summaryrefslogtreecommitdiff
path: root/src/libffmpeg/libavcodec/alpha
diff options
context:
space:
mode:
Diffstat (limited to 'src/libffmpeg/libavcodec/alpha')
-rw-r--r--src/libffmpeg/libavcodec/alpha/asm.h181
-rw-r--r--src/libffmpeg/libavcodec/alpha/dsputil_alpha.c113
-rw-r--r--src/libffmpeg/libavcodec/alpha/mpegvideo_alpha.c3
3 files changed, 160 insertions, 137 deletions
diff --git a/src/libffmpeg/libavcodec/alpha/asm.h b/src/libffmpeg/libavcodec/alpha/asm.h
index 0f4685f11..ceaf0be4a 100644
--- a/src/libffmpeg/libavcodec/alpha/asm.h
+++ b/src/libffmpeg/libavcodec/alpha/asm.h
@@ -20,122 +20,115 @@
#ifndef LIBAVCODEC_ALPHA_ASM_H
#define LIBAVCODEC_ALPHA_ASM_H
-#include <stdint.h>
+#include <inttypes.h>
#define AMASK_BWX (1 << 0)
#define AMASK_FIX (1 << 1)
+#define AMASK_CIX (1 << 2)
#define AMASK_MVI (1 << 8)
-static inline uint64_t BYTE_VEC(uint64_t x)
+inline static uint64_t BYTE_VEC(uint64_t x)
{
x |= x << 8;
x |= x << 16;
x |= x << 32;
return x;
}
-static inline uint64_t WORD_VEC(uint64_t x)
+inline static uint64_t WORD_VEC(uint64_t x)
{
x |= x << 16;
x |= x << 32;
return x;
}
-static inline int32_t ldl(const void* p)
-{
- return *(const int32_t*) p;
-}
-static inline uint64_t ldq(const void* p)
-{
- return *(const uint64_t*) p;
-}
-/* FIXME ccc doesn't seem to get it? Use inline asm? */
-static inline uint64_t ldq_u(const void* p)
-{
- return *(const uint64_t*) ((uintptr_t) p & ~7ul);
-}
-static inline void stl(uint32_t l, void* p)
-{
- *(uint32_t*) p = l;
-}
-static inline void stq(uint64_t l, void* p)
-{
- *(uint64_t*) p = l;
-}
+#define ldq(p) (*(const uint64_t *) (p))
+#define ldl(p) (*(const int32_t *) (p))
+#define stl(l, p) do { *(uint32_t *) (p) = (l); } while (0)
+#define stq(l, p) do { *(uint64_t *) (p) = (l); } while (0)
#ifdef __GNUC__
-#define OPCODE1(name) \
-static inline uint64_t name(uint64_t l) \
-{ \
- uint64_t r; \
- asm (#name " %1, %0" : "=r" (r) : "r" (l)); \
- return r; \
-}
-
-#define OPCODE2(name) \
-static inline uint64_t name(uint64_t l1, uint64_t l2) \
-{ \
- uint64_t r; \
- asm (#name " %1, %2, %0" : "=r" (r) : "r" (l1), "rI" (l2)); \
- return r; \
-}
-
-/* We don't want gcc to move this around or combine it with another
- rpcc, so mark it volatile. */
-static inline uint64_t rpcc(void)
-{
- uint64_t r;
- asm volatile ("rpcc %0" : "=r" (r));
- return r;
-}
-
-static inline uint64_t uldq(const void* v)
-{
- struct foo {
- unsigned long l;
- } __attribute__((packed));
-
- return ((const struct foo*) v)->l;
-}
+#define ASM_ACCEPT_MVI asm (".arch pca56")
+struct unaligned_long { uint64_t l; } __attribute__((packed));
+#define ldq_u(p) (*(const uint64_t *) (((uint64_t) (p)) & ~7ul))
+#define uldq(a) (((const struct unaligned_long *) (a))->l)
+
+#if __GNUC__ >= 3 && __GNUC_MINOR__ >= 2
+#define cmpbge __builtin_alpha_cmpbge
+/* Avoid warnings. */
+#define extql(a, b) __builtin_alpha_extql(a, (uint64_t) (b))
