2[7] */ \
- "and %[ftmp2], %[ftmp2], %[mask] \n\t" \
- "and %[ftmp3], %[ftmp3], %[mask] \n\t" \
+ "pand %[ftmp2], %[ftmp2], %[mask] \n\t" \
+ "pand %[ftmp3], %[ftmp3], %[mask] \n\t" \
"packushb %[ftmp2], %[ftmp2], %[ftmp3] \n\t" \
"gssdlc1 %[ftmp2], 0x07(%[temp2_ptr]) \n\t" \
"gssdrc1 %[ftmp2], 0x00(%[temp2_ptr]) \n\t"
@@ -247,8 +247,8 @@ static const uint8_t bilinear_filters[8][2] = {
"psrlh %[ftmp9], %[ftmp9], %[ftmp14] \n\t" \
\
/* store: temp2[0] ~ temp2[7] */ \
- "and %[ftmp8], %[ftmp8], %[mask] \n\t" \
- "and %[ftmp9], %[ftmp9], %[mask] \n\t" \
+ "pand %[ftmp8], %[ftmp8], %[mask] \n\t" \
+ "pand %[ftmp9], %[ftmp9], %[mask] \n\t" \
"packushb %[ftmp8], %[ftmp8], %[ftmp9] \n\t" \
"gssdlc1 %[ftmp8], 0x07(%[temp2_ptr]) \n\t" \
"gssdrc1 %[ftmp8], 0x00(%[temp2_ptr]) \n\t"
@@ -319,8 +319,8 @@ static const uint8_t bilinear_filters[8][2] = {
"psrlh %[ftmp5], %[ftmp5], %[ftmp14] \n\t" \
\
/* store: temp2[8] ~ temp2[15] */ \
- "and %[ftmp4], %[ftmp4], %[mask] \n\t" \
- "and %[ftmp5], %[ftmp5], %[mask] \n\t" \
+ "pand %[ftmp4], %[ftmp4], %[mask] \n\t" \
+ "pand %[ftmp5], %[ftmp5], %[mask] \n\t" \
"packushb %[ftmp4], %[ftmp4], %[ftmp5] \n\t" \
"gssdlc1 %[ftmp4], 0x0f(%[temp2_ptr]) \n\t" \
"gssdrc1 %[ftmp4], 0x08(%[temp2_ptr]) \n\t"
@@ -343,8 +343,8 @@ static const uint8_t bilinear_filters[8][2] = {
"psrlh %[ftmp11], %[ftmp11], %[ftmp14] \n\t" \
\
/* store: temp2[8] ~ temp2[15] */ \
- "and %[ftmp10], %[ftmp10], %[mask] \n\t" \
- "and %[ftmp11], %[ftmp11], %[mask] \n\t" \
+ "pand %[ftmp10], %[ftmp10], %[mask] \n\t" \
+ "pand %[ftmp11], %[ftmp11], %[mask] \n\t" \
"packushb %[ftmp10], %[ftmp10], %[ftmp11] \n\t" \
"gssdlc1 %[ftmp10], 0x0f(%[temp2_ptr]) \n\t" \
"gssdrc1 %[ftmp10], 0x08(%[temp2_ptr]) \n\t"
@@ -414,13 +414,14 @@ static inline uint32_t vpx_variance64x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp9], %[ftmp9], %[ftmp9] \n\t"
- "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp9], %[ftmp9], %[ftmp9] \n\t"
+ "pxor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
"1: \n\t"
"gsldlc1 %[ftmp1], 0x07(%[src_ptr]) \n\t"
"gsldrc1 %[ftmp1], 0x00(%[src_ptr]) \n\t"
@@ -478,7 +479,7 @@ static inline uint32_t vpx_variance64x(const uint8_t *src_ptr, int src_stride,
"mfc1 %[tmp1], %[ftmp9] \n\t"
"mfhc1 %[tmp2], %[ftmp9] \n\t"
"addu %[sum], %[tmp1], %[tmp2] \n\t"
- "dsrl %[ftmp1], %[ftmp10], %[ftmp11] \n\t"
+ "ssrld %[ftmp1], %[ftmp10], %[ftmp11] \n\t"
"paddw %[ftmp1], %[ftmp1], %[ftmp10] \n\t"
"swc1 %[ftmp1], 0x00(%[sse]) \n\t"
: [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),
@@ -496,6 +497,7 @@ static inline uint32_t vpx_variance64x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse)
: "memory"
);
+ /* clang-format on */
return *sse - (((int64_t)sum * sum) / (64 * high));
}
