ref: 14750b4797a011da07c8cb718b13aa3a02ac3d84
parent: 906dacd34972e42819dab320960ffcfb7b84aada
author: gxw <[email protected]>
date: Mon Aug 13 05:23:03 EDT 2018
Add optimization files for loongson platform 1. Add dct_mmi.c in codec/decoder/core/mips 2. Add vaa_mmi.c in codec/processing/src/mips 3. Add optimization functions in codec/common/src/mc.cpp Change-Id: I9060a5f42ac7903b377b48ef7fe92809a2ba4481
--- a/codec/common/src/mc.cpp
+++ b/codec/common/src/mc.cpp
@@ -43,6 +43,7 @@
#include "cpu_core.h"
#include "ls_defines.h"
#include "macros.h"
+#include "asmdefs_mmi.h"
namespace {@@ -1659,6 +1660,2541 @@
}
#endif
+#if defined(HAVE_MMI)
+#define MMI_LOAD_8P(f0, f2, f4, r0) \
+ "gsldlc1 "#f0", 0x7("#r0") \n\t" \+ "gsldrc1 "#f0", 0x0("#r0") \n\t" \+ "punpckhbh "#f2", "#f0", "#f4" \n\t" \
+ "punpcklbh "#f0", "#f0", "#f4" \n\t"
+
+#define FILTER_HV_W4(f0, f2, f4, f6, f8, f10, f12, f14, f16, f18, \
+ f20, f22, f24, f26, f28, f30, r0, r1, r2) \
+ "paddh "#f0", "#f0", "#f20" \n\t" \
+ "paddh "#f2", "#f2", "#f22" \n\t" \
+ "mov.d "#f28", "#f8" \n\t" \
+ "mov.d "#f30", "#f10" \n\t" \
+ "mov.d "#f24", "#f4" \n\t" \
+ "mov.d "#f26", "#f6" \n\t" \
+ "dmfc1 "#r2", "#f8" \n\t" \
+ "dli "#r1", 0x0010001000100010 \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "paddh "#f0", "#f0", "#f8" \n\t" \
+ "paddh "#f2", "#f2", "#f8" \n\t" \
+ "paddh "#f28", "#f28", "#f12" \n\t" \
+ "paddh "#f30", "#f30", "#f14" \n\t" \
+ "paddh "#f24", "#f24", "#f16" \n\t" \
+ "paddh "#f26", "#f26", "#f18" \n\t" \
+ "dli "#r1", 0x2 \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "psllh "#f28", "#f28", "#f8" \n\t" \
+ "psllh "#f30", "#f30", "#f8" \n\t" \
+ "psubh "#f28", "#f28", "#f24" \n\t" \
+ "psubh "#f30", "#f30", "#f26" \n\t" \
+ "paddh "#f0", "#f0", "#f28" \n\t" \
+ "paddh "#f2", "#f2", "#f30" \n\t" \
+ "psllh "#f28", "#f28", "#f8" \n\t" \
+ "psllh "#f30", "#f30", "#f8" \n\t" \
+ "paddh "#f0", "#f0", "#f28" \n\t" \
+ "paddh "#f2", "#f2", "#f30" \n\t" \
+ "dli "#r1", 0x5 \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "psrah "#f0", "#f0", "#f8" \n\t" \
+ "psrah "#f2", "#f2", "#f8" \n\t" \
+ "xor "#f28", "#f28", "#f28" \n\t" \
+ "packushb "#f0", "#f0", "#f2" \n\t" \
+ "gsswlc1 "#f0", 0x3("#r0") \n\t" \+ "gsswrc1 "#f0", 0x0("#r0") \n\t" \+ "dmtc1 "#r2", "#f8" \n\t"
+
+#define FILTER_HV_W8(f0, f2, f4, f6, f8, f10, f12, f14, f16, f18, \
+ f20, f22, f24, f26, f28, f30, r0, r1, r2) \
+ "paddh "#f0", "#f0", "#f20" \n\t" \
+ "paddh "#f2", "#f2", "#f22" \n\t" \
+ "mov.d "#f28", "#f8" \n\t" \
+ "mov.d "#f30", "#f10" \n\t" \
+ "mov.d "#f24", "#f4" \n\t" \
+ "mov.d "#f26", "#f6" \n\t" \
+ "dmfc1 "#r2", "#f8" \n\t" \
+ "dli "#r1", 0x0010001000100010 \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "paddh "#f0", "#f0", "#f8" \n\t" \
+ "paddh "#f2", "#f2", "#f8" \n\t" \
+ "paddh "#f28", "#f28", "#f12" \n\t" \
+ "paddh "#f30", "#f30", "#f14" \n\t" \
+ "paddh "#f24", "#f24", "#f16" \n\t" \
+ "paddh "#f26", "#f26", "#f18" \n\t" \
+ "dli "#r1", 0x2 \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "psllh "#f28", "#f28", "#f8" \n\t" \
+ "psllh "#f30", "#f30", "#f8" \n\t" \
+ "psubh "#f28", "#f28", "#f24" \n\t" \
+ "psubh "#f30", "#f30", "#f26" \n\t" \
+ "paddh "#f0", "#f0", "#f28" \n\t" \
+ "paddh "#f2", "#f2", "#f30" \n\t" \
+ "psllh "#f28", "#f28", "#f8" \n\t" \
+ "psllh "#f30", "#f30", "#f8" \n\t" \
+ "paddh "#f0", "#f0", "#f28" \n\t" \
+ "paddh "#f2", "#f2", "#f30" \n\t" \
+ "dli "#r1", 0x5 \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "psrah "#f0", "#f0", "#f8" \n\t" \
+ "psrah "#f2", "#f2", "#f8" \n\t" \
+ "xor "#f28", "#f28", "#f28" \n\t" \
+ "packushb "#f0", "#f0", "#f2" \n\t" \
+ "gssdlc1 "#f0", 0x7("#r0") \n\t" \+ "gssdrc1 "#f0", 0x0("#r0") \n\t" \+ "dmtc1 "#r2", "#f8" \n\t"
+
+#define FILTER_VER_ALIGN(f0, f2, f4, f6, f8, f10, f12, f14, f16, f18, \
+ f20, f22, f24, f26, f28, f30, r0, r1, r2, r3, r4) \
+ "paddh "#f0", "#f0", "#f20" \n\t" \
+ "paddh "#f2", "#f2", "#f22" \n\t" \
+ "mov.d "#f24", "#f4" \n\t" \
+ "mov.d "#f26", "#f6" \n\t" \
+ "mov.d "#f28", "#f8" \n\t" \
+ "mov.d "#f30", "#f10" \n\t" \
+ "dli "#r2", 0x2 \n\t" \
+ "paddh "#f24", "#f24", "#f16" \n\t" \
+ "paddh "#f26", "#f26", "#f18" \n\t" \
+ "dmfc1 "#r3", "#f8" \n\t" \
+ "paddh "#f28", "#f28", "#f12" \n\t" \
+ "paddh "#f30", "#f30", "#f14" \n\t" \
+ "dmtc1 "#r2", "#f8" \n\t" \
+ "psubh "#f0", "#f0", "#f24" \n\t" \
+ "psubh "#f2", "#f2", "#f26" \n\t" \
+ "psrah "#f0", "#f0", "#f8" \n\t" \
+ "psrah "#f2", "#f2", "#f8" \n\t" \
+ "paddh "#f0", "#f0", "#f28" \n\t" \
+ "paddh "#f2", "#f2", "#f30" \n\t" \
+ "psubh "#f0", "#f0", "#f24" \n\t" \
+ "psubh "#f2", "#f2", "#f26" \n\t" \
+ "psrah "#f0", "#f0", "#f8" \n\t" \
+ "psrah "#f2", "#f2", "#f8" \n\t" \
+ "dmtc1 "#r4", "#f8" \n\t" \
+ "paddh "#f28", "#f28", "#f0" \n\t" \
+ "paddh "#f30", "#f30", "#f2" \n\t" \
+ "dli "#r2", 0x6 \n\t" \
+ "paddh "#f28", "#f28", "#f8" \n\t" \
+ "paddh "#f30", "#f30", "#f8" \n\t" \
+ "dmtc1 "#r2", "#f8" \n\t" \
+ "psrah "#f28", "#f28", "#f8" \n\t" \
+ "psrah "#f30", "#f30", "#f8" \n\t" \
+ "packushb "#f28", "#f28", "#f30" \n\t" \
+ "gssdxc1 "#f28", 0x0("#r0", "#r1") \n\t" \+ "dmtc1 "#r3", "#f8" \n\t"
+
+#define FILTER_VER_UNALIGN(f0, f2, f4, f6, f8, f10, f12, f14, f16, f18, \
+ f20, f22, f24, f26, f28, f30, r0, r1, r2, r3) \
+ "paddh "#f0", "#f0", "#f20" \n\t" \
+ "paddh "#f2", "#f2", "#f22" \n\t" \
+ "mov.d "#f24", "#f4" \n\t" \
+ "mov.d "#f26", "#f6" \n\t" \
+ "mov.d "#f28", "#f8" \n\t" \
+ "mov.d "#f30", "#f10" \n\t" \
+ "dli "#r1", 0x2 \n\t" \
+ "paddh "#f24", "#f24", "#f16" \n\t" \
+ "paddh "#f26", "#f26", "#f18" \n\t" \
+ "dmfc1 "#r2", "#f8" \n\t" \
+ "paddh "#f28", "#f28", "#f12" \n\t" \
+ "paddh "#f30", "#f30", "#f14" \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "psubh "#f0", "#f0", "#f24" \n\t" \
+ "psubh "#f2", "#f2", "#f26" \n\t" \
+ "psrah "#f0", "#f0", "#f8" \n\t" \
+ "psrah "#f2", "#f2", "#f8" \n\t" \
+ "paddh "#f0", "#f0", "#f28" \n\t" \
+ "paddh "#f2", "#f2", "#f30" \n\t" \
+ "psubh "#f0", "#f0", "#f24" \n\t" \
+ "psubh "#f2", "#f2", "#f26" \n\t" \
+ "psrah "#f0", "#f0", "#f8" \n\t" \
+ "psrah "#f2", "#f2", "#f8" \n\t" \
+ "dmtc1 "#r3", "#f8" \n\t" \
+ "paddh "#f28", "#f28", "#f0" \n\t" \
+ "paddh "#f30", "#f30", "#f2" \n\t" \
+ "dli "#r1", 0x6 \n\t" \
+ "paddh "#f28", "#f28", "#f8" \n\t" \
+ "paddh "#f30", "#f30", "#f8" \n\t" \
+ "dmtc1 "#r1", "#f8" \n\t" \
+ "psrah "#f28", "#f28", "#f8" \n\t" \
+ "psrah "#f30", "#f30", "#f8" \n\t" \
+ "packushb "#f28", "#f28", "#f30" \n\t" \
+ "gssdlc1 "#f28", 0x7("#r0") \n\t" \+ "gssdrc1 "#f28", 0x0("#r0") \n\t" \+ "dmtc1 "#r2", "#f8" \n\t"
+
+void McHorVer20Width5_mmi(const uint8_t *pSrc, int32_t iSrcStride, uint8_t *pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], -0x2 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dli $10, 0x0010001000100010 \n\t"
+ "dli $11, 0x5 \n\t"
+ "1: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "mov.d $f28, $f8 \n\t"
+ "mov.d $f30, $f10 \n\t"
+ "paddh $f28, $f28, $f12 \n\t"
+ "paddh $f30, $f30, $f14 \n\t"
+ "mov.d $f24, $f16 \n\t"
+ "mov.d $f26, $f18 \n\t"
+ "paddh $f24, $f24, $f20 \n\t"
+ "paddh $f26, $f26, $f22 \n\t"
+ "dmfc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "psubh $f24, $f24, $f28 \n\t"
+ "psubh $f26, $f26, $f30 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+
+ "dmtc1 $10, $f12 \n\t"
+ "paddh $f0, $f0, $f12 \n\t"
+ "paddh $f2, $f2, $f12 \n\t"
+ "dmtc1 $11, $f12 \n\t"
+ "psrah $f0, $f0, $f12 \n\t"
+ "psrah $f2, $f2, $f12 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+
+ "gsswlc1 $f0, 0x3(%[pDst]) \n\t"
+ "gsswrc1 $f0, 0x0(%[pDst]) \n\t"
+
+ "gsldlc1 $f0, 0xd(%[pSrc]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x6(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "dmtc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+
+ "paddh $f16, $f16, $f4 \n\t"
+ "paddh $f18, $f18, $f6 \n\t"
+ "paddh $f20, $f20, $f12 \n\t"
+ "paddh $f22, $f22, $f14 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "psubh $f20, $f20, $f16 \n\t"
+ "psubh $f22, $f22, $f18 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+
+ "dmtc1 $10, $f24 \n\t"
+ "paddh $f8, $f8, $f24 \n\t"
+ "paddh $f10, $f10, $f24 \n\t"
+ "dmtc1 $11, $f24 \n\t"
+ "psrah $f8, $f8, $f24 \n\t"
+ "psrah $f10, $f10, $f24 \n\t"
+ "packushb $f8, $f8, $f10 \n\t"
+ "gsswlc1 $f8, 0x4(%[pDst]) \n\t"
+ "gsswrc1 $f8, 0x1(%[pDst]) \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"((unsigned char *)pSrc), [pDst]"+&r"((unsigned char *)pDst),
+ [iWidth]"+&r"((int)iWidth), [iHeight]"+&r"((int)iHeight)
+ : [iSrcStride]"r"((int)iSrcStride), [iDstStride]"r"((int)iDstStride)
+ : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8",
+ "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26",
+ "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void McHorVer20Width9Or17_mmi(const uint8_t *pSrc, int32_t iSrcStride, uint8_t *pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], -0x2 \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dli $9, 0x9 \n\t"
+ "dli $10, 0x0010001000100010 \n\t"
+ "dli $11, 0x5 \n\t"
+ "bne %[iWidth], $9, 2f \n\t"
+ "1: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "mov.d $f28, $f8 \n\t"
+ "mov.d $f30, $f10 \n\t"
+ "paddh $f28, $f28, $f12 \n\t"
+ "paddh $f30, $f30, $f14 \n\t"
+ "mov.d $f24, $f16 \n\t"
+ "mov.d $f26, $f18 \n\t"
+ "paddh $f24, $f24, $f20 \n\t"
+ "paddh $f26, $f26, $f22 \n\t"
+ "dmfc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "psubh $f24, $f24, $f28 \n\t"
+ "psubh $f26, $f26, $f30 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+
+ "dmtc1 $10, $f12 \n\t"
+ "paddh $f0, $f0, $f12 \n\t"
+ "paddh $f2, $f2, $f12 \n\t"
+ "dmtc1 $11, $f12 \n\t"
+ "psrah $f0, $f0, $f12 \n\t"
+ "psrah $f2, $f2, $f12 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+
+ "gsswlc1 $f0, 0x3(%[pDst]) \n\t"
+ "gsswrc1 $f0, 0x0(%[pDst]) \n\t"
+
+ "gsldlc1 $f0, 0xd(%[pSrc]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x6(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "dmtc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+
+ "paddh $f16, $f16, $f4 \n\t"
+ "paddh $f18, $f18, $f6 \n\t"
+ "paddh $f20, $f20, $f12 \n\t"
+ "paddh $f22, $f22, $f14 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "psubh $f20, $f20, $f16 \n\t"
+ "psubh $f22, $f22, $f18 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+
+ "dmtc1 $10, $f24 \n\t"
+ "paddh $f8, $f8, $f24 \n\t"
+ "paddh $f10, $f10, $f24 \n\t"
+ "dmtc1 $11, $f24 \n\t"
+ "psrah $f8, $f8, $f24 \n\t"
+ "psrah $f10, $f10, $f24 \n\t"
+ "packushb $f8, $f8, $f10 \n\t"
+ "gssdlc1 $f8, 0x8(%[pDst]) \n\t"
+ "gssdrc1 $f8, 0x1(%[pDst]) \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ "j 3f \n\t"
+
+ "2: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "dmtc1 $8, $f30 \n\t"
+ "paddh $f8, $f8, $f12 \n\t"
+ "paddh $f10, $f10, $f14 \n\t"
+ "paddh $f16, $f16, $f20 \n\t"
+ "paddh $f18, $f18, $f22 \n\t"
+ "psllh $f16, $f16, $f30 \n\t"
+ "psllh $f18, $f18, $f30 \n\t"
+ "psubh $f16, $f16, $f8 \n\t"
+ "psubh $f18, $f18, $f10 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "psllh $f16, $f16, $f30 \n\t"
+ "psllh $f18, $f18, $f30 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+
+ "dmtc1 $10, $f30 \n\t"
+ "paddh $f0, $f0, $f30 \n\t"
+ "paddh $f2, $f2, $f30 \n\t"
+ "dmtc1 $11, $f30 \n\t"
+ "psrah $f0, $f0, $f30 \n\t"
+ "psrah $f2, $f2, $f30 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+
+ "gsldlc1 $f0, 15(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0x14(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x10(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0x13(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x11(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0x12(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 8(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0xd(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x9(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0xc(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0xb(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "mov.d $f28, $f8 \n\t"
+ "mov.d $f30, $f10 \n\t"
+ "paddh $f28, $f28, $f12 \n\t"
+ "paddh $f30, $f30, $f14 \n\t"
+ "mov.d $f24, $f16 \n\t"
+ "mov.d $f26, $f18 \n\t"
+ "paddh $f24, $f24, $f20 \n\t"
+ "paddh $f26, $f26, $f22 \n\t"
+ "dmfc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "psubh $f24, $f24, $f28 \n\t"
+ "psubh $f26, $f26, $f30 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+
+ "dmtc1 $10, $f30 \n\t"
+ "paddh $f0, $f0, $f30 \n\t"
+ "paddh $f2, $f2, $f30 \n\t"
+ "dmtc1 $11, $f30 \n\t"
+ "psrah $f0, $f0, $f30 \n\t"
+ "psrah $f2, $f2, $f30 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+ "gsswlc1 $f0, 0xb(%[pDst]) \n\t"
+ "gsswrc1 $f0, 0x8(%[pDst]) \n\t"
+
+ "dmtc1 $9, $f12 \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "dli $9, 0x20 \n\t"
+ "gsldlc1 $f0, 0x15(%[pSrc]) \n\t"
+ "dmtc1 $9, $f30 \n\t"
+ "gsldrc1 $f0, 0xE(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+
+ "paddh $f16, $f16, $f4 \n\t"
+ "paddh $f18, $f18, $f6 \n\t"
+ "paddh $f20, $f20, $f12 \n\t"
+ "paddh $f22, $f22, $f14 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "psubh $f20, $f20, $f16 \n\t"
+ "psubh $f22, $f22, $f18 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+
+ "dmtc1 $10, $f24 \n\t"
+ "paddh $f8, $f8, $f24 \n\t"
+ "paddh $f10, $f10, $f24 \n\t"
+ "dmtc1 $11, $f24 \n\t"
+ "psrah $f8, $f8, $f24 \n\t"
+ "psrah $f10, $f10, $f24 \n\t"
+ "packushb $f8, $f8, $f10 \n\t"
+ "gssdlc1 $f8, 0x10(%[pDst]) \n\t"
+ "gssdrc1 $f8, 0x9(%[pDst]) \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 2b \n\t"
+ "3: \n\t"
+ : [pSrc]"+&r"((unsigned char *)pSrc), [pDst]"+&r"((unsigned char *)pDst),
+ [iWidth]"+&r"((int)iWidth), [iHeight]"+&r"((int)iHeight)
+ : [iSrcStride]"r"((int)iSrcStride), [iDstStride]"r"((int)iDstStride)
+ : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8",
+ "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26",
+ "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+//horizontal filter to gain half sample, that is (2, 0) location in quarter sample
+static inline void McHorVer20Width5Or9Or17_mmi(const uint8_t* pSrc, int32_t iSrcStride,
+ uint8_t* pDst, int32_t iDstStride,
+ int32_t iWidth, int32_t iHeight) {+ if (iWidth == 17 || iWidth == 9)
+ McHorVer20Width9Or17_mmi(pSrc, iSrcStride, pDst, iDstStride, iWidth, iHeight);
+ else //if (iWidth == 5)
+ McHorVer20Width5_mmi(pSrc, iSrcStride, pDst, iDstStride, iWidth, iHeight);
+}
+
+void McHorVer02Height5_mmi(const uint8_t *pSrc, int32_t iSrcStride, uint8_t *pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $12, %[pSrc] \n\t"
+ "move $13, %[pDst] \n\t"
+ "move $14, %[iHeight] \n\t"
+
+ "dsrl %[iWidth], %[iWidth], 0x2 \n\t"
+ PTR_ADDU "$10, %[iSrcStride], %[iSrcStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], $10 \n\t"
+
+ "1: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ MMI_LOAD_8P($f0, $f2, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f4, $f6, $f28, $8)
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f8, $f10, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f12, $f14, $f28, $8)
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f16, $f18, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f20, $f22, $f28, $8)
+ FILTER_HV_W4($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f24, $f26, $f28, %[pSrc])
+ "mov.d $f0, $f4 \n\t"
+ "mov.d $f2, $f6 \n\t"
+ "mov.d $f4, $f8 \n\t"
+ "mov.d $f6, $f10 \n\t"
+ "mov.d $f8, $f12 \n\t"
+ "mov.d $f10, $f14 \n\t"
+ "mov.d $f12, $f16 \n\t"
+ "mov.d $f14, $f18 \n\t"
+ "mov.d $f16, $f20 \n\t"
+ "mov.d $f18, $f22 \n\t"
+ "mov.d $f20, $f24 \n\t"
+ "mov.d $f22, $f26 \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+
+ "2: \n\t"
+ FILTER_HV_W4($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f24, $f26, $f28, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W4($f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24,
+ $f26, $f28, $f30, $f0, $f2, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f28, $f30, $f0, $8)
+ FILTER_HV_W4($f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28,
+ $f30, $f0, $f2, $f4, $f6, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f0, $f2, $f4, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W4($f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0,
+ $f2, $f4, $f6, $f8, $f10, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f4, $f6, $f8, $8)
+ FILTER_HV_W4($f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6,
+ $f8, $f10, $f12, $f14, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f8, $f10, $f12, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W4($f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10,
+ $f12, $f14, $f16, $f18, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f12, $f14, $f16, $8)
+ FILTER_HV_W4($f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14,
+ $f16, $f18, $f20, $f22, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f16, $f18, $f20, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W4($f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18,
+ $f20, $f22, $f24, $f26, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f20, $f22, $f24, $8)
+ "j 2b \n\t"
+
+ "3: \n\t"
+ PTR_ADDIU "%[iWidth], %[iWidth], -0x1 \n\t"
+ "beqz %[iWidth], 4f \n\t"
+ "move %[pSrc], $12 \n\t"
+ "move %[pDst], $13 \n\t"
+ "move %[iHeight], $14 \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], $10 \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], 0x4 \n\t"
+ PTR_ADDIU "%[pDst], %[pDst], 0x4 \n\t"
+ "j 1b \n\t"
+ "4: \n\t"