+#define extqh(a, b) __builtin_alpha_extqh(a, (uint64_t) (b))
+#define zap __builtin_alpha_zap
+#define zapnot __builtin_alpha_zapnot
+#define amask __builtin_alpha_amask
+#define implver __builtin_alpha_implver
+#define rpcc __builtin_alpha_rpcc
+#define minub8 __builtin_alpha_minub8
+#define minsb8 __builtin_alpha_minsb8
+#define minuw4 __builtin_alpha_minuw4
+#define minsw4 __builtin_alpha_minsw4
+#define maxub8 __builtin_alpha_maxub8
+#define maxsb8 __builtin_alpha_maxsb8
+#define maxuw4 __builtin_alpha_maxuw4
+#define maxsw4 __builtin_alpha_maxsw4
+#define perr __builtin_alpha_perr
+#define pklb __builtin_alpha_pklb
+#define pkwb __builtin_alpha_pkwb
+#define unpkbl __builtin_alpha_unpkbl
+#define unpkbw __builtin_alpha_unpkbw
+#else
+#define cmpbge(a, b) ({ uint64_t __r; asm ("cmpbge %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
+#define extql(a, b) ({ uint64_t __r; asm ("extql %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
+#define extqh(a, b) ({ uint64_t __r; asm ("extqh %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
+#define zap(a, b) ({ uint64_t __r; asm ("zap %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
+#define zapnot(a, b) ({ uint64_t __r; asm ("zapnot %r1,%2,%0" : "=r" (__r) : "rJ" (a), "rI" (b)); __r; })
+#define amask(a) ({ uint64_t __r; asm ("amask %1,%0" : "=r" (__r) : "rI" (a)); __r; })
+#define implver() ({ uint64_t __r; asm ("implver %0" : "=r" (__r)); __r; })
+#define rpcc() ({ uint64_t __r; asm volatile ("rpcc %0" : "=r" (__r)); __r; })
+#define minub8(a, b) ({ uint64_t __r; asm ("minub8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define minsb8(a, b) ({ uint64_t __r; asm ("minsb8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define minuw4(a, b) ({ uint64_t __r; asm ("minuw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define minsw4(a, b) ({ uint64_t __r; asm ("minsw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define maxub8(a, b) ({ uint64_t __r; asm ("maxub8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define maxsb8(a, b) ({ uint64_t __r; asm ("maxsb8 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define maxuw4(a, b) ({ uint64_t __r; asm ("maxuw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define maxsw4(a, b) ({ uint64_t __r; asm ("maxsw4 %r1,%2,%0" : "=r" (__r) : "%rJ" (a), "rI" (b)); __r; })
+#define perr(a, b) ({ uint64_t __r; asm ("perr %r1,%r2,%0" : "=r" (__r) : "%rJ" (a), "rJ" (b)); __r; })
+#define pklb(a) ({ uint64_t __r; asm ("pklb %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
+#define pkwb(a) ({ uint64_t __r; asm ("pkwb %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
+#define unpkbl(a) ({ uint64_t __r; asm ("unpkbl %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
+#define unpkbw(a) ({ uint64_t __r; asm ("unpkbw %r1,%0" : "=r" (__r) : "rJ" (a)); __r; })
+#endif
-#elif defined(__DECC) /* Compaq "ccc" compiler */
+#elif defined(__DECC) /* Digital/Compaq "ccc" compiler */