@@ -519,13 +521,14 @@ uint32_t vpx_variance32x64_mmi(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
"li %[tmp0], 0x40 \n\t"
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp9], %[ftmp9], %[ftmp9] \n\t"
- "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp9], %[ftmp9], %[ftmp9] \n\t"
+ "pxor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
"1: \n\t"
"gsldlc1 %[ftmp1], 0x07(%[src_ptr]) \n\t"
"gsldrc1 %[ftmp1], 0x00(%[src_ptr]) \n\t"
@@ -559,7 +562,7 @@ uint32_t vpx_variance32x64_mmi(const uint8_t *src_ptr, int src_stride,
"mfc1 %[tmp1], %[ftmp9] \n\t"
"mfhc1 %[tmp2], %[ftmp9] \n\t"
"addu %[sum], %[tmp1], %[tmp2] \n\t"
- "dsrl %[ftmp1], %[ftmp10], %[ftmp11] \n\t"
+ "ssrld %[ftmp1], %[ftmp10], %[ftmp11] \n\t"
"paddw %[ftmp1], %[ftmp1], %[ftmp10] \n\t"
"swc1 %[ftmp1], 0x00(%[sse]) \n\t"
: [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),
@@ -577,6 +580,7 @@ uint32_t vpx_variance32x64_mmi(const uint8_t *src_ptr, int src_stride,
[sse]"r"(sse)
: "memory"
);
+ /* clang-format on */
return *sse - (((int64_t)sum * sum) / 2048);
}
@@ -590,14 +594,15 @@ static inline uint32_t vpx_variance32x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
- "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
- "xor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
+ "pxor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ "pxor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
"1: \n\t"
"gsldlc1 %[ftmp1], 0x07(%[src_ptr]) \n\t"
"gsldrc1 %[ftmp1], 0x00(%[src_ptr]) \n\t"
@@ -625,7 +630,7 @@ static inline uint32_t vpx_variance32x(const uint8_t *src_ptr, int src_stride,
MMI_ADDU(%[ref_ptr], %[ref_ptr], %[ref_stride])
"bnez %[tmp0], 1b \n\t"
- "dsrl %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
+ "ssrld %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
"paddw %[ftmp9], %[ftmp9], %[ftmp8] \n\t"
"swc1 %[ftmp9], 0x00(%[sse]) \n\t"
@@ -636,7 +641,7 @@ static inline uint32_t vpx_variance32x(const uint8_t *src_ptr, int src_stride,
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp5] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp6] \n\t"
- "dsrl %[ftmp0], %[ftmp3], %[ftmp11] \n\t"
+ "ssrld %[ftmp0], %[ftmp3], %[ftmp11] \n\t"
"paddw %[ftmp0], %[ftmp0], %[ftmp3] \n\t"
"swc1 %[ftmp0], 0x00(%[sum]) \n\t"
@@ -653,6 +658,7 @@ static inline uint32_t vpx_variance32x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse), [sum]"r"(&sum)
: "memory"
);
+ /* clang-format on */
return *sse - (((int64_t)sum * sum) / (32 * high));
}
@@ -676,14 +682,15 @@ static inline uint32_t vpx_variance16x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
- "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
- "xor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
+ "pxor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ "pxor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