+ : [pSrc]"+&r"((unsigned char *)pSrc), [pDst]"+&r"((unsigned char *)pDst),
+ [iWidth]"+&r"(iWidth), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$9", "$10", "$12", "$13", "$14", "$f0", "$f2", "$f4",
+ "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22",
+ "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void McHorVer02Height9Or17_mmi(const uint8_t *pSrc, int32_t iSrcStride, uint8_t *pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $12, %[pSrc] \n\t"
+ "move $13, %[pDst] \n\t"
+ "move $14, %[iHeight] \n\t"
+
+ "dsrl %[iWidth], %[iWidth], 0x3 \n\t"
+ PTR_ADDU "$10, %[iSrcStride], %[iSrcStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], $10 \n\t"
+
+ "1: \n\t"
+ "dli $8, 0x20 \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "dmtc1 $8, $f30 \n\t"
+
+ MMI_LOAD_8P($f0, $f2, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f4, $f6, $f28, $8)
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f8, $f10, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f12, $f14, $f28, $8)
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f16, $f18, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f20, $f22, $f28, $8)
+ FILTER_HV_W8($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f24, $f26, $f28, %[pSrc])
+ "mov.d $f0, $f4 \n\t"
+ "mov.d $f2, $f6 \n\t"
+ "mov.d $f4, $f8 \n\t"
+ "mov.d $f6, $f10 \n\t"
+ "mov.d $f8, $f12 \n\t"
+ "mov.d $f10, $f14 \n\t"
+ "mov.d $f12, $f16 \n\t"
+ "mov.d $f14, $f18 \n\t"
+ "mov.d $f16, $f20 \n\t"
+ "mov.d $f18, $f22 \n\t"
+ "mov.d $f20, $f24 \n\t"
+ "mov.d $f22, $f26 \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+
+ "2: \n\t"
+ FILTER_HV_W8($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9)
+ "dmtc1 $9, $f8 \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f24, $f26, $f28, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24,
+ $f26, $f28, $f30, $f0, $f2, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f28, $f30, $f0, $8)
+ FILTER_HV_W8($f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28,
+ $f30, $f0, $f2, $f4, $f6, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f0, $f2, $f4, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0,
+ $f2, $f4, $f6, $f8, $f10, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f4, $f6, $f8, $8)
+ FILTER_HV_W8($f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4,
+ $f6, $f8, $f10, $f12, $f14, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f8, $f10, $f12, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8,
+ $f10, $f12, $f14, $f16, $f18, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f12, $f14, $f16, $8)
+ FILTER_HV_W8($f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12,
+ $f14, $f16, $f18, $f20, $f22, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], $10 \n\t"
+ MMI_LOAD_8P($f16, $f18, $f20, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16,
+ $f18, $f20, $f22, $f24, $f26, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 3f \n\t"
+
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f20, $f22, $f24, $8)
+ "j 2b \n\t"
+
+ "3: \n\t"
+ PTR_ADDIU "%[iWidth], %[iWidth], -0x1 \n\t"
+ "beqz %[iWidth], 4f \n\t"
+
+ "move %[pSrc], $12 \n\t"
+ "move %[pDst], $13 \n\t"
+ "move %[iHeight], $14 \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], $10 \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], 0x8 \n\t"
+ PTR_ADDIU "%[pDst], %[pDst], 0x8 \n\t"
+ "j 1b \n\t"
+ "4: \n\t"
+ : [pSrc]"+&r"((unsigned char *)pSrc), [pDst]"+&r"((unsigned char *)pDst),
+ [iWidth]"+&r"(iWidth), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$9", "$10", "$12", "$13", "$14", "$f0", "$f2", "$f4",
+ "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22",
+ "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+//vertical filter to gain half sample, that is (0, 2) location in quarter sample
+static inline void McHorVer02Height5Or9Or17_mmi(const uint8_t* pSrc, int32_t iSrcStride,
+ uint8_t* pDst, int32_t iDstStride,
+ int32_t iWidth, int32_t iHeight) {+ if (iWidth == 16 || iWidth == 8)
+ McHorVer02Height9Or17_mmi(pSrc, iSrcStride, pDst, iDstStride, iWidth, iHeight );
+ else
+ McHorVer02Height5_mmi (pSrc, iSrcStride, pDst, iDstStride, iWidth, iHeight);
+}
+
+static inline void McHorVer22HorFirst_mmi(const uint8_t *pSrc, int32_t iSrcStride,
+ uint8_t * pTap, int32_t iTapStride,
+ int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "dli $8, 0x9 \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ "bne %[iWidth], $8, 2f \n\t"
+
+ "1: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "mov.d $f28, $f8 \n\t"
+ "mov.d $f30, $f10 \n\t"
+ "paddh $f28, $f28, $f12 \n\t"
+ "paddh $f30, $f30, $f14 \n\t"
+ "mov.d $f24, $f16 \n\t"
+ "mov.d $f26, $f18 \n\t"
+ "paddh $f24, $f24, $f20 \n\t"
+ "paddh $f26, $f26, $f22 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmfc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "psubh $f24, $f24, $f28 \n\t"
+ "psubh $f26, $f26, $f30 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "gsswlc1 $f0, 0x3(%[pTap]) \n\t"
+ "gsswrc1 $f0, 0x0(%[pTap]) \n\t"
+
+ "gsldlc1 $f0, 0xd(%[pSrc]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x6(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+
+ "paddh $f16, $f16, $f4 \n\t"
+ "paddh $f18, $f18, $f6 \n\t"
+ "paddh $f20, $f20, $f12 \n\t"
+ "paddh $f22, $f22, $f14 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "psubh $f20, $f20, $f16 \n\t"
+ "psubh $f22, $f22, $f18 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "gssdlc1 $f8, 0x9(%[pTap]) \n\t"
+ "gssdlc1 $f10, 0x11(%[pTap]) \n\t"
+ "gssdrc1 $f8, 0x2(%[pTap]) \n\t"
+ "gssdrc1 $f10, 0xa(%[pTap]) \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pTap], %[pTap], %[iTapStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ "j 3f \n\t"
+
+ "2: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "dli $8, 0x2 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "dmtc1 $8, $f30 \n\t"
+ "paddh $f8, $f8, $f12 \n\t"
+ "paddh $f10, $f10, $f14 \n\t"
+ "paddh $f16, $f16, $f20 \n\t"
+ "paddh $f18, $f18, $f22 \n\t"
+ "psllh $f16, $f16, $f30 \n\t"
+ "psllh $f18, $f18, $f30 \n\t"
+ "psubh $f16, $f16, $f8 \n\t"
+ "psubh $f18, $f18, $f10 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "psllh $f16, $f16, $f30 \n\t"
+ "psllh $f18, $f18, $f30 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pTap]) \n\t"
+
+ "gsldlc1 $f0, 15(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 8(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "gsldlc1 $f4, 0x14(%[pSrc]) \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "gsldrc1 $f4, 0xd(%[pSrc]) \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "gsldlc1 $f8, 0x10(%[pSrc]) \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "gsldrc1 $f8, 0x9(%[pSrc]) \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "gsldlc1 $f12, 0x13(%[pSrc]) \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "gsldrc1 $f12, 0xc(%[pSrc]) \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "gsldlc1 $f16, 0x11(%[pSrc]) \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "gsldrc1 $f16, 0xa(%[pSrc]) \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "gsldlc1 $f20, 0x12(%[pSrc]) \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "gsldrc1 $f20, 0xb(%[pSrc]) \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+
+ "mov.d $f28, $f8 \n\t"
+ "mov.d $f30, $f10 \n\t"
+ "paddh $f28, $f28, $f12 \n\t"
+ "paddh $f30, $f30, $f14 \n\t"
+ "mov.d $f24, $f16 \n\t"
+ "mov.d $f26, $f18 \n\t"
+ "dli $8, 0x2 \n\t"
+ "paddh $f24, $f24, $f20 \n\t"
+ "paddh $f26, $f26, $f22 \n\t"
+ "dmfc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "psubh $f24, $f24, $f28 \n\t"
+ "psubh $f26, $f26, $f30 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "psllh $f24, $f24, $f12 \n\t"
+ "psllh $f26, $f26, $f12 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "gsswlc1 $f0, 0x13(%[pTap]) \n\t"
+ "gsswrc1 $f0, 0x10(%[pTap]) \n\t"
+
+ "gsldlc1 $f0, 0x15(%[pSrc]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0xE(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $9, $f12 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+
+ "paddh $f16, $f16, $f4 \n\t"
+ "paddh $f18, $f18, $f6 \n\t"
+ "paddh $f20, $f20, $f12 \n\t"
+ "paddh $f22, $f22, $f14 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "psubh $f20, $f20, $f16 \n\t"
+ "psubh $f22, $f22, $f18 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "psllh $f20, $f20, $f24 \n\t"
+ "psllh $f22, $f22, $f24 \n\t"
+ "paddh $f8, $f8, $f20 \n\t"
+ "paddh $f10, $f10, $f22 \n\t"
+ "gssdlc1 $f8, 0x19(%[pTap]) \n\t"
+ "gssdlc1 $f10, 0x21(%[pTap]) \n\t"
+ "gssdrc1 $f8, 0x12(%[pTap]) \n\t"
+ "gssdrc1 $f10, 0x1a(%[pTap]) \n\t"
+
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pTap], %[pTap], %[iTapStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 2b \n\t"
+ "3: \n\t"
+ : [pSrc]"+&r"(pSrc), [pTap]"+&r"(pTap), [iWidth]"+&r"(iWidth),
+ [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iTapStride]"r"(iTapStride)
+ : "memory", "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+static inline void McHorVer22Width8VerLastAlign_mmi(const uint8_t *pTap,
+ int32_t iTapStride, uint8_t * pDst, int32_t iDstStride,
+ int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $10, %[pTap] \n\t"
+ "move $11, %[pDst] \n\t"
+ "move $12, %[iHeight] \n\t"
+ "dsrl %[iWidth], 0x3 \n\t"
+ PTR_ADDU "$13, %[iTapStride], %[iTapStride] \n\t"
+ PTR_ADDU "$14, %[iDstStride], %[iDstStride] \n\t"
+ "dli $15, 0x0020002000200020 \n\t"
+
+ "4: \n\t"
+ "gslqc1 $f2, $f0, 0x0(%[pTap]) \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f6, $f4, 0x0($8) \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f10, $f8, 0x0(%[pTap]) \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f14, $f12, 0x0($8) \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f18, $f16, 0x0(%[pTap]) \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f22, $f20, 0x0($8) \n\t"
+
+ FILTER_VER_ALIGN($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $0, $8, $9, $15)
+
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f26, $f24, 0x0(%[pTap]) \n\t"
+ "mov.d $f0, $f4 \n\t"
+ "mov.d $f2, $f6 \n\t"
+ "mov.d $f4, $f8 \n\t"
+ "mov.d $f6, $f10 \n\t"
+ "mov.d $f8, $f12 \n\t"
+ "mov.d $f10, $f14 \n\t"
+ "mov.d $f12, $f16 \n\t"
+ "mov.d $f14, $f18 \n\t"
+ "mov.d $f16, $f20 \n\t"
+ "mov.d $f18, $f22 \n\t"
+ "mov.d $f20, $f24 \n\t"
+ "mov.d $f22, $f26 \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_SUBU "%[pTap], %[pTap], %[iTapStride] \n\t"
+
+ "5: \n\t"
+ FILTER_VER_ALIGN($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $0, $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f26, $f24, 0x0(%[pTap]) \n\t"
+
+ FILTER_VER_ALIGN($f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24,
+ $f26, $f28, $f30, $f0, $f2, %[pDst], %[iDstStride], $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], $14 \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f30, $f28, 0x0($8) \n\t"
+
+ FILTER_VER_ALIGN($f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28,
+ $f30, $f0, $f2, $f4, $f6, %[pDst], $0, $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f2, $f0, 0x0(%[pTap]) \n\t"
+
+ FILTER_VER_ALIGN($f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0,
+ $f2, $f4, $f6, $f8, $f10, %[pDst], %[iDstStride], $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], $14 \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f6, $f4, 0x0($8) \n\t"
+
+ FILTER_VER_ALIGN($f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4,
+ $f6, $f8, $f10, $f12, $f14, %[pDst], $0, $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f10, $f8, 0x0(%[pTap]) \n\t"
+
+ FILTER_VER_ALIGN($f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8,
+ $f10, $f12, $f14, $f16, $f18, %[pDst], %[iDstStride], $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], $14 \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f14, $f12, 0x0($8) \n\t"
+
+ FILTER_VER_ALIGN($f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12,
+ $f14, $f16, $f18, $f20, $f22, %[pDst], $0, $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gslqc1 $f18, $f16, 0x0(%[pTap]) \n\t"
+
+ FILTER_VER_ALIGN($f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16,
+ $f18, $f20, $f22, $f24, $f26, %[pDst], %[iDstStride], $8, $9, $15)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], $14 \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gslqc1 $f22, $f20, 0x0($8) \n\t"
+ "j 5b \n\t"
+
+ "6: \n\t"
+ PTR_ADDIU "%[iWidth], %[iWidth], -0x1 \n\t"
+ "beqz %[iWidth], 7f \n\t"
+ "move %[pTap], $10 \n\t"
+ "move %[pDst], $11 \n\t"
+ "move %[iHeight], $12 \n\t"
+ PTR_ADDIU "%[pTap], %[pTap], 0x10 \n\t"
+ PTR_ADDIU "%[pDst], %[pDst], 0x8 \n\t"
+ "j 4b \n\t"
+ "7: \n\t"
+ : [pTap]"+&r"((unsigned char *)pTap), [pDst]"+&r"((unsigned char *)pDst),
+ [iWidth]"+&r"((int)iWidth), [iHeight]"+&r"((int)iHeight)
+ : [iTapStride]"r"((int)iTapStride), [iDstStride]"r"((int)iDstStride)
+ : "memory", "$8", "$9", "$10", "$11", "$12", "$13", "$14", "$15", "$f0",
+ "$f2", "$f4", "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18",
+ "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+static inline void McHorVer22Width8VerLastUnAlign_mmi(const uint8_t *pTap,
+ int32_t iTapStride, uint8_t * pDst, int32_t iDstStride,
+ int32_t iWidth, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $10, %[pTap] \n\t"
+ "move $11, %[pDst] \n\t"
+ "move $12, %[iHeight] \n\t"
+ "dsrl %[iWidth], 0x3 \n\t"
+ PTR_ADDU "$13, %[iTapStride], %[iTapStride] \n\t"
+ "dli $14, 0x0020002000200020 \n\t"
+
+ "4: \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f0, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f2, 0xF(%[pTap]) \n\t"
+ "gsldlc1 $f4, 0x7($8) \n\t"
+ "gsldlc1 $f6, 0xF($8) \n\t"
+ "gsldrc1 $f0, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f2, 0x8(%[pTap]) \n\t"
+ "gsldrc1 $f4, 0x0($8) \n\t"
+ "gsldrc1 $f6, 0x8($8) \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f8, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f10, 0xF(%[pTap]) \n\t"
+ "gsldlc1 $f12, 0x7($8) \n\t"
+ "gsldlc1 $f14, 0xF($8) \n\t"
+ "gsldrc1 $f8, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f10, 0x8(%[pTap]) \n\t"
+ "gsldrc1 $f12, 0x0($8) \n\t"
+ "gsldrc1 $f14, 0x8($8) \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f16, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f18, 0xF(%[pTap]) \n\t"
+ "gsldlc1 $f20, 0x7($8) \n\t"
+ "gsldlc1 $f22, 0xF($8) \n\t"
+ "gsldrc1 $f16, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f18, 0x8(%[pTap]) \n\t"
+ "gsldrc1 $f20, 0x0($8) \n\t"
+ "gsldrc1 $f22, 0x8($8) \n\t"
+
+ FILTER_VER_UNALIGN($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18,
+ $f20, $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gsldlc1 $f24, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f26, 0xF(%[pTap]) \n\t"
+ "gsldrc1 $f24, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f26, 0x8(%[pTap]) \n\t"
+ "mov.d $f0, $f4 \n\t"
+ "mov.d $f2, $f6 \n\t"
+ "mov.d $f4, $f8 \n\t"
+ "mov.d $f6, $f10 \n\t"
+ "mov.d $f8, $f12 \n\t"
+ "mov.d $f10, $f14 \n\t"
+ "mov.d $f12, $f16 \n\t"
+ "mov.d $f14, $f18 \n\t"
+ "mov.d $f16, $f20 \n\t"
+ "mov.d $f18, $f22 \n\t"
+ "mov.d $f20, $f24 \n\t"
+ "mov.d $f22, $f26 \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_SUBU "%[pTap], %[pTap], %[iTapStride] \n\t"
+
+ "5: \n\t"
+ FILTER_VER_UNALIGN($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18,
+ $f20, $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9, $14)
+
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gsldlc1 $f24, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f26, 0xF(%[pTap]) \n\t"
+ "gsldrc1 $f24, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f26, 0x8(%[pTap]) \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+
+ FILTER_VER_UNALIGN($f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22,
+ $f24, $f26, $f28, $f30, $f0, $f2, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f28, 0x7($8) \n\t"
+ "gsldlc1 $f30, 0xF($8) \n\t"
+ "gsldrc1 $f28, 0x0($8) \n\t"
+ "gsldrc1 $f30, 0x8($8) \n\t"
+
+ FILTER_VER_UNALIGN($f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26,
+ $f28, $f30, $f0, $f2, $f4, $f6, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gsldlc1 $f0, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f2, 0xF(%[pTap]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f2, 0x8(%[pTap]) \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+
+ FILTER_VER_UNALIGN($f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28,
+ $f30, $f0, $f2, $f4, $f6, $f8, $f10, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f4, 0x7($8) \n\t"
+ "gsldlc1 $f6, 0xF($8) \n\t"
+ "gsldrc1 $f4, 0x0($8) \n\t"
+ "gsldrc1 $f6, 0x8($8) \n\t"
+
+ FILTER_VER_UNALIGN($f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2,
+ $f4, $f6, $f8, $f10, $f12, $f14, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gsldlc1 $f8, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f10, 0xF(%[pTap]) \n\t"
+ "gsldrc1 $f8, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f10, 0x8(%[pTap]) \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+
+ FILTER_VER_UNALIGN($f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6,
+ $f8, $f10, $f12, $f14, $f16, $f18, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f12, 0x7($8) \n\t"
+ "gsldlc1 $f14, 0xF($8) \n\t"
+ "gsldrc1 $f12, 0x0($8) \n\t"
+ "gsldrc1 $f14, 0x8($8) \n\t"
+
+ FILTER_VER_UNALIGN($f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10,
+ $f12, $f14, $f16, $f18, $f20, $f22, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pTap], %[pTap], $13 \n\t"