"1: \n\t"
"gsldlc1 %[ftmp1], 0x07(%[src_ptr]) \n\t"
"gsldrc1 %[ftmp1], 0x00(%[src_ptr]) \n\t"
@@ -701,7 +708,7 @@ static inline uint32_t vpx_variance16x(const uint8_t *src_ptr, int src_stride,
MMI_ADDU(%[ref_ptr], %[ref_ptr], %[ref_stride])
"bnez %[tmp0], 1b \n\t"
- "dsrl %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
+ "ssrld %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
"paddw %[ftmp9], %[ftmp9], %[ftmp8] \n\t"
"swc1 %[ftmp9], 0x00(%[sse]) \n\t"
@@ -712,7 +719,7 @@ static inline uint32_t vpx_variance16x(const uint8_t *src_ptr, int src_stride,
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp5] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp6] \n\t"
- "dsrl %[ftmp0], %[ftmp3], %[ftmp11] \n\t"
+ "ssrld %[ftmp0], %[ftmp3], %[ftmp11] \n\t"
"paddw %[ftmp0], %[ftmp0], %[ftmp3] \n\t"
"swc1 %[ftmp0], 0x00(%[sum]) \n\t"
@@ -729,6 +736,7 @@ static inline uint32_t vpx_variance16x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse), [sum]"r"(&sum)
: "memory"
);
+ /* clang-format on */
return *sse - (((int64_t)sum * sum) / (16 * high));
}
@@ -753,14 +761,15 @@ static inline uint32_t vpx_variance8x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
- "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
- "xor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
+ "pxor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ "pxor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
"1: \n\t"
"gsldlc1 %[ftmp1], 0x07(%[src_ptr]) \n\t"
"gsldrc1 %[ftmp1], 0x00(%[src_ptr]) \n\t"
@@ -773,7 +782,7 @@ static inline uint32_t vpx_variance8x(const uint8_t *src_ptr, int src_stride,
MMI_ADDU(%[ref_ptr], %[ref_ptr], %[ref_stride])
"bnez %[tmp0], 1b \n\t"
- "dsrl %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
+ "ssrld %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
"paddw %[ftmp9], %[ftmp9], %[ftmp8] \n\t"
"swc1 %[ftmp9], 0x00(%[sse]) \n\t"
@@ -784,7 +793,7 @@ static inline uint32_t vpx_variance8x(const uint8_t *src_ptr, int src_stride,
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp5] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp6] \n\t"
- "dsrl %[ftmp0], %[ftmp3], %[ftmp11] \n\t"
+ "ssrld %[ftmp0], %[ftmp3], %[ftmp11] \n\t"
"paddw %[ftmp0], %[ftmp0], %[ftmp3] \n\t"
"swc1 %[ftmp0], 0x00(%[sum]) \n\t"
@@ -801,6 +810,7 @@ static inline uint32_t vpx_variance8x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse), [sum]"r"(&sum)
: "memory"
);
+ /* clang-format on */
return *sse - (((int64_t)sum * sum) / (8 * high));
}
@@ -825,14 +835,15 @@ static inline uint32_t vpx_variance4x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp10] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp6], %[ftmp6], %[ftmp6] \n\t"
- "xor %[ftmp7], %[ftmp7], %[ftmp7] \n\t"