+ "gsldlc1 $f16, 0x7(%[pTap]) \n\t"
+ "gsldlc1 $f18, 0xF(%[pTap]) \n\t"
+ "gsldrc1 $f16, 0x0(%[pTap]) \n\t"
+ "gsldrc1 $f18, 0x8(%[pTap]) \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+
+ FILTER_VER_UNALIGN($f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14,
+ $f16, $f18, $f20, $f22, $f24, $f26, %[pDst], $8, $9, $14)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 6f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pTap], %[iTapStride] \n\t"
+ "gsldlc1 $f20, 0x7($8) \n\t"
+ "gsldlc1 $f22, 0xF($8) \n\t"
+ "gsldrc1 $f20, 0x0($8) \n\t"
+ "gsldrc1 $f22, 0x8($8) \n\t"
+ "j 5b \n\t"
+
+ "6: \n\t"
+ PTR_ADDIU "%[iWidth], %[iWidth], -0x1 \n\t"
+ "beqz %[iWidth], 7f \n\t"
+ "move %[pTap], $10 \n\t"
+ "move %[pDst], $11 \n\t"
+ "move %[iHeight], $12 \n\t"
+ PTR_ADDIU "%[pTap], %[pTap], 0x10 \n\t"
+ PTR_ADDIU "%[pDst], %[pDst], 0x8 \n\t"
+ "j 4b \n\t"
+
+ "7: \n\t"
+ : [pTap]"+&r"((unsigned char *)pTap), [pDst]"+&r"((unsigned char *)pDst),
+ [iWidth]"+&r"((int)iWidth), [iHeight]"+&r"((int)iHeight)
+ : [iTapStride]"r"((int)iTapStride), [iDstStride]"r"((int)iDstStride)
+ : "memory", "$8", "$9", "$10", "$11", "$12", "$13", "$14", "$f0", "$f2",
+ "$f4", "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20",
+ "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+//horizontal and vertical filter to gain half sample, that is (2, 2) location in quarter sample
+static inline void McHorVer22Width5Or9Or17Height5Or9Or17_mmi(const uint8_t* pSrc,
+ int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
+ int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_2D (int16_t, pTap, 22, 24, 16)
+
+ if (iWidth == 17 || iWidth == 9){+ int32_t tmp1 = 2 * (iWidth - 8);
+ McHorVer22HorFirst_mmi(pSrc - 2, iSrcStride, (uint8_t*)pTap, 48, iWidth, iHeight + 5);
+
+ McHorVer22Width8VerLastAlign_mmi((uint8_t*)pTap, 48, pDst, iDstStride, iWidth - 1, iHeight);
+
+ McHorVer22Width8VerLastUnAlign_mmi((uint8_t*)pTap + tmp1, 48, pDst + iWidth - 8,
+ iDstStride, 8, iHeight);
+ } else {+ int16_t iTmp[17 + 5];
+ int32_t i, j, k;
+
+ for (i = 0; i < iHeight; i++) {+ for (j = 0; j < iWidth + 5; j++) {+ iTmp[j] = FilterInput8bitWithStride_c (pSrc - 2 + j, iSrcStride);
+ }
+ for (k = 0; k < iWidth; k++) {+ pDst[k] = WelsClip1 ((HorFilterInput16bit_c (&iTmp[k]) + 512) >> 10);
+ }
+ pSrc += iSrcStride;
+ pDst += iDstStride;
+ }
+ }
+}
+
+void McCopyWidthEq4_mmi(const uint8_t *pSrc, int iSrcStride,
+ uint8_t *pDst, int iDstStride, int iHeight) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ "lwl $8, 0x3(%[pSrc]) \n\t"
+ "lwr $8, 0x0(%[pSrc]) \n\t"
+ "swl $8, 0x3(%[pDst]) \n\t"
+ "swr $8, 0x0(%[pDst]) \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8"
+ );
+}
+
+void McCopyWidthEq8_mmi(const uint8_t *pSrc, int iSrcStride,
+ uint8_t *pDst, int iDstStride, int iHeight) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ "ldl $8, 0x7(%[pSrc]) \n\t"
+ "ldr $8, 0x0(%[pSrc]) \n\t"
+ "sdl $8, 0x7(%[pDst]) \n\t"
+ "sdr $8, 0x0(%[pDst]) \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8"
+ );
+}
+
+void McCopyWidthEq16_mmi(const uint8_t *pSrc, int iSrcStride,
+ uint8_t *pDst, int iDstStride, int iHeight) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ "ldl $8, 0x7(%[pSrc]) \n\t"
+ "ldl $9, 0xF(%[pSrc]) \n\t"
+ "ldr $8, 0x0(%[pSrc]) \n\t"
+ "ldr $9, 0x8(%[pSrc]) \n\t"
+ "sdl $8, 0x7(%[pDst]) \n\t"
+ "sdl $9, 0xF(%[pDst]) \n\t"
+ "sdr $8, 0x0(%[pDst]) \n\t"
+ "sdr $9, 0x8(%[pDst]) \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$9"
+ );
+}
+
+static inline void McCopy_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ if (iWidth == 16)
+ McCopyWidthEq16_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else if (iWidth == 8)
+ McCopyWidthEq8_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else if (iWidth == 4)
+ McCopyWidthEq4_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else
+ McCopyWidthEq2_c (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+}
+
+void McChromaWidthEq4_mmi(const uint8_t *pSrc, int32_t iSrcStride, uint8_t *pDst,
+ int32_t iDstStride, const uint8_t *pABCD, int32_t iHeight) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "gsldlc1 $f6, 0x7(%[pABCD]) \n\t"
+ "gsldrc1 $f6, 0x0(%[pABCD]) \n\t"
+ "xor $f14, $f14, $f14 \n\t"
+ "punpcklbh $f6, $f6, $f6 \n\t"
+ "mov.d $f8, $f6 \n\t"
+ "punpcklhw $f6, $f6, $f6 \n\t"
+ "punpckhhw $f8, $f8, $f8 \n\t"
+ "mov.d $f10, $f6 \n\t"
+ "punpcklbh $f6, $f6, $f14 \n\t"
+ "punpckhbh $f10, $f10, $f14 \n\t"
+
+ "mov.d $f12, $f8 \n\t"
+ "punpcklbh $f8, $f8, $f14 \n\t"
+ "punpckhbh $f12, $f12, $f14 \n\t"
+ PTR_ADDU "%[pABCD], %[pSrc], %[iSrcStride] \n\t"
+ "dli $8, 0x6 \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f2, 0x8(%[pSrc]) \n\t"
+ "dmtc1 $8, $f16 \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f2, 0x1(%[pSrc]) \n\t"
+ "dli $8, 0x0020002000200020 \n\t"
+ "punpcklbh $f0, $f0, $f14 \n\t"
+ "punpcklbh $f2, $f2, $f14 \n\t"
+
+ "dmtc1 $8, $f18 \n\t"
+ "1: \n\t"
+ "pmullh $f0, $f0, $f6 \n\t"
+ "pmullh $f2, $f2, $f10 \n\t"
+ "paddh $f0, $f0, $f2 \n\t"
+
+ "gsldlc1 $f2, 0x7(%[pABCD]) \n\t"
+ "gsldrc1 $f2, 0x0(%[pABCD]) \n\t"
+ "punpcklbh $f2, $f2, $f14 \n\t"
+ "mov.d $f4, $f2 \n\t"
+ "pmullh $f2, $f2, $f8 \n\t"
+ "paddh $f0, $f0, $f2 \n\t"
+ "gsldlc1 $f2, 0x8(%[pABCD]) \n\t"
+ "gsldrc1 $f2, 0x1(%[pABCD]) \n\t"
+ "punpcklbh $f2, $f2, $f14 \n\t"
+ "mov.d $f14, $f2 \n\t"
+ "pmullh $f2, $f2, $f12 \n\t"
+ "paddh $f0, $f0, $f2 \n\t"
+ "mov.d $f2, $f14 \n\t"
+ "paddh $f0, $f0, $f18 \n\t"
+ "psrlh $f0, $f0, $f16 \n\t"
+ "xor $f14, $f14, $f14 \n\t"
+ "packushb $f0, $f0, $f14 \n\t"
+ "gsswlc1 $f0, 0x3(%[pDst]) \n\t"
+ "gsswrc1 $f0, 0x0(%[pDst]) \n\t"
+ "mov.d $f0, $f4 \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "%[pABCD], %[pABCD], %[iSrcStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"((unsigned char *)pSrc), [pDst]"+&r"((unsigned char *)pDst),
+ [pABCD]"+&r"((unsigned char *)pABCD), [iHeight]"+&r"((int)iHeight)
+ : [iSrcStride]"r"((int)iSrcStride), [iDstStride]"r"((int)iDstStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18"
+ );
+}
+
+void McChromaWidthEq8_mmi(const uint8_t *pSrc, int32_t iSrcStride, uint8_t *pDst,
+ int32_t iDstStride, const uint8_t *pABCD, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "gsldlc1 $f12, 0x7(%[pABCD]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f12, 0x0(%[pABCD]) \n\t"
+ "punpcklbh $f12, $f12, $f12 \n\t"
+ "punpckhhw $f14, $f12, $f12 \n\t"
+ "punpcklhw $f12, $f12, $f12 \n\t"
+
+ "mov.d $f16, $f14 \n\t"
+ "punpckhwd $f14, $f12, $f12 \n\t"
+ "punpcklwd $f12, $f12, $f12 \n\t"
+ "punpckhwd $f18, $f16, $f16 \n\t"
+ "punpcklwd $f16, $f16, $f16 \n\t"
+ "mov.d $f20, $f14 \n\t"
+ "mov.d $f24, $f18 \n\t"
+
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpckhbh $f26, $f24, $f28 \n\t"
+ "punpcklbh $f24, $f24, $f28 \n\t"
+
+ PTR_ADDU "%[pABCD], %[pSrc], %[iSrcStride] \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0x8(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x1(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "1: \n\t"
+ "dli $8, 0x20 \n\t"
+ "dmtc1 $8, $f30 \n\t"
+
+ "pmullh $f0, $f0, $f12 \n\t"
+ "pmullh $f2, $f2, $f14 \n\t"
+ "pmullh $f4, $f4, $f20 \n\t"
+ "pmullh $f6, $f6, $f22 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+
+ "gsldlc1 $f4, 0x7(%[pABCD]) \n\t"
+ "gsldrc1 $f4, 0x0(%[pABCD]) \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "mov.d $f8, $f4 \n\t"
+ "mov.d $f10, $f6 \n\t"
+ "pmullh $f4, $f4, $f16 \n\t"
+ "pmullh $f6, $f6, $f18 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+
+ "gsldlc1 $f4, 0x8(%[pABCD]) \n\t"
+ "gsldrc1 $f4, 0x1(%[pABCD]) \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "mov.d $f28, $f4 \n\t"
+ "mov.d $f30, $f6 \n\t"
+ "pmullh $f4, $f4, $f24 \n\t"
+ "pmullh $f6, $f6, $f26 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "mov.d $f4, $f28 \n\t"
+ "mov.d $f6, $f30 \n\t"
+
+ "dli $8, 0x0020002000200020 \n\t"
+ "dmfc1 $9, $f20 \n\t"
+ "dmtc1 $8, $f20 \n\t"
+ "dli $8, 0x6 \n\t"
+ "paddh $f0, $f0, $f20 \n\t"
+ "paddh $f2, $f2, $f20 \n\t"
+ "dmtc1 $8, $f20 \n\t"
+ "psrlh $f0, $f0, $f20 \n\t"
+ "psrlh $f2, $f2, $f20 \n\t"
+
+ "xor $f28, $f28, $f28 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+
+ "mov.d $f0, $f8 \n\t"
+ "mov.d $f2, $f10 \n\t"
+ "dmtc1 $9, $f20 \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "%[pABCD], %[pABCD], %[iSrcStride] \n\t"
+
+ PTR_ADDIU "%[iHeight], %[iHeight], -1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [pABCD]"+&r"(pABCD),
+ [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void McChroma_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int16_t iMvX, int16_t iMvY,
+ int32_t iWidth, int32_t iHeight) {+ static const PMcChromaWidthExtFunc kpMcChromaWidthFuncs[2] = {+ McChromaWidthEq4_mmi,
+ McChromaWidthEq8_mmi
+ };
+ const int32_t kiD8x = iMvX & 0x07;
+ const int32_t kiD8y = iMvY & 0x07;
+ if (kiD8x == 0 && kiD8y == 0) {+ McCopy_mmi (pSrc, iSrcStride, pDst, iDstStride, iWidth, iHeight);
+ return;
+ }
+ if (iWidth != 2) {+ kpMcChromaWidthFuncs[iWidth >> 3] (pSrc, iSrcStride, pDst, iDstStride,
+ g_kuiABCD[kiD8y][kiD8x], iHeight);
+ } else
+ McChromaWithFragMv_c (pSrc, iSrcStride, pDst, iDstStride, iMvX, iMvY,
+ iWidth, iHeight);
+}
+
+void McHorVer20WidthEq8_mmi(const uint8_t *pSrc, int iSrcStride, uint8_t *pDst,
+ int iDstStride, int iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], -0x2 \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "dli $8, 0x0010001000100010 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $8, $f26 \n\t"
+ "dli $8, 0x5 \n\t"
+ "dmtc1 $8, $f30 \n\t"
+ "1: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+ "paddh $f8, $f8, $f12 \n\t"
+ "paddh $f10, $f10, $f14 \n\t"
+ "paddh $f16, $f16, $f20 \n\t"
+ "paddh $f18, $f18, $f22 \n\t"
+ "psllh $f16, $f16, $f26 \n\t"
+ "psllh $f18, $f18, $f26 \n\t"
+ "psubh $f16, $f16, $f8 \n\t"
+ "psubh $f18, $f18, $f10 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "psllh $f16, $f16, $f26 \n\t"
+ "psllh $f18, $f18, $f26 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f24 \n\t"
+ "psrah $f0, $f0, $f30 \n\t"
+ "psrah $f2, $f2, $f30 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void McHorVer20WidthEq16_mmi(const uint8_t *pSrc, int iSrcStride, uint8_t *pDst,
+ int iDstStride, int iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], -0x2 \n\t"
+ "dli $8, 0x0010001000100010 \n\t"
+ "dmtc1 $8, $f24 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $8, $f26 \n\t"
+ "dli $8, 0x5 \n\t"
+ "dmtc1 $8, $f30 \n\t"
+ "1: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+ "paddh $f8, $f8, $f12 \n\t"
+ "paddh $f10, $f10, $f14 \n\t"
+ "paddh $f16, $f16, $f20 \n\t"
+ "paddh $f18, $f18, $f22 \n\t"
+ "psllh $f16, $f16, $f26 \n\t"
+ "psllh $f18, $f18, $f26 \n\t"
+ "psubh $f16, $f16, $f8 \n\t"
+ "psubh $f18, $f18, $f10 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "psllh $f16, $f16, $f26 \n\t"
+ "psllh $f18, $f18, $f26 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f24 \n\t"
+ "psrah $f0, $f0, $f30 \n\t"
+ "psrah $f2, $f2, $f30 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+ "gsldlc1 $f0, 0xF(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0x14(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x10(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0x13(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x11(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0x12(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x8(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0xd(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x9(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0xc(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0xb(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+ "paddh $f8, $f8, $f12 \n\t"
+ "paddh $f10, $f10, $f14 \n\t"
+ "paddh $f16, $f16, $f20 \n\t"
+ "paddh $f18, $f18, $f22 \n\t"
+ "psllh $f16, $f16, $f26 \n\t"
+ "psllh $f18, $f18, $f26 \n\t"
+ "psubh $f16, $f16, $f8 \n\t"
+ "psubh $f18, $f18, $f10 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "psllh $f16, $f16, $f26 \n\t"
+ "psllh $f18, $f18, $f26 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f24 \n\t"
+ "psrah $f0, $f0, $f30 \n\t"
+ "psrah $f2, $f2, $f30 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+ "gssdlc1 $f0, 0xF(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x8(%[pDst]) \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void McHorVer20WidthEq4_mmi(const uint8_t *pSrc, int iSrcStride, uint8_t *pDst,
+ int iDstStride, int iHeight) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ PTR_ADDIU "%[pSrc], %[pSrc], -0x2 \n\t"
+ "xor $f14, $f14, $f14 \n\t"
+ "dli $8, 0x0010001000100010 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "1: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f2, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f6, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f10, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f2, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f6, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f10, 0x3(%[pSrc]) \n\t"
+ "dli $8, 0x2 \n\t"
+ "punpcklbh $f0, $f0, $f14 \n\t"
+ "punpcklbh $f2, $f2, $f14 \n\t"
+ "punpcklbh $f4, $f4, $f14 \n\t"
+ "punpcklbh $f6, $f6, $f14 \n\t"
+ "punpcklbh $f8, $f8, $f14 \n\t"
+ "punpcklbh $f10, $f10, $f14 \n\t"
+ "dmtc1 $8, $f16 \n\t"
+ "paddh $f4, $f4, $f6 \n\t"
+ "paddh $f8, $f8, $f10 \n\t"
+ "psllh $f8, $f8, $f16 \n\t"
+ "psubh $f8, $f8, $f4 \n\t"
+ "paddh $f0, $f0, $f2 \n\t"
+ "paddh $f0, $f0, $f8 \n\t"
+ "dli $8, 0x5 \n\t"
+ "psllh $f8, $f8, $f16 \n\t"
+ "paddh $f0, $f0, $f8 \n\t"
+ "paddh $f0, $f0, $f12 \n\t"
+ "dmtc1 $8, $f16 \n\t"
+ "psrah $f0, $f0, $f16 \n\t"
+ "packushb $f0, $f0, $f14 \n\t"
+ "gsswlc1 $f0, 0x3(%[pDst]) \n\t"
+ "gsswrc1 $f0, 0x0(%[pDst]) \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16"
+ );
+}
+
+static inline void McHorVer20_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ if (iWidth == 16)
+ McHorVer20WidthEq16_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else if (iWidth == 8)
+ McHorVer20WidthEq8_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else
+ McHorVer20WidthEq4_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+}
+
+void McHorVer02WidthEq8_mmi(const uint8_t *pSrc, int iSrcStride, uint8_t *pDst,
+ int iDstStride, int iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ MMI_LOAD_8P($f0, $f2, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f4, $f6, $f28, $8)
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f8, $f10, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f12, $f14, $f28, $8)
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f16, $f18, $f28, %[pSrc])
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f20, $f22, $f28, $8)
+
+ "1: \n\t"
+ FILTER_HV_W8($f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20,
+ $f22, $f24, $f26, $f28, $f30, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f24, $f26, $f28, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f4, $f6, $f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24,
+ $f26, $f28, $f30, $f0, $f2, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f28, $f30, $f0, $8)
+ FILTER_HV_W8($f8, $f10, $f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28,
+ $f30, $f0, $f2, $f4, $f6, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f0, $f2, $f4, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f12, $f14, $f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0,
+ $f2, $f4, $f6, $f8, $f10, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f4, $f6, $f8, $8)
+ FILTER_HV_W8($f16, $f18, $f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4,
+ $f6, $f8, $f10, $f12, $f14, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f8, $f10, $f12, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f20, $f22, $f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8,
+ $f10, $f12, $f14, $f16, $f18, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f12, $f14, $f16, $8)
+ FILTER_HV_W8($f24, $f26, $f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12,
+ $f14, $f16, $f18, $f20, $f22, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f16, $f18, $f20, %[pSrc])
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ FILTER_HV_W8($f28, $f30, $f0, $f2, $f4, $f6, $f8, $f10, $f12, $f14, $f16,
+ $f18, $f20, $f22, $f24, $f26, %[pDst], $8, $9)
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "beqz %[iHeight], 2f \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "$8, %[pSrc], %[iSrcStride] \n\t"
+ MMI_LOAD_8P($f20, $f22, $f24, $8)
+ "j 1b \n\t"
+ "2: \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$9", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+static inline void McHorVer02WidthEq16_mmi(const uint8_t* pSrc, int32_t iSrcStride,
+ uint8_t* pDst, int32_t iDstStride, int32_t iHeight) {+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ McHorVer02WidthEq8_mmi (&pSrc[8], iSrcStride, &pDst[8], iDstStride, iHeight);
+}
+
+static inline void McHorVer02_mmi(const uint8_t* pSrc, int32_t iSrcStride,
+ uint8_t* pDst, int32_t iDstStride, int32_t iWidth,
+ int32_t iHeight) {+ if (iWidth == 16)
+ McHorVer02WidthEq16_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else if (iWidth == 8)
+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else
+ McHorVer02_c (pSrc, iSrcStride, pDst, iDstStride, 4, iHeight);
+}
+
+void McHorVer22Width8HorFirst_mmi(const uint8_t *pSrc, int16_t iSrcStride,
+ uint8_t *pDst, int32_t iDstStride, int32_t iHeight) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_SUBU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $8, $f30 \n\t"