- "xor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp6], %[ftmp6], %[ftmp6] \n\t"
+ "pxor %[ftmp7], %[ftmp7], %[ftmp7] \n\t"
+ "pxor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
"1: \n\t"
"gsldlc1 %[ftmp1], 0x07(%[src_ptr]) \n\t"
"gsldrc1 %[ftmp1], 0x00(%[src_ptr]) \n\t"
@@ -845,7 +856,7 @@ static inline uint32_t vpx_variance4x(const uint8_t *src_ptr, int src_stride,
MMI_ADDU(%[ref_ptr], %[ref_ptr], %[ref_stride])
"bnez %[tmp0], 1b \n\t"
- "dsrl %[ftmp9], %[ftmp6], %[ftmp10] \n\t"
+ "ssrld %[ftmp9], %[ftmp6], %[ftmp10] \n\t"
"paddw %[ftmp9], %[ftmp9], %[ftmp6] \n\t"
"swc1 %[ftmp9], 0x00(%[sse]) \n\t"
@@ -856,7 +867,7 @@ static inline uint32_t vpx_variance4x(const uint8_t *src_ptr, int src_stride,
"paddw %[ftmp3], %[ftmp3], %[ftmp4] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp5] \n\t"
"psubw %[ftmp3], %[ftmp3], %[ftmp6] \n\t"
- "dsrl %[ftmp0], %[ftmp3], %[ftmp10] \n\t"
+ "ssrld %[ftmp0], %[ftmp3], %[ftmp10] \n\t"
"paddw %[ftmp0], %[ftmp0], %[ftmp3] \n\t"
"swc1 %[ftmp0], 0x00(%[sum]) \n\t"
: [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),
@@ -872,6 +883,7 @@ static inline uint32_t vpx_variance4x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse), [sum]"r"(&sum)
: "memory"
);
+ /* clang-format on */
return *sse - (((int64_t)sum * sum) / (4 * high));
}
@@ -894,12 +906,13 @@ static inline uint32_t vpx_mse16x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
"1: \n\t"
VARIANCE_SSE_16
@@ -909,7 +922,7 @@ static inline uint32_t vpx_mse16x(const uint8_t *src_ptr, int src_stride,
MMI_ADDU(%[ref_ptr], %[ref_ptr], %[ref_stride])
"bnez %[tmp0], 1b \n\t"
- "dsrl %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
+ "ssrld %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
"paddw %[ftmp9], %[ftmp9], %[ftmp8] \n\t"
"swc1 %[ftmp9], 0x00(%[sse]) \n\t"
: [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),
@@ -925,6 +938,7 @@ static inline uint32_t vpx_mse16x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse)
: "memory"
);
+ /* clang-format on */
return *sse;
}
@@ -947,12 +961,13 @@ static inline uint32_t vpx_mse8x(const uint8_t *src_ptr, int src_stride,
*sse = 0;
+ /* clang-format off */
__asm__ volatile (
"li %[tmp0], 0x20 \n\t"
"mtc1 %[tmp0], %[ftmp11] \n\t"
MMI_L(%[tmp0], %[high], 0x00)
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
- "xor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp8], %[ftmp8], %[ftmp8] \n\t"
"1: \n\t"
VARIANCE_SSE_8
@@ -962,7 +977,7 @@ static inline uint32_t vpx_mse8x(const uint8_t *src_ptr, int src_stride,
MMI_ADDU(%[ref_ptr], %[ref_ptr], %[ref_stride])
"bnez %[tmp0], 1b \n\t"
- "dsrl %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
+ "ssrld %[ftmp9], %[ftmp8], %[ftmp11] \n\t"
"paddw %[ftmp9], %[ftmp9], %[ftmp8] \n\t"
"swc1 %[ftmp9], 0x00(%[sse]) \n\t"