+ "1: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrc]) \n\t"
+ "gsldlc1 $f4, 0xc(%[pSrc]) \n\t"
+ "gsldlc1 $f8, 0x8(%[pSrc]) \n\t"
+ "gsldlc1 $f12, 0xb(%[pSrc]) \n\t"
+ "gsldlc1 $f16, 0x9(%[pSrc]) \n\t"
+ "gsldlc1 $f20, 0xa(%[pSrc]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrc]) \n\t"
+ "gsldrc1 $f4, 0x5(%[pSrc]) \n\t"
+ "gsldrc1 $f8, 0x1(%[pSrc]) \n\t"
+ "gsldrc1 $f12, 0x4(%[pSrc]) \n\t"
+ "gsldrc1 $f16, 0x2(%[pSrc]) \n\t"
+ "gsldrc1 $f20, 0x3(%[pSrc]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpckhbh $f10, $f8, $f28 \n\t"
+ "punpckhbh $f14, $f12, $f28 \n\t"
+ "punpckhbh $f18, $f16, $f28 \n\t"
+ "punpckhbh $f22, $f20, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "punpcklbh $f8, $f8, $f28 \n\t"
+ "punpcklbh $f12, $f12, $f28 \n\t"
+ "punpcklbh $f16, $f16, $f28 \n\t"
+ "punpcklbh $f20, $f20, $f28 \n\t"
+ "paddh $f8, $f8, $f12 \n\t"
+ "paddh $f10, $f10, $f14 \n\t"
+ "paddh $f16, $f16, $f20 \n\t"
+ "paddh $f18, $f18, $f22 \n\t"
+ "psllh $f16, $f16, $f30 \n\t"
+ "psllh $f18, $f18, $f30 \n\t"
+ "psubh $f16, $f16, $f8 \n\t"
+ "psubh $f18, $f18, $f10 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "psllh $f16, $f16, $f30 \n\t"
+ "psllh $f18, $f18, $f30 \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdlc1 $f2, 0xF(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+ "gssdrc1 $f2, 0x8(%[pDst]) \n\t"
+ PTR_ADDU "%[pSrc], %[pSrc], %[iSrcStride] \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pSrc]"+&r"(pSrc), [pDst]"+&r"(pDst), [iHeight]"+&r"(iHeight)
+ : [iSrcStride]"r"(iSrcStride), [iDstStride]"r"(iDstStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+static inline void McHorVer22WidthEq8_mmi(const uint8_t* pSrc, int32_t iSrcStride,
+ uint8_t* pDst, int32_t iDstStride, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_2D (int16_t, iTap, 21, 8, 16)
+ McHorVer22Width8HorFirst_mmi (pSrc - 2, iSrcStride, (uint8_t*)iTap, 16, iHeight + 5);
+ McHorVer22Width8VerLastAlign_mmi ((uint8_t*)iTap, 16, pDst, iDstStride, 8, iHeight);
+}
+
+static inline void McHorVer22WidthEq16_mmi(const uint8_t* pSrc, int32_t iSrcStride,
+ uint8_t* pDst, int32_t iDstStride, int32_t iHeight) {+ McHorVer22WidthEq8_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ McHorVer22WidthEq8_mmi (&pSrc[8], iSrcStride, &pDst[8], iDstStride, iHeight);
+}
+
+static inline void McHorVer22_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ if (iWidth == 16)
+ McHorVer22WidthEq16_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else if (iWidth == 8)
+ McHorVer22WidthEq8_mmi (pSrc, iSrcStride, pDst, iDstStride, iHeight);
+ else
+ McHorVer22_c (pSrc, iSrcStride, pDst, iDstStride, 4, iHeight);
+}
+
+void PixelAvgWidthEq4_mmi(uint8_t *pDst, int iDstStride, const uint8_t *pSrcA,
+ int iSrcAStride, const uint8_t *pSrcB, int iSrcBStride, int iHeight ) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrcB]) \n\t"
+ "gsldlc1 $f2, 0x7(%[pSrcA]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrcB]) \n\t"
+ "gsldrc1 $f2, 0x0(%[pSrcA]) \n\t"
+ "pavgb $f0, $f0, $f2 \n\t"
+ "gsswlc1 $f0, 0x3(%[pDst]) \n\t"
+ "gsswrc1 $f0, 0x0(%[pDst]) \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x1 \n\t"
+ PTR_ADDU "%[pDst], %[pDst], %[iDstStride] \n\t"
+ PTR_ADDU "%[pSrcA], %[pSrcA], %[iSrcAStride] \n\t"
+ PTR_ADDU "%[pSrcB], %[pSrcB], %[iSrcBStride] \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pDst]"+&r"((unsigned char *)pDst), [pSrcA]"+&r"((unsigned char *)pSrcA),
+ [pSrcB]"+&r"((unsigned char *)pSrcB), [iHeight]"+&r"((int)iHeight)
+ : [iDstStride]"r"((int)iDstStride), [iSrcAStride]"r"((int)iSrcAStride),
+ [iSrcBStride]"r"((int)iSrcBStride)
+ : "memory", "$8", "$9", "$10", "$f0", "$f2"
+ );
+}
+
+void PixelAvgWidthEq8_mmi(uint8_t *pDst, int iDstStride, const uint8_t *pSrcA,
+ int iSrcAStride, const uint8_t *pSrcB, int iSrcBStride, int iHeight ) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrcA]) \n\t"
+ "gsldlc1 $f2, 0x7(%[pSrcB]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrcA]) \n\t"
+ "gsldrc1 $f2, 0x0(%[pSrcB]) \n\t"
+ "pavgb $f0, $f0, $f2 \n\t"
+ PTR_ADDU "$8, %[pSrcA], %[iSrcAStride] \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ PTR_ADDU "$9, %[pSrcB], %[iSrcBStride] \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+ "gsldlc1 $f0, 0x7($8) \n\t"
+ "gsldlc1 $f2, 0x7($9) \n\t"
+ "gsldrc1 $f0, 0x0($8) \n\t"
+ "gsldrc1 $f2, 0x0($9) \n\t"
+ "pavgb $f0, $f0, $f2 \n\t"
+ PTR_ADDU "$10, %[pDst], %[iDstStride] \n\t"
+ "gssdlc1 $f0, 0x7($10) \n\t"
+ PTR_ADDU "%[pSrcA], $8, %[iSrcAStride] \n\t"
+ "gssdrc1 $f0, 0x0($10) \n\t"
+ PTR_ADDU "%[pSrcB], $9, %[iSrcBStride] \n\t"
+ PTR_ADDU "%[pDst], $10, %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x2 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pDst]"+&r"((unsigned char *)pDst), [pSrcA]"+&r"((unsigned char *)pSrcA),
+ [pSrcB]"+&r"((unsigned char *)pSrcB), [iHeight]"+&r"((int)iHeight)
+ : [iDstStride]"r"((int)iDstStride), [iSrcAStride]"r"((int)iSrcAStride),
+ [iSrcBStride]"r"((int)iSrcBStride)
+ : "memory", "$8", "$9", "$10", "$f0", "$f2"
+ );
+}
+
+void PixelAvgWidthEq16_mmi(uint8_t *pDst, int iDstStride, const uint8_t *pSrcA,
+ int iSrcAStride, const uint8_t *pSrcB, int iSrcBStride, int iHeight ) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "1: \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrcA]) \n\t"
+ "gsldlc1 $f2, 0xF(%[pSrcA]) \n\t"
+ "gsldlc1 $f4, 0x7(%[pSrcB]) \n\t"
+ "gsldlc1 $f6, 0xF(%[pSrcB]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrcA]) \n\t"
+ "gsldrc1 $f2, 0x8(%[pSrcA]) \n\t"
+ "gsldrc1 $f4, 0x0(%[pSrcB]) \n\t"
+ "gsldrc1 $f6, 0x8(%[pSrcB]) \n\t"
+ "pavgb $f0, $f0, $f4 \n\t"
+ "pavgb $f2, $f2, $f6 \n\t"
+ PTR_ADDU "$8, %[pSrcA], %[iSrcAStride] \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdlc1 $f2, 0xF(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+ "gssdrc1 $f2, 0x8(%[pDst]) \n\t"
+ PTR_ADDU "$9, %[pSrcB], %[iSrcBStride] \n\t"
+ "gsldlc1 $f0, 0x7($8) \n\t"
+ "gsldlc1 $f2, 0xF($8) \n\t"
+ "gsldrc1 $f0, 0x0($8) \n\t"
+ "gsldrc1 $f2, 0x8($8) \n\t"
+ PTR_ADDU "$10, %[pDst], %[iDstStride] \n\t"
+ "gsldlc1 $f4, 0x7($9) \n\t"
+ "gsldlc1 $f6, 0xF($9) \n\t"
+ "gsldrc1 $f4, 0x0($9) \n\t"
+ "gsldrc1 $f6, 0x8($9) \n\t"
+ "pavgb $f0, $f0, $f4 \n\t"
+ "pavgb $f2, $f2, $f6 \n\t"
+ "gssdlc1 $f0, 0x7($10) \n\t"
+ "gssdlc1 $f2, 0xF($10) \n\t"
+ "gssdrc1 $f0, 0x0($10) \n\t"
+ "gssdrc1 $f2, 0x8($10) \n\t"
+
+ PTR_ADDU "%[pSrcA], $8, %[iSrcAStride] \n\t"
+ PTR_ADDU "%[pSrcB], $9, %[iSrcBStride] \n\t"
+ PTR_ADDU "%[pDst], $10, %[iDstStride] \n\t"
+ "gsldlc1 $f0, 0x7(%[pSrcA]) \n\t"
+ "gsldlc1 $f2, 0xF(%[pSrcA]) \n\t"
+ "gsldlc1 $f4, 0x7(%[pSrcB]) \n\t"
+ "gsldlc1 $f6, 0xF(%[pSrcB]) \n\t"
+ "gsldrc1 $f0, 0x0(%[pSrcA]) \n\t"
+ "gsldrc1 $f2, 0x8(%[pSrcA]) \n\t"
+ "gsldrc1 $f4, 0x0(%[pSrcB]) \n\t"
+ "gsldrc1 $f6, 0x8(%[pSrcB]) \n\t"
+ "pavgb $f0, $f0, $f4 \n\t"
+ "pavgb $f2, $f2, $f6 \n\t"
+ PTR_ADDU "$8, %[pSrcA], %[iSrcAStride] \n\t"
+ PTR_ADDU "$9, %[pSrcB], %[iSrcBStride] \n\t"
+ "gssdlc1 $f0, 0x7(%[pDst]) \n\t"
+ "gssdlc1 $f2, 0xF(%[pDst]) \n\t"
+ "gssdrc1 $f0, 0x0(%[pDst]) \n\t"
+ "gssdrc1 $f2, 0x8(%[pDst]) \n\t"
+ "gsldlc1 $f0, 0x7($8) \n\t"
+ "gsldlc1 $f2, 0xF($8) \n\t"
+ "gsldlc1 $f4, 0x7($9) \n\t"
+ "gsldlc1 $f6, 0xF($9) \n\t"
+ "gsldrc1 $f0, 0x0($8) \n\t"
+ "gsldrc1 $f2, 0x8($8) \n\t"
+ "gsldrc1 $f4, 0x0($9) \n\t"
+ "gsldrc1 $f6, 0x8($9) \n\t"
+ PTR_ADDU "$10, %[pDst], %[iDstStride] \n\t"
+ "pavgb $f0, $f0, $f4 \n\t"
+ "pavgb $f2, $f2, $f6 \n\t"
+ "gssdlc1 $f0, 0x7($10) \n\t"
+ "gssdlc1 $f2, 0xF($10) \n\t"
+ "gssdrc1 $f0, 0x0($10) \n\t"
+ "gssdrc1 $f2, 0x8($10) \n\t"
+ PTR_ADDU "%[pSrcA], $8, %[iSrcAStride] \n\t"
+ PTR_ADDU "%[pSrcB], $9, %[iSrcBStride] \n\t"
+ PTR_ADDU "%[pDst], $10, %[iDstStride] \n\t"
+ PTR_ADDIU "%[iHeight], %[iHeight], -0x4 \n\t"
+ "bnez %[iHeight], 1b \n\t"
+ : [pDst]"+&r"((unsigned char *)pDst), [pSrcA]"+&r"((unsigned char *)pSrcA),
+ [pSrcB]"+&r"((unsigned char *)pSrcB), [iHeight]"+&r"((int)iHeight)
+ : [iDstStride]"r"((int)iDstStride), [iSrcAStride]"r"((int)iSrcAStride),
+ [iSrcBStride]"r"((int)iSrcBStride)
+ : "memory", "$8", "$9", "$10", "$f0", "$f2", "$f4", "$f6"
+ );
+}
+
+static inline void McHorVer01_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer02WidthEq16_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pSrc, iSrcStride, pTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pSrc, iSrcStride, pTmp, 16, iHeight);
+ } else {+ McHorVer02_c (pSrc, iSrcStride, pTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pSrc, iSrcStride, pTmp, 16, iHeight);
+ }
+}
+
+static inline void McHorVer03_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer02WidthEq16_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pSrc + iSrcStride, iSrcStride, pTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pSrc + iSrcStride, iSrcStride, pTmp, 16, iHeight);
+ } else {+ McHorVer02_c (pSrc, iSrcStride, pTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pSrc + iSrcStride, iSrcStride, pTmp, 16, iHeight);
+ }
+}
+
+static inline void McHorVer10_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pSrc, iSrcStride, pTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pSrc, iSrcStride, pTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc, iSrcStride, pTmp, 16, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pSrc, iSrcStride, pTmp, 16, iHeight);
+ }
+}
+
+static inline void McHorVer11_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pVerTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq16_mmi (pSrc, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02_c (pSrc, iSrcStride, pVerTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ }
+}
+
+static inline void McHorVer12_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pVerTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pCtrTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer02WidthEq16_mmi (pSrc, iSrcStride, pVerTmp, 16, iHeight);
+ McHorVer22WidthEq16_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pVerTmp, 16, pCtrTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pVerTmp, 16, iHeight);
+ McHorVer22WidthEq8_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pVerTmp, 16, pCtrTmp, 16, iHeight);
+ } else {+ McHorVer02_c (pSrc, iSrcStride, pVerTmp, 16, 4, iHeight);
+ McHorVer22_c (pSrc, iSrcStride, pCtrTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pVerTmp, 16, pCtrTmp, 16, iHeight);
+ }
+}
+static inline void McHorVer13_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pVerTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq16_mmi (pSrc, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq8_mmi (pSrc, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02_c (pSrc, iSrcStride, pVerTmp, 16, 4 , iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ }
+}
+static inline void McHorVer21_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pCtrTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer22WidthEq16_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pHorTmp, 16, pCtrTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer22WidthEq8_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pHorTmp, 16, pCtrTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer22_c (pSrc, iSrcStride, pCtrTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pHorTmp, 16, pCtrTmp, 16, iHeight);
+ }
+}
+
+static inline void McHorVer23_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pCtrTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer22WidthEq16_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pHorTmp, 16, pCtrTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer22WidthEq8_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pHorTmp, 16, pCtrTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer22_c (pSrc, iSrcStride, pCtrTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pHorTmp, 16, pCtrTmp, 16, iHeight);
+ }
+}
+static inline void McHorVer30_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pSrc + 1, iSrcStride, pHorTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pSrc + 1, iSrcStride, pHorTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pSrc + 1, iSrcStride, pHorTmp, 16, iHeight);
+ }
+}
+static inline void McHorVer31_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pVerTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq16_mmi (pSrc + 1, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq8_mmi (pSrc + 1, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02_c (pSrc + 1, iSrcStride, pVerTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ }
+}
+static inline void McHorVer32_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pVerTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pCtrTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer02WidthEq16_mmi (pSrc + 1, iSrcStride, pVerTmp, 16, iHeight);
+ McHorVer22WidthEq16_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pVerTmp, 16, pCtrTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer02WidthEq8_mmi (pSrc + 1, iSrcStride, pVerTmp, 16, iHeight);
+ McHorVer22WidthEq8_mmi (pSrc, iSrcStride, pCtrTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pVerTmp, 16, pCtrTmp, 16, iHeight);
+ } else {+ McHorVer02_c (pSrc + 1, iSrcStride, pVerTmp, 16, 4, iHeight);
+ McHorVer22_c (pSrc, iSrcStride, pCtrTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pVerTmp, 16, pCtrTmp, 16, iHeight);
+ }
+}
+static inline void McHorVer33_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst,
+ int32_t iDstStride, int32_t iWidth, int32_t iHeight) {+ ENFORCE_STACK_ALIGN_1D (uint8_t, pHorTmp, 256, 16);
+ ENFORCE_STACK_ALIGN_1D (uint8_t, pVerTmp, 256, 16);
+ if (iWidth == 16) {+ McHorVer20WidthEq16_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq16_mmi (pSrc + 1, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq16_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else if (iWidth == 8) {+ McHorVer20WidthEq8_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02WidthEq8_mmi (pSrc + 1, iSrcStride, pVerTmp, 16, iHeight);
+ PixelAvgWidthEq8_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ } else {+ McHorVer20WidthEq4_mmi (pSrc + iSrcStride, iSrcStride, pHorTmp, 16, iHeight);
+ McHorVer02_c (pSrc + 1, iSrcStride, pVerTmp, 16, 4, iHeight);
+ PixelAvgWidthEq4_mmi (pDst, iDstStride, pHorTmp, 16, pVerTmp, 16, iHeight);
+ }
+}
+
+void McLuma_mmi(const uint8_t* pSrc, int32_t iSrcStride, uint8_t* pDst, int32_t iDstStride,
+ int16_t iMvX, int16_t iMvY, int32_t iWidth, int32_t iHeight) {+ static const PWelsMcWidthHeightFunc pWelsMcFunc[4][4] = { //[x][y]+ {McCopy_mmi, McHorVer01_mmi, McHorVer02_mmi, McHorVer03_mmi},+ {McHorVer10_mmi, McHorVer11_mmi, McHorVer12_mmi, McHorVer13_mmi},+ {McHorVer20_mmi, McHorVer21_mmi, McHorVer22_mmi, McHorVer23_mmi},+ {McHorVer30_mmi, McHorVer31_mmi, McHorVer32_mmi, McHorVer33_mmi},+ };
+
+ pWelsMcFunc[iMvX & 0x03][iMvY & 0x03] (pSrc, iSrcStride, pDst, iDstStride, iWidth, iHeight);
+}
+
+void PixelAvg_mmi(uint8_t* pDst, int32_t iDstStride, const uint8_t* pSrcA, int32_t iSrcAStride,
+ const uint8_t* pSrcB, int32_t iSrcBStride, int32_t iWidth, int32_t iHeight) {+ static const PWelsSampleWidthAveragingFunc kpfFuncs[2] = {+ PixelAvgWidthEq8_mmi,
+ PixelAvgWidthEq16_mmi
+ };
+ kpfFuncs[iWidth >> 4] (pDst, iDstStride, pSrcA, iSrcAStride, pSrcB, iSrcBStride, iHeight);
+}
+#endif//HAVE_MMI
} // anon ns.
void WelsCommon::InitMcFunc (SMcFunc* pMcFuncs, uint32_t uiCpuFlag) {@@ -1716,4 +4252,15 @@
pMcFuncs->pfLumaHalfpelCen = McHorVer22Width5Or9Or17Height5Or9Or17_AArch64_neon;//iWidth+1/heigh+1
}
#endif
+
+#if defined(HAVE_MMI)
+ if (uiCpuFlag & WELS_CPU_MMI) {+ pMcFuncs->pfLumaHalfpelHor = McHorVer20Width5Or9Or17_mmi;
+ pMcFuncs->pfLumaHalfpelVer = McHorVer02Height5Or9Or17_mmi;
+ pMcFuncs->pfLumaHalfpelCen = McHorVer22Width5Or9Or17Height5Or9Or17_mmi;
+ pMcFuncs->pfSampleAveraging = PixelAvg_mmi;
+ pMcFuncs->pMcChromaFunc = McChroma_mmi;
+ pMcFuncs->pMcLumaFunc = McLuma_mmi;
+ }
+#endif//HAVE_MMI
}
--- a/codec/decoder/core/inc/decode_mb_aux.h
+++ b/codec/decoder/core/inc/decode_mb_aux.h
@@ -63,6 +63,10 @@
#endif
+#if defined(HAVE_MMI)
+void IdctResAddPred_mmi (uint8_t* pPred, const int32_t kiStride, int16_t* pRs);
+#endif//HAVE_MMI
+
#if defined(__cplusplus)
}
#endif//__cplusplus
--- a/codec/decoder/core/inc/get_intra_predictor.h
+++ b/codec/decoder/core/inc/get_intra_predictor.h
@@ -166,6 +166,20 @@
void WelsDecoderIChromaPredPlane_AArch64_neon (uint8_t* pPred, const int32_t kiStride);
void WelsDecoderIChromaPredDcTop_AArch64_neon (uint8_t* pPred, const int32_t kiStride);
#endif//HAVE_NEON_AARCH64
+
+#if defined(HAVE_MMI)
+void WelsDecoderI16x16LumaPredDc_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderI16x16LumaPredPlane_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderI16x16LumaPredH_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderI16x16LumaPredV_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderI16x16LumaPredDcTop_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderI16x16LumaPredDcNA_mmi (uint8_t* pPred, const int32_t kiStride);
+
+void WelsDecoderIChromaPredDcTop_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderIChromaPredPlane_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderIChromaPredDc_mmi (uint8_t* pPred, const int32_t kiStride);
+void WelsDecoderI4x4LumaPredH_mmi (uint8_t* pPred, const int32_t kiStride);
+#endif//HAVE_MMI
#if defined(__cplusplus)
}
#endif//__cplusplus
--- /dev/null
+++ b/codec/decoder/core/mips/dct_mmi.c
@@ -1,0 +1,786 @@
+/*!