: [ftmp0]"=&f"(ftmp[0]), [ftmp1]"=&f"(ftmp[1]),
@@ -978,6 +993,7 @@ static inline uint32_t vpx_mse8x(const uint8_t *src_ptr, int src_stride,
[high]"r"(&high), [sse]"r"(sse)
: "memory"
);
+ /* clang-format on */
return *sse;
}
@@ -1021,22 +1037,39 @@ static inline void var_filter_block2d_bil_16x(const uint8_t *src_ptr,
uint8_t *temp2_ptr = temp2;
mips_reg l_counter = counter;
double ftmp[15];
+ double ff_ph_40, mask;
+ double filter_x0, filter_x1, filter_y0, filter_y1;
mips_reg tmp[2];
- DECLARE_ALIGNED(8, const uint64_t, ff_ph_40) = { 0x0040004000400040ULL };
- DECLARE_ALIGNED(8, const uint64_t, mask) = { 0x00ff00ff00ff00ffULL };
+ uint64_t x0, x1, y0, y1, all;
const uint8_t *filter_x = bilinear_filters[x_offset];
const uint8_t *filter_y = bilinear_filters[y_offset];
+ x0 = (uint64_t)filter_x[0];
+ x1 = (uint64_t)filter_x[1];
+ y0 = (uint64_t)filter_y[0];
+ y1 = (uint64_t)filter_y[1];
+ all = x0 | x1 << 8 | y0 << 16 | y1 << 24;
+ /* clang-format off */
__asm__ volatile (
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ MMI_MTC1(%[all], %[ftmp14])
+ "punpcklbh %[ftmp14], %[ftmp14], %[ftmp0] \n\t"
+ "pshufh %[filter_x0], %[ftmp14], %[ftmp0] \n\t"
+ MMI_LI(%[tmp0], 0x10)
+ MMI_MTC1(%[tmp0], %[mask])
+ "ssrld %[ftmp14], %[ftmp14], %[mask] \n\t"
+ "pshufh %[filter_x1], %[ftmp14], %[ftmp0] \n\t"
+ "ssrld %[ftmp14], %[ftmp14], %[mask] \n\t"
+ "pshufh %[filter_y0], %[ftmp14], %[ftmp0] \n\t"
+ "ssrld %[ftmp14], %[ftmp14], %[mask] \n\t"
+ "pshufh %[filter_y1], %[ftmp14], %[ftmp0] \n\t"
MMI_LI(%[tmp0], 0x07)
MMI_MTC1(%[tmp0], %[ftmp14])
- "pshufh %[filter_x0], %[filter_x0], %[ftmp0] \n\t"
- "pshufh %[filter_x1], %[filter_x1], %[ftmp0] \n\t"
- "pshufh %[filter_y0], %[filter_y0], %[ftmp0] \n\t"
- "pshufh %[filter_y1], %[filter_y1], %[ftmp0] \n\t"
-
+ MMI_LI(%[tmp0], 0x0040004000400040)
+ MMI_MTC1(%[tmp0], %[ff_ph_40])
+ MMI_LI(%[tmp0], 0x00ff00ff00ff00ff)
+ MMI_MTC1(%[tmp0], %[mask])
// fdata3: fdata3[0] ~ fdata3[15]
VAR_FILTER_BLOCK2D_BIL_FIRST_PASS_16_A
@@ -1072,15 +1105,13 @@ static inline void var_filter_block2d_bil_16x(const uint8_t *src_ptr,
[ftmp11] "=&f"(ftmp[11]), [ftmp12] "=&f"(ftmp[12]),
[ftmp13] "=&f"(ftmp[13]), [ftmp14] "=&f"(ftmp[14]),
[tmp0] "=&r"(tmp[0]), [src_ptr] "+&r"(src_ptr), [temp2_ptr] "+&r"(temp2_ptr),
- [counter]"+&r"(l_counter)
- : [filter_x0] "f"((uint64_t)filter_x[0]),
- [filter_x1] "f"((uint64_t)filter_x[1]),
- [filter_y0] "f"((uint64_t)filter_y[0]),
- [filter_y1] "f"((uint64_t)filter_y[1]),
- [src_stride] "r"((mips_reg)src_stride), [ff_ph_40] "f"(ff_ph_40),
- [mask] "f"(mask)
+ [counter]"+&r"(l_counter), [ff_ph_40] "=&f"(ff_ph_40), [mask] "=&f"(mask),
+ [filter_x0] "=&f"(filter_x0), [filter_x1] "=&f"(filter_x1),
+ [filter_y0] "=&f"(filter_y0), [filter_y1] "=&f"(filter_y1)
+ : [src_stride] "r"((mips_reg)src_stride), [all] "r"(all)