+ * \copy
+ * Copyright (c) 2009-2018, Cisco Systems
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ *
+ *
+ * \file dct_mmi.c
+ *
+ * \brief Loongson optimization
+ *
+ * \date 17/07/2018 Created
+ *
+ *************************************************************************************
+ */
+#include <stdint.h>
+#include "asmdefs_mmi.h"
+
+#define LOAD_2_LEFT_AND_ADD \
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t" \
+ "lbu $9, -0x1(%[pPred]) \n\t" \
+ PTR_ADDU "$8, $8, $9 \n\t" \
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t" \
+ "lbu $9, -0x1(%[pPred]) \n\t" \
+ PTR_ADDU "$8, $8, $9 \n\t"
+
+unsigned char mmi_dc_0x80[16] __attribute__((aligned(16))) = {+ 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
+ 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80};
+
+short mmi_wd_0x02[8] __attribute__((aligned(16))) = {2, 2, 2, 2, 2, 2, 2, 2};+short mmi_plane_inc_minus[8]__attribute__((aligned(16))) = {-7, -6, -5, -4, -3, -2, -1, 0};+short mmi_plane_inc[8]__attribute__((aligned(16))) = {1, 2, 3, 4, 5, 6, 7, 8};+short mmi_plane_dec[8]__attribute__((aligned(16))) = {8, 7, 6, 5, 4, 3, 2, 1};+
+short mmi_plane_inc_c[4]__attribute__((aligned(16))) = {1, 2, 3, 4};+short mmi_plane_dec_c[4]__attribute__((aligned(16))) = {4, 3, 2, 1};+short mmi_plane_mul_b_c[8]__attribute__((aligned(16))) = {-3, -2, -1, 0, 1, 2, 3, 4};+
+unsigned char mmi_01bytes[16]__attribute__((aligned(16))) = {1, 1, 1, 1, 1, 1, 1, 1,+ 1, 1, 1, 1, 1, 1, 1, 1};
+
+void IdctResAddPred_mmi(uint8_t *pPred, const int32_t kiStride, int16_t *pRs) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "dli $8, 0x1 \n\t"
+ "gsldxc1 $f0, 0x0(%[pRs], $0) \n\t"
+ "gsldxc1 $f2, 0x8(%[pRs], $0) \n\t"
+ "gsldxc1 $f4, 0x10(%[pRs], $0) \n\t"
+ "gsldxc1 $f6, 0x18(%[pRs], $0) \n\t"
+ "dmtc1 $8, $f14 \n\t"
+
+ MMI_Trans4x4H_SINGLE($f0, $f2, $f4, $f6, $f8)
+ MMI_IDCT_SINGLE($f2, $f4, $f6, $f8, $f0, $f12, $f14)
+ MMI_Trans4x4H_SINGLE($f2, $f6, $f0, $f8, $f4)
+ MMI_IDCT_SINGLE($f6, $f0, $f8, $f4, $f2, $f12, $f14)
+
+ "dli $8, 0x20 \n\t"
+ "xor $f14, $f14, $f14 \n\t"
+ "dmtc1 $8, $f12 \n\t"
+ "pshufh $f12, $f12, $f14 \n\t"
+ "dli $8, 0x6 \n\t"
+ "dmtc1 $8, $f16 \n\t"
+
+ MMI_StoreDiff4P_SINGLE($f6, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ MMI_StoreDiff4P_SINGLE($f8, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ MMI_StoreDiff4P_SINGLE($f2, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ MMI_StoreDiff4P_SINGLE($f4, $f0, $f12, $f14, %[pPred], %[pPred], $f16)
+ : [pPred]"+&r"((unsigned char *)pPred)
+ : [pRs]"r"((unsigned char *)pRs), [kiStride]"r"((int)kiStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10", "$f12",
+ "$f14", "$f16"
+ );
+}
+
+void WelsDecoderI16x16LumaPredDc_mmi(uint8_t *pPred, const int32_t kiStride) {+ __asm__ volatile(
+ ".set arch=loongson3a \n\t"
+ "dli $8, 0x5 \n\t"
+ "gsldxc1 $f10, 0x0(%[mmi_01bytes], $0) \n\t"
+ "dmtc1 $8, $f8 \n\t"
+
+ "move $10, %[pPred] \n\t"
+ PTR_SUBU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gslqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ "xor $f4, $f4, $f4 \n\t"
+ "pasubub $f0, $f0, $f4 \n\t"
+ "pasubub $f2, $f2, $f4 \n\t"
+ "biadd $f0, $f0 \n\t"
+ "biadd $f2, $f2 \n\t"
+ "paddh $f0, $f0, $f2 \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $8, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+
+ LOAD_2_LEFT_AND_ADD
+ LOAD_2_LEFT_AND_ADD
+ LOAD_2_LEFT_AND_ADD
+ LOAD_2_LEFT_AND_ADD
+ LOAD_2_LEFT_AND_ADD
+ LOAD_2_LEFT_AND_ADD
+ LOAD_2_LEFT_AND_ADD
+
+ PTR_ADDIU "$8, $8, 0x10 \n\t"
+ "dmtc1 $8, $f4 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "psrlw $f0, $f0, $f8 \n\t"
+ "pmuluw $f0, $f0, $f10 \n\t"
+ "punpcklwd $f0, $f0, $f0 \n\t"
+ "mov.d $f2, $f0 \n\t"
+
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0($10) \n\t"
+ : [pPred] "+&r"((unsigned char *)pPred)
+ : [kiStride] "r"((int)kiStride),
+ [mmi_01bytes] "r"((unsigned char *)mmi_01bytes)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10"
+ );
+}
+
+void WelsDecoderI16x16LumaPredPlane_mmi(uint8_t *pPred, const int32_t kiStride) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $10, %[pPred] \n\t"
+ PTR_ADDIU "%[pPred], %[pPred], -0x1 \n\t"
+ PTR_SUBU "%[pPred], %[pPred], %[kiStride] \n\t"
+
+ "gsldlc1 $f0, 0x7(%[pPred]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x0(%[pPred]) \n\t"
+ "gslqc1 $f22, $f20, 0x0(%[mmi_plane_dec]) \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "pmullh $f0, $f0, $f20 \n\t"
+ "gsldlc1 $f4, 0x10(%[pPred]) \n\t"
+ "pmullh $f2, $f2, $f22 \n\t"
+ "gsldrc1 $f4, 0x9(%[pPred]) \n\t"
+ "gslqc1 $f26, $f24, 0x0(%[mmi_plane_inc]) \n\t"
+ "punpckhbh $f6, $f4, $f28 \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "pmullh $f4, $f4, $f24 \n\t"
+ "pmullh $f6, $f6, $f26 \n\t"
+ "psubh $f4, $f4, $f0 \n\t"
+ "psubh $f6, $f6, $f2 \n\t"
+
+ SUMH_HORIZON($f4, $f6, $f0, $f2, $f8)
+ "dmfc1 $8, $f4 \n\t"
+ "seh $8, $8 \n\t"
+ "mul $8, $8, 0x5 \n\t"
+ PTR_ADDIU "$8, $8, 0x20 \n\t"
+ "sra $8, $8, 0x6 \n\t"
+ MMI_Copy8Times($f4, $f6, $f28, $8)
+
+ "lbu $9, 0x10(%[pPred]) \n\t"
+ PTR_ADDIU "%[pPred], %[pPred], -0x3 \n\t"
+ LOAD_COLUMN($f0, $f2, $f8, $f10, $f12, $f14, $f16, $f18, %[pPred],
+ %[kiStride], $11)
+
+ PTR_ADDIU "%[pPred], %[pPred], 0x3 \n\t"
+ "dsll $11, %[kiStride], 0x3 \n\t"
+ PTR_ADDU "$11, $11, %[pPred] \n\t"
+ "lbu $8, 0x0($11) \n\t"
+ PTR_ADDU "$9, $9, $8 \n\t"
+ "dsll $9, $9, 0x4 \n\t"
+
+ PTR_ADDIU "%[pPred], %[pPred], -0x3 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ LOAD_COLUMN($f28, $f30, $f8, $f10, $f12, $f14, $f16, $f18, %[pPred],
+ %[kiStride], $11)
+
+ "xor $f16, $f16, $f16 \n\t"
+ "punpcklbh $f0, $f2, $f16 \n\t"
+ "punpckhbh $f2, $f2, $f16 \n\t"
+ "pmullh $f0, $f0, $f20 \n\t"
+ "pmullh $f2, $f2, $f22 \n\t"
+ "punpcklbh $f28, $f30, $f16 \n\t"
+ "punpckhbh $f30, $f30, $f16 \n\t"
+ "pmullh $f28, $f28, $f24 \n\t"
+ "pmullh $f30, $f30, $f26 \n\t"
+ "psubh $f28, $f28, $f0 \n\t"
+ "psubh $f30, $f30, $f2 \n\t"
+
+ "xor $f8, $f8, $f8 \n\t"
+
+ SUMH_HORIZON($f28, $f30, $f0, $f2, $f8)
+ "dmfc1 $8, $f28 \n\t"
+ "seh $8, $8 \n\t"
+
+ "mul $8, $8, 0x5 \n\t"
+ PTR_ADDIU "$8, $8, 0x20 \n\t"
+ "sra $8, $8, 0x6 \n\t"
+ MMI_Copy8Times($f16, $f18, $f8, $8)
+
+ "move %[pPred], $10 \n\t"
+ PTR_ADDIU "$9, $9, 0x10 \n\t"
+ "mul $8, $8, -0x7 \n\t"
+ PTR_ADDU "$9, $9, $8 \n\t"
+ MMI_Copy8Times($f0, $f2, $f8, $9)
+
+ "xor $8, $8, $8 \n\t"
+ "gslqc1 $f22, $f20, 0x0(%[mmi_plane_inc_minus]) \n\t"
+
+ "dli $11, 0x5 \n\t"
+ "dmtc1 $11, $f30 \n\t"
+ "1: \n\t"
+ "pmullh $f8, $f4, $f20 \n\t"
+ "pmullh $f10, $f6, $f22 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "psrah $f8, $f8, $f30 \n\t"
+ "psrah $f10, $f10, $f30 \n\t"
+ "pmullh $f12, $f4, $f24 \n\t"
+ "pmullh $f14, $f6, $f26 \n\t"
+ "paddh $f12, $f12, $f0 \n\t"
+ "paddh $f14, $f14, $f2 \n\t"
+ "psrah $f12, $f12, $f30 \n\t"
+ "psrah $f14, $f14, $f30 \n\t"
+ "packushb $f8, $f8, $f10 \n\t"
+ "packushb $f10, $f12, $f14 \n\t"
+ "gssqc1 $f10, $f8, 0x0(%[pPred]) \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ PTR_ADDIU "$8, $8, 0x1 \n\t"
+ PTR_ADDIU "$11, $8, -0x10 \n\t"
+ "bnez $11, 1b \n\t"
+ "nop \n\t"
+ : [pPred]"+&r"((unsigned char *)pPred)
+ : [kiStride]"r"((int)kiStride), [mmi_plane_inc_minus]"r"(mmi_plane_inc_minus),
+ [mmi_plane_inc]"r"(mmi_plane_inc), [mmi_plane_dec]"r"(mmi_plane_dec)
+ : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8",
+ "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26",
+ "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+#define COPY_16_TIMES(r0, f0, f2, f4, f6, f8) \
+ "gslqc1 "#f2", "#f0", -0x10("#r0") \n\t" \+ "dsrl "#f0", "#f2", "#f4" \n\t" \
+ "pmuluw "#f0", "#f0", "#f6" \n\t" \
+ "punpcklwd "#f0", "#f0", "#f0" \n\t" \
+ "mov.d "#f2", "#f0" \n\t"
+
+#define MMI_PRED_H_16X16_TWO_LINE_DEC \
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t" \
+ COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8) \
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t" \
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t" \
+ COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8) \
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+
+void WelsDecoderI16x16LumaPredH_mmi(uint8_t *pPred, const int32_t kiStride) {+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "dli $8, 56 \n\t"
+ "dmtc1 $8, $f4 \n\t"
+ "gsldxc1 $f6, 0x0(%[mmi_01bytes], $0) \n\t"
+ "xor $f8, $f8, $f8 \n\t"
+
+ COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8)
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ COPY_16_TIMES(%[pPred], $f0, $f2, $f4, $f6, $f8)
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ MMI_PRED_H_16X16_TWO_LINE_DEC
+ : [pPred]"+&r"((unsigned char *)pPred)
+ : [kiStride]"r"((int)kiStride),
+ [mmi_01bytes]"r"((unsigned char *)mmi_01bytes)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6", "$f8"
+ );
+}
+
+void WelsDecoderI16x16LumaPredV_mmi(uint8_t *pPred, const int32_t kiStride) {+ __asm__ volatile(
+ ".set arch=loongson3a \n\t"
+ PTR_SUBU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gslqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ : [pPred] "+&r"((unsigned char *)pPred)
+ : [kiStride] "r"((int)kiStride)
+ : "memory", "$f0", "$f2"
+ );
+}
+
+void WelsDecoderI16x16LumaPredDcTop_mmi(uint8_t *pPred, const int32_t kiStride) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ PTR_SUBU "$8, %[pPred], %[kiStride] \n\t"
+ "gslqc1 $f2, $f0, 0x0($8) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "pasubub $f0, $f0, $f28 \n\t"
+ "pasubub $f2, $f2, $f28 \n\t"
+ "biadd $f0, $f0 \n\t"
+ "biadd $f2, $f2 \n\t"
+ "paddh $f0, $f0, $f2 \n\t"
+ "dmfc1 $8, $f0 \n\t"
+
+ PTR_ADDIU "$8, $8, 0x8 \n\t"
+ "dsra $8, $8, 0x4 \n\t"
+ MMI_Copy16Times($f4, $f6, $f28, $8)
+ "mov.d $f0, $f4 \n\t"
+ "mov.d $f2, $f6 \n\t"
+
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ : [pPred]"+&r"((unsigned char *)pPred)
+ : [kiStride]"r"((int)kiStride)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6"
+ );
+ RECOVER_REG;
+}
+
+void WelsDecoderI16x16LumaPredDcNA_mmi(uint8_t *pPred, const int32_t kiStride) {+ __asm__ volatile(
+ ".set arch=loongson3a \n\t"
+ "gslqc1 $f2, $f0, 0x0(%[mmi_dc_0x80]) \n\t"
+ "mov.d $f4, $f0 \n\t"
+ "mov.d $f6, $f2 \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f2, $f0, 0x0(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[pPred]) \n\t"
+ : [pPred] "+&r"((unsigned char *)pPred)
+ : [kiStride] "r"((int)kiStride), [mmi_dc_0x80] "r"(mmi_dc_0x80)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6"
+ );
+}
+
+void WelsDecoderIChromaPredPlane_mmi(uint8_t *pPred, const int32_t kiStride) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $10, %[pPred] \n\t"
+ PTR_ADDIU "%[pPred], %[pPred], -0x1 \n\t"
+ PTR_SUBU "%[pPred], %[pPred], %[kiStride] \n\t"
+
+ "gsldlc1 $f0, 0x7(%[pPred]) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "gsldrc1 $f0, 0x0(%[pPred]) \n\t"
+ "gsldxc1 $f20, 0x0(%[mmi_plane_dec_c], $0) \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "gsldlc1 $f4, 0xc(%[pPred]) \n\t"
+ "pmullh $f0, $f0, $f20 \n\t"
+ "gsldrc1 $f4, 0x5(%[pPred]) \n\t"
+ "gsldxc1 $f24, 0x0(%[mmi_plane_inc_c], $0) \n\t"
+ "punpcklbh $f4, $f4, $f28 \n\t"
+ "pmullh $f4, $f4, $f24 \n\t"
+ "psubh $f4, $f4, $f0 \n\t"
+
+ "xor $f6, $f6, $f6 \n\t"
+ "xor $f8, $f8, $f8 \n\t"
+ SUMH_HORIZON($f4, $f6, $f0, $f2, $f8)
+ "dmfc1 $8, $f4 \n\t"
+ "seh $8, $8 \n\t"
+ "mul $8, $8, 0x11 \n\t"
+ PTR_ADDIU "$8, $8, 0x10 \n\t"
+ "sra $8, $8, 0x5 \n\t"
+ MMI_Copy8Times($f4, $f6, $f8, $8)
+
+ "lbu $9, 0x8(%[pPred]) \n\t"
+ PTR_ADDIU "%[pPred], %[pPred], -0x3 \n\t"
+ LOAD_COLUMN_C($f0, $f8, $f12, $f16, %[pPred], %[kiStride], $11)
+
+ PTR_ADDIU "%[pPred], %[pPred], 0x3 \n\t"
+ "dsll $11, %[kiStride], 0x2 \n\t"
+ PTR_ADDU "$11, $11, %[pPred] \n\t"
+ "lbu $8, 0x0($11) \n\t"
+ PTR_ADDU "$9, $9, $8 \n\t"
+ "dsll $9, $9, 0x4 \n\t"
+
+ PTR_ADDIU "%[pPred], %[pPred], -0x3 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ LOAD_COLUMN_C($f28, $f8, $f12, $f16, %[pPred], %[kiStride], $11)
+ "xor $f16, $f16, $f16 \n\t"
+ "punpckhbh $f0, $f0, $f16 \n\t"
+ "pmullh $f0, $f0, $f20 \n\t"
+ "punpckhbh $f28, $f28, $f16 \n\t"
+ "pmullh $f28, $f28, $f24 \n\t"
+ "psubh $f28, $f28, $f0 \n\t"
+
+ "xor $f30, $f30, $f30 \n\t"
+ "xor $f8, $f8, $f8 \n\t"
+ SUMH_HORIZON($f28, $f30, $f0, $f2, $f8)
+ "dmfc1 $8, $f28 \n\t"
+ "seh $8, $8 \n\t"
+
+ "mul $8, $8, 0x11 \n\t"
+ PTR_ADDIU "$8, $8, 0x10 \n\t"
+ "sra $8, $8, 0x5 \n\t"
+ MMI_Copy8Times($f16, $f18, $f8, $8)
+
+ "move %[pPred], $10 \n\t"
+ PTR_ADDIU "$9, $9, 0x10 \n\t"
+ "mul $8, $8, -0x3 \n\t"
+ PTR_ADDU "$9, $9, $8 \n\t"
+ MMI_Copy8Times($f0, $f2, $f8, $9)
+
+ "xor $8, $8, $8 \n\t"
+ "gslqc1 $f22, $f20, 0x0(%[mmi_plane_mul_b_c]) \n\t"
+
+ "dli $11, 0x5 \n\t"
+ "dmtc1 $11, $f30 \n\t"
+ "1: \n\t"
+ "pmullh $f8, $f4, $f20 \n\t"
+ "pmullh $f10, $f6, $f22 \n\t"
+ "paddh $f8, $f8, $f0 \n\t"
+ "paddh $f10, $f10, $f2 \n\t"
+ "psrah $f8, $f8, $f30 \n\t"
+ "psrah $f10, $f10, $f30 \n\t"
+ "packushb $f8, $f8, $f10 \n\t"
+ "gssdxc1 $f8, 0x0(%[pPred], $0) \n\t"
+ "paddh $f0, $f0, $f16 \n\t"
+ "paddh $f2, $f2, $f18 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ PTR_ADDIU "$8, $8, 0x1 \n\t"
+ PTR_ADDIU "$11, $8, -0x8 \n\t"