: "memory"
);
+ /* clang-format on */
}
#define SUBPIX_VAR16XN(H) \
@@ -1105,19 +1136,38 @@ static inline void var_filter_block2d_bil_8x(const uint8_t *src_ptr,
mips_reg l_counter = counter;
double ftmp[15];
mips_reg tmp[2];
- DECLARE_ALIGNED(8, const uint64_t, ff_ph_40) = { 0x0040004000400040ULL };
- DECLARE_ALIGNED(8, const uint64_t, mask) = { 0x00ff00ff00ff00ffULL };
+ double ff_ph_40, mask;
+ uint64_t x0, x1, y0, y1, all;
+ double filter_x0, filter_x1, filter_y0, filter_y1;
const uint8_t *filter_x = bilinear_filters[x_offset];
const uint8_t *filter_y = bilinear_filters[y_offset];
+ x0 = (uint64_t)filter_x[0];
+ x1 = (uint64_t)filter_x[1];
+ y0 = (uint64_t)filter_y[0];
+ y1 = (uint64_t)filter_y[1];
+ all = x0 | x1 << 8 | y0 << 16 | y1 << 24;
+ /* clang-format off */
__asm__ volatile (
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ MMI_MTC1(%[all], %[ftmp14])
+ "punpcklbh %[ftmp14], %[ftmp14], %[ftmp0] \n\t"
+ "pshufh %[filter_x0], %[ftmp14], %[ftmp0] \n\t"
+ MMI_LI(%[tmp0], 0x10)
+ MMI_MTC1(%[tmp0], %[mask])
+ "ssrld %[ftmp14], %[ftmp14], %[mask] \n\t"
+ "pshufh %[filter_x1], %[ftmp14], %[ftmp0] \n\t"
+ "ssrld %[ftmp14], %[ftmp14], %[mask] \n\t"
+ "pshufh %[filter_y0], %[ftmp14], %[ftmp0] \n\t"
+ "ssrld %[ftmp14], %[ftmp14], %[mask] \n\t"
+ "pshufh %[filter_y1], %[ftmp14], %[ftmp0] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
MMI_LI(%[tmp0], 0x07)
MMI_MTC1(%[tmp0], %[ftmp14])
- "pshufh %[filter_x0], %[filter_x0], %[ftmp0] \n\t"
- "pshufh %[filter_x1], %[filter_x1], %[ftmp0] \n\t"
- "pshufh %[filter_y0], %[filter_y0], %[ftmp0] \n\t"
- "pshufh %[filter_y1], %[filter_y1], %[ftmp0] \n\t"
+ MMI_LI(%[tmp0], 0x0040004000400040)
+ MMI_MTC1(%[tmp0], %[ff_ph_40])
+ MMI_LI(%[tmp0], 0x00ff00ff00ff00ff)
+ MMI_MTC1(%[tmp0], %[mask])
// fdata3: fdata3[0] ~ fdata3[7]
VAR_FILTER_BLOCK2D_BIL_FIRST_PASS_8_A
@@ -1154,15 +1204,13 @@ static inline void var_filter_block2d_bil_8x(const uint8_t *src_ptr,
[ftmp11] "=&f"(ftmp[11]), [ftmp12] "=&f"(ftmp[12]),
[ftmp13] "=&f"(ftmp[13]), [ftmp14] "=&f"(ftmp[14]),
[tmp0] "=&r"(tmp[0]), [src_ptr] "+&r"(src_ptr), [temp2_ptr] "+&r"(temp2_ptr),
- [counter]"+&r"(l_counter)
- : [filter_x0] "f"((uint64_t)filter_x[0]),
- [filter_x1] "f"((uint64_t)filter_x[1]),
- [filter_y0] "f"((uint64_t)filter_y[0]),
- [filter_y1] "f"((uint64_t)filter_y[1]),
- [src_stride] "r"((mips_reg)src_stride), [ff_ph_40] "f"(ff_ph_40),
- [mask] "f"(mask)
+ [counter]"+&r"(l_counter), [ff_ph_40] "=&f"(ff_ph_40), [mask] "=&f"(mask),
+ [filter_x0] "=&f"(filter_x0), [filter_x1] "=&f"(filter_x1),
+ [filter_y0] "=&f"(filter_y0), [filter_y1] "=&f"(filter_y1)
+ : [src_stride] "r"((mips_reg)src_stride), [all] "r"(all)
: "memory"
);
+ /* clang-format on */
}
#define SUBPIX_VAR8XN(H) \