+ "bnez $11, 1b \n\t"
+ "nop \n\t"
+ : [pPred]"+&r"((unsigned char *)pPred)
+ : [kiStride]"r"((int)kiStride), [mmi_plane_mul_b_c]"r"(mmi_plane_mul_b_c),
+ [mmi_plane_inc_c]"r"(mmi_plane_inc_c), [mmi_plane_dec_c]"r"(mmi_plane_dec_c)
+ : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10",
+ "$f12", "$f14", "$f16", "$f18", "$f20", "$f22", "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void WelsDecoderIChromaPredDc_mmi(uint8_t *pPred, const int32_t kiStride) {+ __asm__ volatile(
+ ".set arch=loongson3a \n\t"
+ "move $10, %[pPred] \n\t"
+
+ PTR_SUBU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gsldxc1 $f0, 0x0(%[pPred], $0) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $8, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+ "dmtc1 $8, $f2 \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $8, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "lbu $9, -0x1(%[pPred]) \n\t"
+ PTR_ADDU "$8, $8, $9 \n\t"
+ "dmtc1 $8, $f4 \n\t"
+
+ "xor $f8, $f8, $f8 \n\t"
+ "punpcklwd $f6, $f0, $f8 \n\t"
+ "punpckhwd $f0, $f0, $f8 \n\t"
+ "pasubub $f0, $f0, $f8 \n\t"
+ "pasubub $f6, $f6, $f8 \n\t"
+ "biadd $f0, $f0 \n\t"
+ "biadd $f6, $f6 \n\t"
+
+ "paddd $f6, $f6, $f2 \n\t"
+ "paddd $f2, $f4, $f0 \n\t"
+
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $8, $f8 \n\t"
+ "gsldxc1 $f12, 0x0(%[mmi_01bytes], $0) \n\t"
+ "dli $8, 0x3 \n\t"
+ "dmtc1 $8, $f10 \n\t"
+
+ "paddd $f0, $f0, $f8 \n\t"
+ "dsrl $f0, $f0, $f8 \n\t"
+
+ "paddd $f4, $f4, $f8 \n\t"
+ "dsrl $f4, $f4, $f8 \n\t"
+
+ "paddd $f6, $f6, $f8 \n\t"
+ "paddd $f6, $f6, $f8 \n\t"
+ "dsrl $f6, $f6, $f10 \n\t"
+
+ "paddd $f2, $f2, $f8 \n\t"
+ "paddd $f2, $f2, $f8 \n\t"
+ "dsrl $f2, $f2, $f10 \n\t"
+
+ "dli $8, 0x20 \n\t"
+ "dmtc1 $8, $f8 \n\t"
+ "pmuluw $f0, $f0, $f12 \n\t"
+ "pmuluw $f6, $f6, $f12 \n\t"
+ "dsll $f0, $f0, $f8 \n\t"
+ "xor $f0, $f0, $f6 \n\t"
+
+ "pmuluw $f4, $f4, $f12 \n\t"
+ "pmuluw $f2, $f2, $f12 \n\t"
+ "dsll $f2, $f2, $f8 \n\t"
+ "xor $f2, $f2, $f4 \n\t"
+
+ "gssdxc1 $f0, 0x0($10, $0) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0($10, $0) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0($10, $0) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0($10, $0) \n\t"
+
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f2, 0x0($10, $0) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f2, 0x0($10, $0) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f2, 0x0($10, $0) \n\t"
+ PTR_ADDU "$10, $10, %[kiStride] \n\t"
+ "gssdxc1 $f2, 0x0($10, $0) \n\t"
+ : [pPred] "+&r"((unsigned char *)pPred)
+ : [kiStride] "r"((int)kiStride),
+ [mmi_01bytes] "r"((unsigned char *)mmi_01bytes)
+ : "memory", "$8", "$9", "$10", "$f0", "$f2", "$f4", "$f6", "$f8", "$f10",
+ "$f12"
+ );
+}
+
+void WelsDecoderIChromaPredDcTop_mmi(uint8_t *pPred, const int32_t kiStride) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "dli $8, 0x4e \n\t"
+ "dmtc1 $8, $f16 \n\t"
+ "dli $8, 0xb1 \n\t"
+ "dmtc1 $8, $f18 \n\t"
+ "dli $8, 0x2 \n\t"
+ "dmtc1 $8, $f20 \n\t"
+ PTR_SUBU "$8, %[pPred], %[kiStride] \n\t"
+ "gsldxc1 $f0, 0x0($8, $0) \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "punpckhbh $f2, $f0, $f28 \n\t"
+ "punpcklbh $f0, $f0, $f28 \n\t"
+ "pshufh $f4, $f0, $f16 \n\t"
+ "pshufh $f6, $f2, $f16 \n\t"
+ "paddh $f0, $f0, $f4 \n\t"
+ "paddh $f2, $f2, $f6 \n\t"
+
+ "pshufh $f8, $f0, $f18 \n\t"
+ "pshufh $f14, $f2, $f18 \n\t"
+ "paddh $f2, $f2, $f14 \n\t"
+ "paddh $f0, $f0, $f8 \n\t"
+
+ "gslqc1 $f26, $f24, 0x0(%[mmi_wd_0x02]) \n\t"
+ "paddh $f0, $f0, $f24 \n\t"
+ "paddh $f2, $f2, $f26 \n\t"
+ "psrah $f0, $f0, $f20 \n\t"
+ "psrah $f2, $f2, $f20 \n\t"
+ "packushb $f0, $f0, $f2 \n\t"
+
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ PTR_ADDU "%[pPred], %[pPred], %[kiStride] \n\t"
+ "gssdxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ : [pPred] "+&r"((unsigned char *)pPred)
+ : [kiStride] "r"((int)kiStride), [mmi_wd_0x02] "r"((short *)mmi_wd_0x02)
+ : "memory", "$8", "$f0", "$f2", "$f4", "$f6"
+ );
+ RECOVER_REG;
+}
+
+void WelsDecoderI4x4LumaPredH_mmi(uint8_t *pPred, const int32_t kiStride) {+ __asm__ volatile(
+ ".set arch=loongson3a \n\t"
+ "gsldxc1 $f8, 0x0(%[mmi_01bytes], $0) \n\t"
+ "lbu $8, -0x1(%[pPred]) \n\t"
+ "dmtc1 $8, $f0 \n\t"
+ "pmuluw $f0, $f0, $f8 \n\t"
+
+ PTR_ADDU "$9, %[pPred], %[kiStride] \n\t"
+ "lbu $8, -0x1($9) \n\t"
+ "dmtc1 $8, $f2 \n\t"
+ "pmuluw $f2, $f2, $f8 \n\t"
+
+ PTR_ADDU "$10, $9, %[kiStride] \n\t"
+ "lbu $8, -0x1($10) \n\t"
+ "dmtc1 $8, $f4 \n\t"
+ "pmuluw $f4, $f4, $f8 \n\t"
+
+ PTR_ADDU "$11, $10, %[kiStride] \n\t"
+ "lbu $8, -0x1($11) \n\t"
+ "dmtc1 $8, $f6 \n\t"
+ "pmuluw $f6, $f6, $f8 \n\t"
+
+ "gsswxc1 $f0, 0x0(%[pPred], $0) \n\t"
+ "gsswxc1 $f2, 0x0($9, $0) \n\t"
+ "gsswxc1 $f4, 0x0($10, $0) \n\t"
+ "gsswxc1 $f6, 0x0($11, $0) \n\t"
+ : [pPred] "+&r"((unsigned char *)pPred)
+ : [kiStride] "r"((int)kiStride),
+ [mmi_01bytes] "r"((unsigned char *)mmi_01bytes)
+ : "memory", "$8", "$9", "$10", "$11", "$f0", "$f2", "$f4", "$f6", "$f8"
+ );
+}
--- a/codec/decoder/core/src/decoder.cpp
+++ b/codec/decoder/core/src/decoder.cpp
@@ -1023,6 +1023,23 @@
#endif
#endif
+
+#if defined(HAVE_MMI)
+ if (uiCpuFlag & WELS_CPU_MMI) {+ pCtx->pIdctResAddPredFunc = IdctResAddPred_mmi;
+
+ pCtx->pGetI16x16LumaPredFunc[I16_PRED_DC] = WelsDecoderI16x16LumaPredDc_mmi;
+ pCtx->pGetI16x16LumaPredFunc[I16_PRED_P] = WelsDecoderI16x16LumaPredPlane_mmi;
+ pCtx->pGetI16x16LumaPredFunc[I16_PRED_H] = WelsDecoderI16x16LumaPredH_mmi;
+ pCtx->pGetI16x16LumaPredFunc[I16_PRED_V] = WelsDecoderI16x16LumaPredV_mmi;
+ pCtx->pGetI16x16LumaPredFunc[I16_PRED_DC_T ] = WelsDecoderI16x16LumaPredDcTop_mmi;
+ pCtx->pGetI16x16LumaPredFunc[I16_PRED_DC_128] = WelsDecoderI16x16LumaPredDcNA_mmi;
+ pCtx->pGetIChromaPredFunc[C_PRED_P ] = WelsDecoderIChromaPredPlane_mmi;
+ pCtx->pGetIChromaPredFunc[C_PRED_DC] = WelsDecoderIChromaPredDc_mmi;
+ pCtx->pGetIChromaPredFunc[C_PRED_DC_T] = WelsDecoderIChromaPredDcTop_mmi;
+ pCtx->pGetI4x4LumaPredFunc[I4_PRED_H] = WelsDecoderI4x4LumaPredH_mmi;
+ }
+#endif//HAVE_MMI
}
//reset decoder number related statistics info
--- a/codec/decoder/targets.mk
+++ b/codec/decoder/targets.mk
@@ -53,10 +53,22 @@
endif
OBJS += $(DECODER_OBJSARM64)
+DECODER_ASM_MIPS_SRCS=\
+ $(DECODER_SRCDIR)/core/mips/dct_mmi.c\
+
+DECODER_OBJSMIPS += $(DECODER_ASM_MIPS_SRCS:.c=.$(OBJ))
+ifeq ($(ASM_ARCH), mips)
+DECODER_OBJS += $(DECODER_OBJSMIPS)
+endif
+OBJS += $(DECODER_OBJSMIPS)
+
OBJS += $(DECODER_OBJS)
$(DECODER_SRCDIR)/%.$(OBJ): $(DECODER_SRCDIR)/%.cpp
$(QUIET_CXX)$(CXX) $(CFLAGS) $(CXXFLAGS) $(INCLUDES) $(DECODER_CFLAGS) $(DECODER_INCLUDES) -c $(CXX_O) $<
+
+$(DECODER_SRCDIR)/%.$(OBJ): $(DECODER_SRCDIR)/%.c
+ $(QUIET_CC)$(CC) $(CFLAGS) $(INCLUDES) $(DECODER_CFLAGS) $(DECODER_INCLUDES) -c $(CXX_O) $<
$(DECODER_SRCDIR)/%.$(OBJ): $(DECODER_SRCDIR)/%.asm
$(QUIET_ASM)$(ASM) $(ASMFLAGS) $(ASM_INCLUDES) $(DECODER_ASMFLAGS) $(DECODER_ASM_INCLUDES) -o $@ $<
--- /dev/null
+++ b/codec/processing/src/mips/vaa_mmi.c
@@ -1,0 +1,892 @@
+/*!
+ * \copy
+ * Copyright (c) 2009-2018, Cisco Systems
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ *
+ * * Redistributions of source code must retain the above copyright
+ * notice, this list of conditions and the following disclaimer.
+ *
+ * * Redistributions in binary form must reproduce the above copyright
+ * notice, this list of conditions and the following disclaimer in
+ * the documentation and/or other materials provided with the
+ * distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
+ * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
+ * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
+ * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
+ * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
+ * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ * POSSIBILITY OF SUCH DAMAGE.
+ *
+ *
+ * \file vaa_mmi.c
+ *
+ * \brief Loongson optimization
+ *
+ * \date 23/07/2018 Created
+ *
+ *************************************************************************************
+ */
+#include <stdint.h>
+#include "asmdefs_mmi.h"
+
+//f4 is 0x1, f6 is 0x8
+#define WELS_MAX_REG_MMI(f0, f2, f4, f6) \
+ "punpckhwd $f4, "#f0", "#f0" \n\t" \
+ "punpckhwd $f6, "#f2", "#f2" \n\t" \
+ "pmaxub "#f0", "#f0", $f4 \n\t" \
+ "pmaxub "#f2", "#f2", $f6 \n\t" \
+ "pshufh $f4, "#f0", "#f4" \n\t" \
+ "pshufh $f6, "#f2", "#f4" \n\t" \
+ "pmaxub "#f0", "#f0", $f4 \n\t" \
+ "pmaxub "#f2", "#f2", $f6 \n\t" \
+ "dsrl $f4, "#f0", "#f6" \n\t" \
+ "dsrl $f6, "#f2", "#f6" \n\t" \
+ "pmaxub "#f0", "#f0", $f4 \n\t" \
+ "pmaxub "#f2", "#f2", $f6 \n\t"
+
+#define WELS_SAD_SD_MAD_16x1_MMI(f0, f2, f4, f6, f8, f10, f12, f14, r0, r1, r2) \
+ "gslqc1 $f6, $f4, 0x0("#r0") \n\t" \+ "gslqc1 $f10, $f8, 0x0("#r1") \n\t" \+ "pasubub $f12, $f4, $f0 \n\t" \
+ "pasubub $f14, $f6, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw "#f4", "#f4", $f12 \n\t" \
+ "paddw "#f6", "#f6", $f14 \n\t" \
+ "pasubub $f12, $f8, $f0 \n\t" \
+ "pasubub $f14, $f10, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw "#f8", "#f8", $f12 \n\t" \
+ "paddw "#f10", "#f10", $f14 \n\t" \
+ "pasubub $f12, $f4, $f8 \n\t" \
+ "pasubub $f14, $f6, $f10 \n\t" \
+ "pmaxub "#f12", "#f12", $f12 \n\t" \
+ "pmaxub "#f14", "#f14", $f14 \n\t" \
+ "pasubub $f12, $f12, $f0 \n\t" \
+ "pasubub $f14, $f14, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw "#f0", "#f0", $f12 \n\t" \
+ "paddw "#f2", "#f2", $f14 \n\t" \
+ PTR_ADDU ""#r0", "#r0", "#r2" \n\t" \
+ PTR_ADDU ""#r1", "#r1", "#r2" \n\t"
+
+#define WELS_SAD_16x2_MMI(f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, r1, r2, r3) \
+ "gslqc1 "#f1", "#f2", 0x00("#r1") \n\t" \+ "gslqc1 "#f3", "#f4", 0x00("#r2") \n\t" \+ PTR_ADDU ""#r1", "#r1", "#r3" \n\t" \
+ "gslqc1 "#f5", "#f6", 0x00("#r1") \n\t" \+ PTR_ADDU ""#r2", "#r2", "#r3" \n\t" \
+ "gslqc1 "#f7", "#f8", 0x00("#r2") \n\t" \+ "pasubub "#f1", "#f1", "#f3" \n\t" \
+ "pasubub "#f2", "#f2", "#f4" \n\t" \
+ "biadd "#f1", "#f1" \n\t" \
+ "biadd "#f2", "#f2" \n\t" \
+ "pasubub "#f5", "#f5", "#f7" \n\t" \
+ "pasubub "#f6", "#f6", "#f8" \n\t" \
+ "biadd "#f5", "#f5" \n\t" \
+ "biadd "#f6", "#f6" \n\t" \
+ "paddw "#f9", "#f9", "#f1" \n\t" \
+ "paddw "#f9", "#f9", "#f5" \n\t" \
+ "paddw "#f10", "#f10", "#f2" \n\t" \
+ "paddw "#f10", "#f10", "#f6" \n\t" \
+ PTR_ADDU ""#r1", "#r1", "#r3" \n\t" \
+ PTR_ADDU ""#r2", "#r2", "#r3" \n\t"
+
+#define WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI(r0, r1, r2) \
+ "gslqc1 $f6, $f4, 0x0("#r0") \n\t" \+ "gslqc1 $f10, $f8, 0x0("#r1") \n\t" \+ "pasubub $f12, $f4, $f8 \n\t" \
+ "pasubub $f14, $f6, $f10 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw $f28, $f28, $f12 \n\t" \
+ "paddw $f30, $f30, $f14 \n\t" \
+ "pasubub $f12, $f4, $f8 \n\t" \
+ "pasubub $f14, $f6, $f10 \n\t" \
+ "pasubub $f8, $f4, $f0 \n\t" \
+ "pasubub $f10, $f6, $f2 \n\t" \
+ "biadd $f8, $f8 \n\t" \
+ "biadd $f10, $f10 \n\t" \
+ "paddw $f24, $f24, $f8 \n\t" \
+ "paddw $f26, $f26, $f10 \n\t" \
+ "punpcklbh $f8, $f6, $f2 \n\t" \
+ "punpckhbh $f10, $f6, $f2 \n\t" \
+ "punpckhbh $f6, $f4, $f0 \n\t" \
+ "punpcklbh $f4, $f4, $f0 \n\t" \
+ "pmaddhw $f4, $f4, $f4 \n\t" \
+ "pmaddhw $f6, $f6, $f6 \n\t" \
+ "pmaddhw $f8, $f8, $f8 \n\t" \
+ "pmaddhw $f10, $f10, $f10 \n\t" \
+ "paddw $f20, $f20, $f4 \n\t" \
+ "paddw $f22, $f22, $f6 \n\t" \
+ "paddw $f20, $f20, $f8 \n\t" \
+ "paddw $f22, $f22, $f10 \n\t" \
+ "punpcklbh $f4, $f12, $f0 \n\t" \
+ "punpckhbh $f6, $f12, $f0 \n\t" \
+ "punpcklbh $f12, $f14, $f2 \n\t" \
+ "punpckhbh $f14, $f14, $f2 \n\t" \
+ "pmaddhw $f4, $f4, $f4 \n\t" \
+ "pmaddhw $f6, $f6, $f6 \n\t" \
+ "pmaddhw $f12, $f12, $f12 \n\t" \
+ "pmaddhw $f14, $f14, $f14 \n\t" \
+ "paddw $f16, $f16, $f4 \n\t" \
+ "paddw $f18, $f18, $f6 \n\t" \
+ "paddw $f16, $f16, $f12 \n\t" \
+ "paddw $f18, $f18, $f14 \n\t" \
+ PTR_ADDU ""#r0", "#r0", "#r2" \n\t" \
+ PTR_ADDU ""#r1", "#r1", "#r2" \n\t"
+
+#define WELS_SAD_BGD_SQDIFF_16x1_MMI(f0, f2, f4, f6, f8, f10, f12, f14, r0, r1, r2) \
+ "gslqc1 $f6, $f4, 0x0("#r0") \n\t" \+ "punpcklbh $f8, $f4, $f0 \n\t" \
+ "punpckhbh $f10, $f4, $f0 \n\t" \
+ "punpcklbh $f12, $f6, $f2 \n\t" \
+ "punpckhbh $f14, $f6, $f2 \n\t" \
+ "pmaddhw $f8, $f8, $f8 \n\t" \
+ "pmaddhw $f10, $f10, $f10 \n\t" \
+ "pmaddhw $f12, $f12, $f12 \n\t" \
+ "pmaddhw $f14, $f14, $f14 \n\t" \
+ "paddw $f8, $f8, $f12 \n\t" \
+ "paddw $f10, $f10, $f14 \n\t" \
+ "punpckhwd $f12, $f0, $f8 \n\t" \
+ "punpckhwd $f14, $f0, $f10 \n\t" \
+ "punpcklwd $f8, $f0, $f8 \n\t" \
+ "punpcklwd $f10, $f0, $f10 \n\t" \
+ "paddw $f8, $f8, $f12 \n\t" \
+ "paddw $f10, $f10, $f14 \n\t" \
+ "paddw "#f0", "#f0", $f8 \n\t" \
+ "paddw "#f2", "#f2", $f10 \n\t" \
+ "gslqc1 $f10, $f8, 0x0("#r1") \n\t" \+ "pasubub $f12, $f4, $f0 \n\t" \
+ "pasubub $f14, $f6, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw "#f4", "#f4", $f12 \n\t" \
+ "paddw "#f6", "#f6", $f14 \n\t" \
+ "pasubub $f12, $f8, $f0 \n\t" \
+ "pasubub $f14, $f10, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "punpcklwd $f14, $f14, $f14 \n\t" \
+ "punpckhwd $f14, $f12, $f14 \n\t" \
+ "punpcklwd $f12, $f0, $f12 \n\t" \
+ "paddw "#f4", "#f4", $f12 \n\t" \
+ "paddw "#f6", "#f6", $f14 \n\t" \
+ "pasubub $f12, $f4, $f8 \n\t" \