@@ -1188,19 +1236,38 @@ static inline void var_filter_block2d_bil_4x(const uint8_t *src_ptr,
mips_reg l_counter = counter;
double ftmp[7];
mips_reg tmp[2];
- DECLARE_ALIGNED(8, const uint64_t, ff_ph_40) = { 0x0040004000400040ULL };
- DECLARE_ALIGNED(8, const uint64_t, mask) = { 0x00ff00ff00ff00ffULL };
+ double ff_ph_40, mask;
+ uint64_t x0, x1, y0, y1, all;
+ double filter_x0, filter_x1, filter_y0, filter_y1;
const uint8_t *filter_x = bilinear_filters[x_offset];
const uint8_t *filter_y = bilinear_filters[y_offset];
+ x0 = (uint64_t)filter_x[0];
+ x1 = (uint64_t)filter_x[1];
+ y0 = (uint64_t)filter_y[0];
+ y1 = (uint64_t)filter_y[1];
+ all = x0 | x1 << 8 | y0 << 16 | y1 << 24;
+ /* clang-format off */
__asm__ volatile (
- "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ MMI_MTC1(%[all], %[ftmp6])
+ "punpcklbh %[ftmp6], %[ftmp6], %[ftmp0] \n\t"
+ "pshufh %[filter_x0], %[ftmp6], %[ftmp0] \n\t"
+ MMI_LI(%[tmp0], 0x10)
+ MMI_MTC1(%[tmp0], %[mask])
+ "ssrld %[ftmp6], %[ftmp6], %[mask] \n\t"
+ "pshufh %[filter_x1], %[ftmp6], %[ftmp0] \n\t"
+ "ssrld %[ftmp6], %[ftmp6], %[mask] \n\t"
+ "pshufh %[filter_y0], %[ftmp6], %[ftmp0] \n\t"
+ "ssrld %[ftmp6], %[ftmp6], %[mask] \n\t"
+ "pshufh %[filter_y1], %[ftmp6], %[ftmp0] \n\t"
+ "pxor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
MMI_LI(%[tmp0], 0x07)
MMI_MTC1(%[tmp0], %[ftmp6])
- "pshufh %[filter_x0], %[filter_x0], %[ftmp0] \n\t"
- "pshufh %[filter_x1], %[filter_x1], %[ftmp0] \n\t"
- "pshufh %[filter_y0], %[filter_y0], %[ftmp0] \n\t"
- "pshufh %[filter_y1], %[filter_y1], %[ftmp0] \n\t"
+ MMI_LI(%[tmp0], 0x0040004000400040)
+ MMI_MTC1(%[tmp0], %[ff_ph_40])
+ MMI_LI(%[tmp0], 0x00ff00ff00ff00ff)
+ MMI_MTC1(%[tmp0], %[mask])
// fdata3: fdata3[0] ~ fdata3[3]
VAR_FILTER_BLOCK2D_BIL_FIRST_PASS_4_A
@@ -1232,15 +1299,14 @@ static inline void var_filter_block2d_bil_4x(const uint8_t *src_ptr,
: [ftmp0] "=&f"(ftmp[0]), [ftmp1] "=&f"(ftmp[1]), [ftmp2] "=&f"(ftmp[2]),
[ftmp3] "=&f"(ftmp[3]), [ftmp4] "=&f"(ftmp[4]), [ftmp5] "=&f"(ftmp[5]),
[ftmp6] "=&f"(ftmp[6]), [tmp0] "=&r"(tmp[0]), [src_ptr] "+&r"(src_ptr),
- [temp2_ptr] "+&r"(temp2_ptr), [counter]"+&r"(l_counter)
- : [filter_x0] "f"((uint64_t)filter_x[0]),
- [filter_x1] "f"((uint64_t)filter_x[1]),
- [filter_y0] "f"((uint64_t)filter_y[0]),
- [filter_y1] "f"((uint64_t)filter_y[1]),
- [src_stride] "r"((mips_reg)src_stride), [ff_ph_40] "f"(ff_ph_40),
- [mask] "f"(mask)
+ [temp2_ptr] "+&r"(temp2_ptr), [counter]"+&r"(l_counter),
+ [ff_ph_40] "=&f"(ff_ph_40), [mask] "=&f"(mask),
+ [filter_x0] "=&f"(filter_x0), [filter_x1] "=&f"(filter_x1),
+ [filter_y0] "=&f"(filter_y0), [filter_y1] "=&f"(filter_y1)
+ : [src_stride] "r"((mips_reg)src_stride), [all] "r"(all)
: "memory"
);
+ /* clang-format on */
}
#define SUBPIX_VAR4XN(H) \
generated by cgit v1.2.3 (git 2.25.1) at 2024-06-23 15:28:23 +0000