+ "pasubub $f14, $f6, $f10 \n\t" \
+ "pmaxub "#f8", "#f8", $f12 \n\t" \
+ "pmaxub "#f10", "#f10", $f14 \n\t" \
+ "paddw $f4, $f0, $f12 \n\t" \
+ "paddw $f6, $f0, $f14 \n\t" \
+ "pasubub $f12, $f12, $f0 \n\t" \
+ "pasubub $f14, $f14, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw "#f0", "#f0", $f12 \n\t" \
+ "paddw "#f2", "#f2", $f14 \n\t" \
+ "paddw $f12, $f0, $f4 \n\t" \
+ "paddw $f14, $f0, $f6 \n\t" \
+ "punpcklbh $f4, $f12, $f0 \n\t" \
+ "punpckhbh $f6, $f12, $f0 \n\t" \
+ "punpcklbh $f12, $f14, $f2 \n\t" \
+ "punpckhbh $f14, $f14, $f2 \n\t" \
+ "pmaddhw $f4, $f4, $f4 \n\t" \
+ "pmaddhw $f6, $f6, $f6 \n\t" \
+ "pmaddhw $f12, $f12, $f12 \n\t" \
+ "pmaddhw $f14, $f14, $f14 \n\t" \
+ "paddw "#f12", "#f12", $f4 \n\t" \
+ "paddw "#f14", "#f14", $f6 \n\t" \
+ "paddw "#f12", "#f12", $f12 \n\t" \
+ "paddw "#f14", "#f14", $f14 \n\t" \
+ PTR_ADDU ""#r0", "#r0", "#r2" \n\t" \
+ PTR_ADDU ""#r1", "#r1", "#r2" \n\t"
+
+#define WELS_SAD_SUM_SQSUM_16x1_MMI(r0, r1, r2) \
+ "gslqc1 $f6, $f4, 0x0("#r0") \n\t" \+ "gslqc1 $f10, $f8, 0x0("#r1") \n\t" \+ "pasubub $f12, $f4, $f8 \n\t" \
+ "pasubub $f14, $f6, $f10 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw $f24, $f24, $f12 \n\t" \
+ "paddw $f26, $f26, $f14 \n\t" \
+ "pasubub $f12, $f4, $f0 \n\t" \
+ "pasubub $f14, $f6, $f2 \n\t" \
+ "biadd $f12, $f12 \n\t" \
+ "biadd $f14, $f14 \n\t" \
+ "paddw $f20, $f20, $f12 \n\t" \
+ "paddw $f22, $f22, $f14 \n\t" \
+ "punpcklbh $f8, $f6, $f2 \n\t" \
+ "punpckhbh $f10, $f6, $f2 \n\t" \
+ "punpckhbh $f6, $f4, $f0 \n\t" \
+ "punpcklbh $f4, $f4, $f0 \n\t" \
+ "pmaddhw $f4, $f4, $f4 \n\t" \
+ "pmaddhw $f6, $f6, $f6 \n\t" \
+ "pmaddhw $f8, $f8, $f8 \n\t" \
+ "pmaddhw $f10, $f10, $f10 \n\t" \
+ "paddw $f16, $f16, $f4 \n\t" \
+ "paddw $f18, $f18, $f6 \n\t" \
+ "paddw $f16, $f16, $f8 \n\t" \
+ "paddw $f18, $f18, $f10 \n\t" \
+ PTR_ADDU ""#r0", "#r0", "#r2" \n\t" \
+ PTR_ADDU ""#r1", "#r1", "#r2" \n\t"
+
+void VAACalcSad_mmi(const uint8_t* pCurData, const uint8_t* pRefData,
+ int32_t iPicWidth, int32_t iPicHeight, int32_t iPicStride,
+ int32_t* pFrameSad, int32_t* pSad8x8) {+ double ftmp[13];
+ uint64_t tmp[2];
+ mips_reg addr[3];
+
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ PTR_SRL "%[iPicWidth], %[iPicWidth], 0x04 \n\t"
+ PTR_SRL "%[iPicHeight], %[iPicHeight], 0x04 \n\t"
+ "move %[addr2], %[iPicStride] \n\t"
+ PTR_SLL "%[iPicStride], %[iPicStride], 0x04 \n\t"
+ "xor %[ftmp0], %[ftmp0], %[ftmp0] \n\t"
+ "xor %[ftmp11], %[ftmp11], %[ftmp11] \n\t"
+ "xor %[ftmp12], %[ftmp12], %[ftmp12] \n\t"
+ "1: \n\t"
+ "move %[addr0], %[pCurData] \n\t"
+ "move %[addr1], %[pRefData] \n\t"
+ "move %[tmp0], %[iPicWidth] \n\t"
+ "2: \n\t"
+ "xor %[ftmp9], %[ftmp9], %[ftmp9] \n\t"
+ "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ "paddw %[ftmp11], %[ftmp11], %[ftmp9] \n\t"
+ "paddw %[ftmp12], %[ftmp12], %[ftmp10] \n\t"
+ "swc1 %[ftmp10], 0x00(%[pSad8x8]) \n\t"
+ "swc1 %[ftmp9], 0x04(%[pSad8x8]) \n\t"
+
+ "xor %[ftmp9], %[ftmp9], %[ftmp9] \n\t"
+ "xor %[ftmp10], %[ftmp10], %[ftmp10] \n\t"
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ WELS_SAD_16x2_MMI(%[ftmp1], %[ftmp2], %[ftmp3], %[ftmp4], %[ftmp5],
+ %[ftmp6], %[ftmp7], %[ftmp8], %[ftmp9], %[ftmp10],
+ %[addr0], %[addr1], %[addr2])
+ "paddw %[ftmp11], %[ftmp11], %[ftmp9] \n\t"
+ "paddw %[ftmp12], %[ftmp12], %[ftmp10] \n\t"
+ "swc1 %[ftmp10], 0x08(%[pSad8x8]) \n\t"
+ "swc1 %[ftmp9], 0x0c(%[pSad8x8]) \n\t"
+
+ PTR_ADDU "%[pSad8x8], %[pSad8x8], 0x10 \n\t"
+ PTR_SUBU "%[addr0], %[addr0], %[iPicStride] \n\t"
+ PTR_SUBU "%[addr1], %[addr1], %[iPicStride] \n\t"
+ PTR_ADDI "%[tmp0], %[tmp0], -0x01 \n\t"
+ PTR_ADDU "%[addr0], %[addr0], 0x10 \n\t"
+ PTR_ADDU "%[addr1], %[addr1], 0x10 \n\t"
+ "bnez %[tmp0], 2b \n\t"
+
+ PTR_ADDI "%[iPicHeight], %[iPicHeight], -0x01 \n\t"
+ PTR_ADDU "%[pCurData], %[pCurData], %[iPicStride] \n\t"
+ PTR_ADDU "%[pRefData], %[pRefData], %[iPicStride] \n\t"
+ "bnez %[iPicHeight], 1b \n\t"
+
+ "paddw %[ftmp11], %[ftmp11], %[ftmp12] \n\t"
+ "swc1 %[ftmp11], 0x00(%[pFrameSad]) \n\t"
+ : [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]), [ftmp7]"=&f"(ftmp[7]),
+ [ftmp8]"=&f"(ftmp[8]), [ftmp9]"=&f"(ftmp[9]),
+ [ftmp10]"=&f"(ftmp[10]), [ftmp11]"=&f"(ftmp[11]),
+ [ftmp12]"=&f"(ftmp[12]), [tmp0]"=&r"(tmp[0]),
+ [addr0]"=&r"(addr[0]), [addr1]"=&r"(addr[1]),
+ [pCurData]"+&r"(pCurData), [pRefData]"+&r"(pRefData),
+ [iPicHeight]"+&r"(iPicHeight), [iPicWidth]"+&r"(iPicWidth),
+ [pSad8x8]"+&r"(pSad8x8), [iPicStride]"+&r"(iPicStride),
+ [addr2]"=&r"(addr[2])
+ : [pFrameSad]"r"(pFrameSad)
+ : "memory"
+ );
+}
+
+void VAACalcSadBgd_mmi(const uint8_t *cur_data, const uint8_t *ref_data,
+ int32_t iPicWidth, int32_t iPicHeight, int32_t iPicStride,
+ int32_t *psadframe, int32_t *psad8x8, int32_t *p_sd8x8,
+ uint8_t *p_mad8x8) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $15, %[cur_data] \n\t"
+ "dsrl %[iPicWidth], %[iPicWidth], 0x4 \n\t"
+ "dsrl %[iPicHeight], %[iPicHeight], 0x4 \n\t"
+ "dsll $13, %[iPicStride], 0x4 \n\t"
+ "xor $f0, $f0, $f0 \n\t"
+ "xor $f2, $f2, $f2 \n\t"
+ "xor $14, $14, $14 \n\t"
+ "1: \n\t"
+ "move $9, %[iPicWidth] \n\t"
+ "move $10, $15 \n\t"
+ "move $11, %[ref_data] \n\t"
+ "2: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "xor $f30, $f30, $f30 \n\t"
+ "xor $f24, $f24, $f24 \n\t"
+ "xor $f26, $f26, $f26 \n\t"
+ "xor $f20, $f20, $f20 \n\t"
+ "xor $f22, $f22, $f22 \n\t"
+ "xor $f16, $f16, $f16 \n\t"
+ "xor $f18, $f18, $f18 \n\t"
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+
+ "dli $8, 0x1 \n\t"
+ "dmtc1 $8, $f8 \n\t"
+ "dli $8, 0x8 \n\t"
+ "dmtc1 $8, $f10 \n\t"
+ WELS_MAX_REG_MMI($f16, $f18, $f8, $f10)
+
+ "dmfc1 $8, $f16 \n\t"
+ "sb $8, 0x0(%[p_mad8x8]) \n\t"
+ "dmfc1 $8, $f18 \n\t"
+ "sb $8, 0x1(%[p_mad8x8]) \n\t"
+ PTR_ADDIU "%[p_mad8x8], %[p_mad8x8], 0x2 \n\t"
+
+ "xor $f16, $f16, $f16 \n\t"
+ "xor $f18, $f18, $f18 \n\t"
+ "punpcklwd $f30, $f30, $f30 \n\t"
+ "punpcklwd $f26, $f26, $f26 \n\t"
+ "punpcklwd $f22, $f22, $f22 \n\t"
+
+ "punpckhwd $f30, $f28, $f30 \n\t"
+ "punpckhwd $f26, $f24, $f26 \n\t"
+ "punpckhwd $f22, $f20, $f22 \n\t"
+
+ "punpcklwd $f28, $f16, $f28 \n\t"
+ "punpcklwd $f24, $f16, $f24 \n\t"
+ "punpcklwd $f20, $f16, $f20 \n\t"
+
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+ WELS_SAD_SD_MAD_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16, $f18,
+ $15, %[ref_data], %[iPicStride])
+
+ "dli $8, 0x1 \n\t"
+ "dmtc1 $8, $f8 \n\t"
+ "dli $8, 0x8 \n\t"
+ "dmtc1 $8, $f10 \n\t"
+ WELS_MAX_REG_MMI($f16, $f18, $f8, $f10)
+
+ "dmfc1 $8, $f16 \n\t"
+ "sb $8, 0x0(%[p_mad8x8]) \n\t"
+ "dmfc1 $8, $f18 \n\t"
+ "sb $8, 0x1(%[p_mad8x8]) \n\t"
+ "punpckhwd $f4, $f28, $f30 \n\t"
+ PTR_ADDIU "%[p_mad8x8], %[p_mad8x8], 0x2 \n\t"
+
+ "punpcklwd $f6, $f28, $f30 \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[psad8x8]) \n\t"
+ PTR_ADDIU "%[psad8x8], %[psad8x8], 0x10 \n\t"
+
+ "paddw $f6, $f6, $f30 \n\t"
+ "paddw $f4, $f4, $f28 \n\t"
+ "punpckhwd $f8, $f6, $f6 \n\t"
+ "paddw $f4, $f4, $f8 \n\t"
+ "dmtc1 $14, $f6 \n\t"
+ "paddw $f6, $f6, $f4 \n\t"
+ "dmfc1 $14, $f6 \n\t"
+
+ "psubw $f24, $f24, $f20 \n\t"
+ "psubw $f26, $f26, $f22 \n\t"
+ "punpckhwd $f4, $f24, $f26 \n\t"
+ "punpcklwd $f6, $f24, $f26 \n\t"
+ "gssqc1 $f6, $f4, 0x0(%[p_sd8x8]) \n\t"
+ PTR_ADDIU "%[p_sd8x8], %[p_sd8x8], 0x10 \n\t"
+
+ PTR_SUBU "$15, $15, $13 \n\t"
+ PTR_SUBU "%[ref_data], %[ref_data], $13 \n\t"
+ PTR_ADDIU "$15, $15, 0x10 \n\t"
+ PTR_ADDIU "%[ref_data], %[ref_data], 0x10 \n\t"
+
+ PTR_ADDIU "%[iPicWidth], %[iPicWidth], -0x1 \n\t"
+ "bnez %[iPicWidth], 2b \n\t"
+ "move %[iPicWidth], $9 \n\t"
+ "move $15, $10 \n\t"
+ "move %[ref_data], $11 \n\t"
+ PTR_ADDU "$15, $15, $13 \n\t"
+ PTR_ADDU "%[ref_data], %[ref_data], $13 \n\t"
+
+ PTR_ADDIU "%[iPicHeight], %[iPicHeight], -0x1 \n\t"
+ "bnez %[iPicHeight], 1b \n\t"
+
+ "swl $14, 0x3(%[psadframe]) \n\t"
+ "swr $14, 0x0(%[psadframe]) \n\t"
+ : [ref_data]"+&r"((unsigned char *)ref_data), [iPicWidth]"+&r"((int)iPicWidth),
+ [iPicHeight]"+&r"((int)iPicHeight), [psad8x8]"+&r"((int *)psad8x8),
+ [p_sd8x8]"+&r"((int *)p_sd8x8), [p_mad8x8]"+&r"((unsigned char *)p_mad8x8)
+ : [cur_data]"r"((unsigned char *)cur_data), [iPicStride]"r"((int)iPicStride),
+ [psadframe]"r"((int *)psadframe)
+ : "memory", "$8", "$9", "$10", "$11", "$12", "$13", "$14", "$15", "$f0", "$f2",
+ "$f4", "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22",
+ "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void VAACalcSadSsd_mmi(const uint8_t *cur_data, const uint8_t *ref_data,
+ int32_t iPicWidth, int32_t iPicHeight, int32_t iPicStride,
+ int32_t *psadframe, int32_t *psad8x8, int32_t *psum16x16,
+ int32_t *psqsum16x16, int32_t *psqdiff16x16) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $15, %[cur_data] \n\t"
+ "dsrl %[iPicWidth], %[iPicWidth], 0x4 \n\t"
+ "dsrl %[iPicHeight], %[iPicHeight], 0x4 \n\t"
+ "dsll $13, %[iPicStride], 0x4 \n\t"
+ "xor $f0, $f0, $f0 \n\t"
+ "xor $f2, $f2, $f2 \n\t"
+ "xor $12, $12, $12 \n\t"
+ "xor $14, $14, $14 \n\t"
+ "1: \n\t"
+ "move $9, %[iPicWidth] \n\t"
+ "move $10, $15 \n\t"
+ "move $11, %[ref_data] \n\t"
+ "2: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "xor $f30, $f30, $f30 \n\t"
+ "xor $f24, $f24, $f24 \n\t"
+ "xor $f26, $f26, $f26 \n\t"
+ "xor $f20, $f20, $f20 \n\t"
+ "xor $f22, $f22, $f22 \n\t"
+ "xor $f16, $f16, $f16 \n\t"
+ "xor $f18, $f18, $f18 \n\t"
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ "dmfc1 $8, $f28 \n\t"
+ "sw $8, 0x0(%[psad8x8]) \n\t"
+ "dmfc1 $8, $f30 \n\t"
+ "sw $8, 0x4(%[psad8x8]) \n\t"
+ "paddw $f4, $f28, $f30 \n\t"
+ "dmfc1 $12, $f4 \n\t"
+ PTR_ADDU "$14, $14, $12 \n\t"
+
+ "xor $f28, $f28, $f28 \n\t"
+ "xor $f30, $f30, $f30 \n\t"
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_MMI($15, %[ref_data], %[iPicStride])
+ "dmfc1 $8, $f28 \n\t"
+ "sw $8, 0x8(%[psad8x8]) \n\t"
+ "dmfc1 $8, $f30 \n\t"
+ "paddw $f4, $f28, $f30 \n\t"
+ "sw $8, 0xc(%[psad8x8]) \n\t"
+ "dmfc1 $12, $f4 \n\t"
+ PTR_ADDU "$14, $14, $12 \n\t"
+ PTR_ADDIU "%[psad8x8], %[psad8x8], 0x10 \n\t"
+
+ "paddw $f24, $f24, $f26 \n\t"
+ "dmfc1 $8, $f24 \n\t"
+ "sw $8, 0x0(%[psum16x16]) \n\t"
+ PTR_ADDIU "%[psum16x16], %[psum16x16], 0x4 \n\t"
+ "paddw $f24, $f20, $f22 \n\t"
+ "punpcklwd $f20, $f24, $f24 \n\t"
+ "punpckhwd $f22, $f24, $f24 \n\t"
+ "paddw $f20, $f20, $f22 \n\t"
+ "dmfc1 $8, $f20 \n\t"
+ "sw $8, 0x0(%[psqsum16x16]) \n\t"
+ PTR_ADDIU "%[psqsum16x16], %[psqsum16x16], 0x4 \n\t"
+
+ "paddw $f20, $f16, $f18 \n\t"
+ "punpcklwd $f16, $f20, $f20 \n\t"
+ "punpckhwd $f18, $f20, $f20 \n\t"
+ "paddw $f16, $f16, $f18 \n\t"
+ "dmfc1 $8, $f16 \n\t"
+ "sw $8, 0x0(%[psqdiff16x16]) \n\t"
+ PTR_ADDIU "%[psqdiff16x16], %[psqdiff16x16], 0x4 \n\t"
+
+ PTR_SUBU "$15, $15, $13 \n\t"
+ PTR_SUBU "%[ref_data], %[ref_data], $13 \n\t"
+ PTR_ADDIU "$15, $15, 0x10 \n\t"
+ PTR_ADDIU "%[ref_data], %[ref_data], 0x10 \n\t"
+
+ PTR_ADDIU "%[iPicWidth], %[iPicWidth], -0x1 \n\t"
+ "bnez %[iPicWidth], 2b \n\t"
+ "nop \n\t"
+ "move %[iPicWidth], $9 \n\t"
+ "move $15, $10 \n\t"
+ "move %[ref_data], $11 \n\t"
+ PTR_ADDU "$15, $15, $13 \n\t"
+ PTR_ADDU "%[ref_data], %[ref_data], $13 \n\t"
+
+ PTR_ADDIU "%[iPicHeight], %[iPicHeight], -0x1 \n\t"
+ "bnez %[iPicHeight], 1b \n\t"
+ "nop \n\t"
+
+ "sw $14, 0x0(%[psadframe]) \n\t"
+ : [ref_data]"+&r"((unsigned char *)ref_data), [iPicWidth]"+&r"((int)iPicWidth),
+ [iPicHeight]"+&r"((int)iPicHeight), [psum16x16]"+&r"((int *)psum16x16),
+ [psqsum16x16]"+&r"((int *)psqsum16x16), [psqdiff16x16]"+&r"((int *)psqdiff16x16)
+ : [cur_data]"r"((unsigned char *)cur_data), [iPicStride]"r"((int)iPicStride),
+ [psadframe]"r"((int *)psadframe), [psad8x8]"r"((int *)psad8x8)
+ : "memory", "$8", "$9", "$10", "$11", "$12", "$13", "$14", "$15", "$f0", "$f2",
+ "$f4", "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22",
+ "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void VAACalcSadSsdBgd_mmi(const uint8_t *cur_data, const uint8_t *ref_data,
+ int32_t iPicWidth, int32_t iPicHeight, int32_t iPicStride,
+ int32_t *psadframe, int32_t *psad8x8, int32_t *psum16x16,
+ int32_t *psqsum16x16, int32_t *psqdiff16x16, int32_t *p_sd8x8,
+ uint8_t *p_mad8x8) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $15, %[cur_data] \n\t"
+ "dsrl %[iPicWidth], %[iPicWidth], 0x4 \n\t"
+ "dsrl %[iPicHeight], %[iPicHeight], 0x4 \n\t"
+ "dsll $13, %[iPicStride], 0x4 \n\t"
+ "xor $f0, $f0, $f0 \n\t"
+ "xor $f2, $f2, $f2 \n\t"
+ "xor $12, $12, $12 \n\t"
+ "xor $14, $14, $14 \n\t"
+ "1: \n\t"
+ "move $9, %[iPicWidth] \n\t"
+ "move $10, $15 \n\t"
+ "move $11, %[ref_data] \n\t"
+ "2: \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "xor $f30, $f30, $f30 \n\t"
+ "xor $f24, $f24, $f24 \n\t"
+ "xor $f26, $f26, $f26 \n\t"
+ "xor $f20, $f20, $f20 \n\t"
+ "xor $f22, $f22, $f22 \n\t"
+ "xor $f16, $f16, $f16 \n\t"
+ "xor $f18, $f18, $f18 \n\t"
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+
+ "dmfc1 $8, $f28 \n\t"
+ "sw $8, 0x0(%[psad8x8]) \n\t"
+ "dmfc1 $8, $f30 \n\t"
+ "sw $8, 0x4(%[psad8x8]) \n\t"
+ PTR_ADDIU "%[psad8x8], %[psad8x8], 0x8 \n\t"
+
+ "paddw $f4, $f28, $f30 \n\t"
+ "dmfc1 $12, $f4 \n\t"
+ PTR_ADDU "$14, $14, $12 \n\t"
+
+ "paddw $f4, $f24, $f26 \n\t"
+ "dmfc1 $8, $f4 \n\t"
+ "sw $8, 0x0(%[psum16x16]) \n\t"
+
+ "punpckhwd $f4, $f24, $f26 \n\t"
+ "punpcklwd $f6, $f24, $f26 \n\t"
+ "psubw $f6, $f6, $f4 \n\t"
+ "dmfc1 $8, $f6 \n\t"
+ PTR_S "$8, 0x0(%[p_sd8x8]) \n\t"
+ PTR_ADDIU "%[p_sd8x8], %[p_sd8x8], 0x8 \n\t"
+
+ "dli $8, 0x1 \n\t"
+ "dmtc1 $8, $f8 \n\t"
+ "dli $8, 0x8 \n\t"
+ "dmtc1 $8, $f10 \n\t"
+ WELS_MAX_REG_MMI($f20, $f22, $f8, $f10)
+
+ "dmfc1 $8, $f20 \n\t"
+ "sb $8, 0x0(%[p_mad8x8]) \n\t"
+ "dmfc1 $8, $f22 \n\t"
+ "sb $8, 0x1(%[p_mad8x8]) \n\t"
+ PTR_ADDIU "%[p_mad8x8], %[p_mad8x8], 0x2 \n\t"
+
+ "xor $f20, $f20, $f20 \n\t"
+ "xor $f22, $f22, $f22 \n\t"
+ "punpckhwd $f28, $f20, $f28 \n\t"
+ "xor $f24, $f24, $f24 \n\t"
+ "xor $f26, $f26, $f26 \n\t"
+ "punpckhwd $f30, $f20, $f30 \n\t"
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+ WELS_SAD_BGD_SQDIFF_16x1_MMI($f28, $f30, $f24, $f26, $f20, $f22, $f16,
+ $f18, $15, %[ref_data], %[iPicStride])
+
+ "dmfc1 $8, $f28 \n\t"
+ "sw $8, 0x0(%[psad8x8]) \n\t"
+ "dmfc1 $8, $f30 \n\t"
+ "sw $8, 0x4(%[psad8x8]) \n\t"
+ PTR_ADDIU "%[psad8x8], %[psad8x8], 0x8 \n\t"
+
+ "paddw $f4, $f28, $f30 \n\t"
+ "dmfc1 $12, $f4 \n\t"
+ PTR_ADDU "$14, $14, $12 \n\t"
+
+ "paddw $f4, $f24, $f26 \n\t"
+ "dmfc1 $8, $f4 \n\t"
+ "lw $12, 0x0(%[psum16x16]) \n\t"
+ PTR_ADDU "$8, $8, $12 \n\t"
+ "sw $8, 0x0(%[psum16x16]) \n\t"
+ "xor $f8, $f8, $f8 \n\t"
+ PTR_ADDIU "%[psum16x16], %[psum16x16], 0x4 \n\t"
+
+ "punpckhwd $f30, $f30, $f8 \n\t"
+ "punpckhwd $f28, $f28, $f8 \n\t"
+ "paddw $f8, $f28, $f30 \n\t"
+ "dmfc1 $8, $f8 \n\t"
+ "sw $8, 0x0(%[psqsum16x16]) \n\t"
+ PTR_ADDIU "%[psqsum16x16], %[psqsum16x16], 0x4 \n\t"
+
+ "punpckhwd $f4, $f24, $f26 \n\t"
+ "punpcklwd $f6, $f24, $f26 \n\t"
+ "psubw $f6, $f6, $f4 \n\t"
+ "dmfc1 $8, $f6 \n\t"
+ PTR_S "$8, 0x0(%[p_sd8x8]) \n\t"
+ PTR_ADDIU "%[p_sd8x8], %[p_sd8x8], 0x8 \n\t"
+
+ "dli $8, 0x1 \n\t"
+ "dmtc1 $8, $f8 \n\t"
+ "dli $8, 0x8 \n\t"
+ "dmtc1 $8, $f10 \n\t"
+ WELS_MAX_REG_MMI($f20, $f22, $f8, $f10)
+
+ "dmfc1 $8, $f20 \n\t"
+ "sb $8, 0x0(%[p_mad8x8]) \n\t"
+ "dmfc1 $8, $f22 \n\t"
+ "sb $8, 0x1(%[p_mad8x8]) \n\t"
+ PTR_ADDIU "%[p_mad8x8], %[p_mad8x8], 0x2 \n\t"
+
+ "paddw $f20, $f16, $f18 \n\t"
+ "punpcklwd $f16, $f20, $f20 \n\t"
+ "punpckhwd $f18, $f20, $f20 \n\t"
+ "paddw $f16, $f16, $f18 \n\t"
+ "dmfc1 $8, $f16 \n\t"
+ "sw $8, 0x0(%[psqdiff16x16]) \n\t"
+ PTR_ADDIU "%[psqdiff16x16], %[psqdiff16x16], 0x4 \n\t"
+
+ PTR_SUBU "$15, $15, $13 \n\t"
+ PTR_SUBU "%[ref_data], %[ref_data], $13 \n\t"
+ PTR_ADDIU "$15, $15, 0x10 \n\t"
+ PTR_ADDIU "%[ref_data], %[ref_data], 0x10 \n\t"
+
+ PTR_ADDIU "%[iPicWidth], %[iPicWidth], -0x1 \n\t"
+ "bnez %[iPicWidth], 2b \n\t"
+ "nop \n\t"
+ "move %[iPicWidth], $9 \n\t"
+ "move $15, $10 \n\t"
+ "move %[ref_data], $11 \n\t"
+ PTR_ADDU "$15, $15, $13 \n\t"
+ PTR_ADDU "%[ref_data], %[ref_data], $13 \n\t"
+
+ PTR_ADDIU "%[iPicHeight], %[iPicHeight], -0x1 \n\t"
+ "bnez %[iPicHeight], 1b \n\t"
+ "nop \n\t"
+
+ "sw $14, 0x0(%[psadframe]) \n\t"
+ : [ref_data]"+&r"((unsigned char *)ref_data), [iPicWidth]"+&r"((int)iPicWidth),
+ [iPicHeight]"+&r"((int)iPicHeight), [psad8x8]"+&r"((int *)psad8x8),
+ [psum16x16]"+&r"((int *)psum16x16), [psqsum16x16]"+&r"((int *)psqsum16x16),
+ [psqdiff16x16]"+&r"((int *)psqdiff16x16), [p_sd8x8]"+&r"((int *)p_sd8x8),
+ [p_mad8x8]"+&r"((unsigned char *)p_mad8x8)
+ : [cur_data]"r"((unsigned char *)cur_data), [iPicStride]"r"((int)iPicStride),
+ [psadframe]"r"((int *)psadframe)
+ : "memory", "$8", "$9", "$10", "$11", "$12", "$13", "$14", "$15", "$f0", "$f2",
+ "$f4", "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22",
+ "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
+
+void VAACalcSadVar_mmi(const uint8_t *cur_data, const uint8_t *ref_data,
+ int32_t iPicWidth, int32_t iPicHeight, int32_t iPicStride,
+ int32_t *psadframe, int32_t *psad8x8, int32_t *psum16x16,
+ int32_t *psqsum16x16) {+ BACKUP_REG;
+ __asm__ volatile (
+ ".set arch=loongson3a \n\t"
+ "move $15, %[cur_data] \n\t"
+ "dsrl %[iPicWidth], %[iPicWidth], 0x4 \n\t"
+ "dsrl %[iPicHeight], %[iPicHeight], 0x4 \n\t"
+ "dsll $13, %[iPicStride], 0x4 \n\t"
+ "xor $f0, $f0, $f0 \n\t"
+ "xor $f2, $f2, $f2 \n\t"
+ "xor $f28, $f28, $f28 \n\t"
+ "xor $f30, $f30, $f30 \n\t"
+ "xor $14, $14, $14 \n\t"
+ "1: \n\t"
+ "move $9, %[iPicWidth] \n\t"
+ "move $10, $15 \n\t"
+ "move $11, %[ref_data] \n\t"
+ "2: \n\t"
+ "xor $f24, $f24, $f24 \n\t"
+ "xor $f26, $f26, $f26 \n\t"
+ "xor $f20, $f20, $f20 \n\t"
+ "xor $f22, $f22, $f22 \n\t"
+ "xor $f16, $f16, $f16 \n\t"
+ "xor $f18, $f18, $f18 \n\t"
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ "paddw $f28, $f24, $f28 \n\t"
+ "paddw $f30, $f26, $f30 \n\t"
+ "dmfc1 $8, $f24 \n\t"
+ "sw $8, 0x0(%[psad8x8]) \n\t"
+ "dmfc1 $8, $f26 \n\t"
+ "sw $8, 0x4(%[psad8x8]) \n\t"
+
+ "xor $f24, $f24, $f24 \n\t"
+ "xor $f26, $f26, $f26 \n\t"
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ WELS_SAD_SUM_SQSUM_16x1_MMI($15, %[ref_data], %[iPicStride])
+ "paddw $f28, $f24, $f28 \n\t"
+ "paddw $f30, $f26, $f30 \n\t"
+ "dmfc1 $8, $f24 \n\t"
+ "sw $8, 0x8(%[psad8x8]) \n\t"
+ "dmfc1 $8, $f26 \n\t"
+ "sw $8, 0xc(%[psad8x8]) \n\t"
+ PTR_ADDIU "%[psad8x8], %[psad8x8], 0x10 \n\t"
+
+ "paddw $f20, $f20, $f22 \n\t"
+ "dmfc1 $8, $f20 \n\t"
+ "sw $8, 0x0(%[psum16x16]) \n\t"
+ PTR_ADDIU "%[psum16x16], %[psum16x16], 0x4 \n\t"
+
+ "paddw $f20, $f16, $f18 \n\t"
+ "punpcklwd $f16, $f20, $f20 \n\t"
+ "punpckhwd $f18, $f20, $f20 \n\t"
+ "paddw $f16, $f16, $f18 \n\t"
+ "dmfc1 $8, $f16 \n\t"
+ "sw $8, 0x0(%[psqsum16x16]) \n\t"
+ PTR_ADDIU "%[psqsum16x16], %[psqsum16x16], 0x4 \n\t"
+
+ PTR_SUBU "$15, $15, $13 \n\t"
+ PTR_SUBU "%[ref_data], %[ref_data], $13 \n\t"
+ PTR_ADDIU "$15, $15, 0x10 \n\t"
+ PTR_ADDIU "%[ref_data], %[ref_data], 0x10 \n\t"
+
+ PTR_ADDIU "%[iPicWidth], %[iPicWidth], -0x1 \n\t"
+ "bnez %[iPicWidth], 2b \n\t"
+ "nop \n\t"
+ "move %[iPicWidth], $9 \n\t"
+ "move $15, $10 \n\t"
+ "move %[ref_data], $11 \n\t"
+ PTR_ADDU "$15, $15, $13 \n\t"
+ PTR_ADDU "%[ref_data], %[ref_data], $13 \n\t"
+
+ PTR_ADDIU "%[iPicHeight], %[iPicHeight], -0x1 \n\t"
+ "bnez %[iPicHeight], 1b \n\t"
+ "nop \n\t"
+
+ "paddw $f28, $f28, $f30 \n\t"
+ "dmfc1 $8, $f28 \n\t"
+ "sw $8, 0x0(%[psadframe]) \n\t"
+ : [ref_data]"+&r"((unsigned char *)ref_data), [iPicWidth]"+&r"((int)iPicWidth),
+ [iPicHeight]"+&r"((int)iPicHeight), [psum16x16]"+&r"((int *)psum16x16),
+ [psqsum16x16]"+&r"((int *)psqsum16x16)
+ : [cur_data]"r"((unsigned char *)cur_data), [iPicStride]"r"((int)iPicStride),
+ [psadframe]"r"((int *)psadframe), [psad8x8]"r"((int *)psad8x8)
+ : "memory", "$8", "$9", "$10", "$11", "$12", "$13", "$14", "$15", "$f0", "$f2",
+ "$f4", "$f6", "$f8", "$f10", "$f12", "$f14", "$f16", "$f18", "$f20", "$f22",
+ "$f24", "$f26", "$f28", "$f30"
+ );
+ RECOVER_REG;
+}
--- a/codec/processing/src/vaacalc/vaacalculation.cpp
+++ b/codec/processing/src/vaacalc/vaacalculation.cpp
@@ -93,6 +93,16 @@
sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_AArch64_neon;
}
#endif//HAVE_NEON_AARCH64
+
+#ifdef HAVE_MMI
+ if ((iCpuFlag & WELS_CPU_MMI) == WELS_CPU_MMI) {+ sVaaFuncs.pfVAACalcSad = VAACalcSad_mmi;
+ sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_mmi;
+ sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_mmi;
+ sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_mmi;
+ sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_mmi;
+ }
+#endif//HAVE_MMI
}
EResult CVAACalculation::Process (int32_t iType, SPixMap* pSrcPixMap, SPixMap* pRefPixMap) {--- a/codec/processing/src/vaacalc/vaacalculation.h
+++ b/codec/processing/src/vaacalc/vaacalculation.h
@@ -132,6 +132,16 @@
WELSVP_EXTERN_C_END
#endif
+#ifdef HAVE_MMI
+WELSVP_EXTERN_C_BEGIN
+VAACalcSadBgdFunc VAACalcSadBgd_mmi;
+VAACalcSadSsdBgdFunc VAACalcSadSsdBgd_mmi;
+VAACalcSadFunc VAACalcSad_mmi;
+VAACalcSadVarFunc VAACalcSadVar_mmi;
+VAACalcSadSsdFunc VAACalcSadSsd_mmi;
+WELSVP_EXTERN_C_END
+#endif
+
class CVAACalculation : public IStrategy {public:
CVAACalculation (int32_t iCpuFlag);
--- a/codec/processing/targets.mk
+++ b/codec/processing/targets.mk
@@ -55,10 +55,22 @@
endif
OBJS += $(PROCESSING_OBJSARM64)
+PROCESSING_ASM_MIPS_SRCS=\
+ $(PROCESSING_SRCDIR)/src/mips/vaa_mmi.c\
+
+PROCESSING_OBJSMIPS += $(PROCESSING_ASM_MIPS_SRCS:.c=.$(OBJ))
+ifeq ($(ASM_ARCH), mips)
+PROCESSING_OBJS += $(PROCESSING_OBJSMIPS)
+endif
+OBJS += $(PROCESSING_OBJSMIPS)
+
OBJS += $(PROCESSING_OBJS)
$(PROCESSING_SRCDIR)/%.$(OBJ): $(PROCESSING_SRCDIR)/%.cpp
$(QUIET_CXX)$(CXX) $(CFLAGS) $(CXXFLAGS) $(INCLUDES) $(PROCESSING_CFLAGS) $(PROCESSING_INCLUDES) -c $(CXX_O) $<
+
+$(PROCESSING_SRCDIR)/%.$(OBJ): $(PROCESSING_SRCDIR)/%.c
+ $(QUIET_CC)$(CC) $(CFLAGS) $(INCLUDES) $(PROCESSING_CFLAGS) $(PROCESSING_INCLUDES) -c $(CXX_O) $<
$(PROCESSING_SRCDIR)/%.$(OBJ): $(PROCESSING_SRCDIR)/%.asm
$(QUIET_ASM)$(ASM) $(ASMFLAGS) $(ASM_INCLUDES) $(PROCESSING_ASMFLAGS) $(PROCESSING_ASM_INCLUDES) -o $@ $<
--- a/test/decoder/DecUT_IdctResAddPred.cpp
+++ b/test/decoder/DecUT_IdctResAddPred.cpp
@@ -154,6 +154,10 @@
GENERATE_IDCTRESADDPRED (IdctResAddPred_AArch64_neon, WELS_CPU_NEON)
#endif
+#if defined(HAVE_MMI)
+GENERATE_IDCTRESADDPRED (IdctResAddPred_mmi, WELS_CPU_MMI)
+#endif
+
#define GENERATE_SETNONZEROCOUNT(method, flag) \
TEST(DecoderDecodeMbAux, method) \
{\--- a/test/decoder/DecUT_IntraPrediction.cpp
+++ b/test/decoder/DecUT_IntraPrediction.cpp
@@ -649,3 +649,16 @@
GENERATE_8x8_UT (WelsDecoderIChromaPredPlane_AArch64_neon, WelsIChromaPredPlane_ref, 1, WELS_CPU_NEON)
GENERATE_8x8_UT (WelsDecoderIChromaPredDcTop_AArch64_neon, WelsIChromaPredDcTop_ref, 1, WELS_CPU_NEON)
#endif
+
+#if defined(HAVE_MMI)
+GENERATE_4x4_UT (WelsDecoderI4x4LumaPredH_mmi, LumaI4x4PredH, 1, WELS_CPU_MMI)
+GENERATE_8x8_UT (WelsDecoderIChromaPredDcTop_mmi, WelsIChromaPredDcTop_ref, 1, WELS_CPU_MMI)
+GENERATE_8x8_UT (WelsDecoderIChromaPredDc_mmi, WelsIChromaPredDc_ref, 1, WELS_CPU_MMI)
+GENERATE_8x8_UT (WelsDecoderIChromaPredPlane_mmi, WelsIChromaPredPlane_ref, 1, WELS_CPU_MMI)
+GENERATE_16x16_UT (WelsDecoderI16x16LumaPredPlane_mmi, WelsI16x16LumaPredPlane_ref, 1, WELS_CPU_MMI)
+GENERATE_16x16_UT (WelsDecoderI16x16LumaPredH_mmi, LumaI16x16PredH, 1, WELS_CPU_MMI)
+GENERATE_16x16_UT (WelsDecoderI16x16LumaPredV_mmi, LumaI16x16PredV, 1, WELS_CPU_MMI)
+GENERATE_16x16_UT (WelsDecoderI16x16LumaPredDc_mmi, LumaI16x16PredDC, 1, WELS_CPU_MMI)
+GENERATE_16x16_UT (WelsDecoderI16x16LumaPredDcTop_mmi, LumaI16x16PredDCTop, 1, WELS_CPU_MMI)
+GENERATE_16x16_UT (WelsDecoderI16x16LumaPredDcNA_mmi, LumaI16x16PredDCNone, 1, WELS_CPU_MMI)
+#endif
--- a/test/encoder/EncUT_EncoderMbAux.cpp
+++ b/test/encoder/EncUT_EncoderMbAux.cpp
@@ -74,6 +74,39 @@
FREE_MEMORY (iDct);
}
#endif
+#ifdef HAVE_MMI
+TEST (EncodeMbAuxTest, WelsScan4x4Ac_mmi) {+ CMemoryAlign cMemoryAlign (0);
+ ALLOC_MEMORY (int16_t, iLevelA, 16);
+ ALLOC_MEMORY (int16_t, iLevelB, 16);
+ ALLOC_MEMORY (int16_t, iDct, 16);
+ for (int i = 0; i < 16; i++) {+ iDct[i] = rand() % 256 + 1;
+ }
+ WelsScan4x4Ac_c (iLevelA, iDct);
+ WelsScan4x4Ac_mmi (iLevelB, iDct);
+ for (int j = 0; j < 16; j++)
+ EXPECT_EQ (iLevelA[j], iLevelB[j]);
+ FREE_MEMORY (iLevelA);
+ FREE_MEMORY (iLevelB);
+ FREE_MEMORY (iDct);
+}
+TEST (EncodeMbAuxTest, WelsScan4x4DcAc_mmi) {+ CMemoryAlign cMemoryAlign (0);
+ ALLOC_MEMORY (int16_t, iLevelA, 32);
+ ALLOC_MEMORY (int16_t, iLevelB, 32);
+ ALLOC_MEMORY (int16_t, iDct, 32);
+ for (int i = 0; i < 32; i++)
+ iDct[i] = (rand() & 32767) - 16384;
+ WelsScan4x4DcAc_mmi (iLevelA, iDct);
+ WelsScan4x4DcAc_c (iLevelB, iDct);
+ for (int i = 0; i < 16; i++)
+ EXPECT_EQ (iLevelA[i], iLevelB[i]);
+ FREE_MEMORY (iLevelA);
+ FREE_MEMORY (iLevelB);
+ FREE_MEMORY (iDct);
+}
+#endif
TEST (EncodeMbAuxTest, TestScan_4x4_dcc) {CMemoryAlign cMemoryAlign (0);
ALLOC_MEMORY (int16_t, iLevel, 16);
@@ -230,6 +263,29 @@
iDctC[i] = iDctS[i] = (rand() & 65535) - 32768;
WelsCalculateSingleCtr4x4_c (iDctC);
WelsCalculateSingleCtr4x4_sse2 (iDctS);
+ for (int i = 0; i < 16; i++)
+ EXPECT_EQ (iDctC[i], iDctS[i]);
+ FREE_MEMORY (iDctC);
+ FREE_MEMORY (iDctS);
+}
+#endif
+#ifdef HAVE_MMI
+TEST (EncodeMbAuxTest, WelsDctT4_mmi) {+ TestDctT4 (WelsDctT4_mmi);
+}
+
+TEST (EncodeMbAuxTest, WelsDctFourT4_mmi) {+ TestDctFourT4 (WelsDctFourT4_mmi);
+}
+
+TEST (EncodeMbAuxTest, WelsCalculateSingleCtr4x4_mmi) {+ CMemoryAlign cMemoryAlign (0);
+ ALLOC_MEMORY (int16_t, iDctC, 16);
+ ALLOC_MEMORY (int16_t, iDctS, 16);
+ for (int i = 0; i < 16; i++)
+ iDctC[i] = iDctS[i] = (rand() & 65535) - 32768;
+ WelsCalculateSingleCtr4x4_c (iDctC);
+ WelsCalculateSingleCtr4x4_mmi (iDctS);
for (int i = 0; i < 16; i++)
EXPECT_EQ (iDctC[i], iDctS[i]);
FREE_MEMORY (iDctC);
--- a/test/processing/ProcessUT_VaaCalc.cpp
+++ b/test/processing/ProcessUT_VaaCalc.cpp
@@ -863,3 +863,11 @@
GENERATE_VAACalcSadSsd_UT (VAACalcSadSsd_AArch64_neon, 1, WELS_CPU_NEON)
GENERATE_VAACalcSadVar_UT (VAACalcSadVar_AArch64_neon, 1, WELS_CPU_NEON)
#endif
+
+#if defined(HAVE_MMI)
+GENERATE_VAACalcSad_UT (VAACalcSad_mmi, 1, WELS_CPU_MMI)
+GENERATE_VAACalcSadBgd_UT (VAACalcSadBgd_mmi, 1, WELS_CPU_MMI)
+GENERATE_VAACalcSadSsdBgd_UT (VAACalcSadSsdBgd_mmi, 1, WELS_CPU_MMI)
+GENERATE_VAACalcSadSsd_UT (VAACalcSadSsd_mmi, 1, WELS_CPU_MMI)
+GENERATE_VAACalcSadVar_UT (VAACalcSadVar_mmi, 1, WELS_CPU_MMI)
+#endif