#include <c_asm.h>
-#define OPCODE1(name) \
-static inline uint64_t name(uint64_t l) \
-{ \
- return asm (#name " %a0, %v0", l); \
-}
-
-#define OPCODE2(name) \
-static inline uint64_t name(uint64_t l1, uint64_t l2) \
-{ \
- return asm (#name " %a0, %a1, %v0", l1, l2); \
-}
-
-static inline uint64_t rpcc(void)
-{
- return asm ("rpcc %v0");
-}
-
-static inline uint64_t uldq(const void* v)
-{
- return *(const __unaligned uint64_t *) v;
-}
-
+#define ASM_ACCEPT_MVI
+#define ldq_u(a) asm ("ldq_u %v0,0(%a0)", a)
+#define uldq(a) (*(const __unaligned uint64_t *) (a))
+#define cmpbge(a, b) asm ("cmpbge %a0,%a1,%v0", a, b)
+#define extql(a, b) asm ("extql %a0,%a1,%v0", a, b)
+#define extqh(a, b) asm ("extqh %a0,%a1,%v0", a, b)
+#define zap(a, b) asm ("zap %a0,%a1,%v0", a, b)
+#define zapnot(a, b) asm ("zapnot %a0,%a1,%v0", a, b)
+#define amask(a) asm ("amask %a0,%v0", a)
+#define implver() asm ("implver %v0")
+#define rpcc() asm ("rpcc %v0")
+#define minub8(a, b) asm ("minub8 %a0,%a1,%v0", a, b)
+#define minsb8(a, b) asm ("minsb8 %a0,%a1,%v0", a, b)
+#define minuw4(a, b) asm ("minuw4 %a0,%a1,%v0", a, b)
+#define minsw4(a, b) asm ("minsw4 %a0,%a1,%v0", a, b)
+#define maxub8(a, b) asm ("maxub8 %a0,%a1,%v0", a, b)
+#define maxsb8(a, b) asm ("maxsb8 %a0,%a1,%v0", a, b)
+#define maxuw4(a, b) asm ("maxuw4 %a0,%a1,%v0", a, b)
+#define maxsw4(a, b) asm ("maxsw4 %a0,%a1,%v0", a, b)
+#define perr(a, b) asm ("perr %a0,%a1,%v0", a, b)
+#define pklb(a) asm ("pklb %a0,%v0", a)
+#define pkwb(a) asm ("pkwb %a0,%v0", a)
+#define unpkbl(a) asm ("unpkbl %a0,%v0", a)
+#define unpkbw(a) asm ("unpkbw %a0,%v0", a)
+
+#else
+#error "Unknown compiler!"
#endif
-OPCODE1(amask);
-OPCODE1(unpkbw);
-OPCODE1(pkwb);
-OPCODE2(extql);
-OPCODE2(extqh);
-OPCODE2(zap);
-OPCODE2(cmpbge);
-OPCODE2(minsw4);
-OPCODE2(minuw4);
-OPCODE2(minub8);
-OPCODE2(maxsw4);
-OPCODE2(maxuw4);
-OPCODE2(perr);
-
#endif /* LIBAVCODEC_ALPHA_ASM_H */
diff --git a/src/libffmpeg/libavcodec/alpha/dsputil_alpha.c b/src/libffmpeg/libavcodec/alpha/dsputil_alpha.c
index 3a54904f4..5e1aa2093 100644
--- a/src/libffmpeg/libavcodec/alpha/dsputil_alpha.c
+++ b/src/libffmpeg/libavcodec/alpha/dsputil_alpha.c
@@ -22,58 +22,86 @@
void simple_idct_axp(DCTELEM *block);
-static void put_pixels_clamped_axp(const DCTELEM *block, UINT8 *pixels,
- int line_size)
+void put_pixels_clamped_mvi_asm(const DCTELEM *block, uint8_t *pixels,
+ int line_size);
+void add_pixels_clamped_mvi_asm(const DCTELEM *block, uint8_t *pixels,
+ int line_size);
+
+#if 0
+/* These functions were the base for the optimized assembler routines,
+ and remain here for documentation purposes. */
+static void put_pixels_clamped_mvi(const DCTELEM *block, uint8_t *pixels,
+ int line_size)
{
int i = 8;
+ uint64_t clampmask = zap(-1, 0xaa); /* 0x00ff00ff00ff00ff */
+
+ ASM_ACCEPT_MVI;
+
do {
- UINT64 shorts;
+ uint64_t shorts0, shorts1;
- shorts = ldq(block);
- shorts = maxsw4(shorts, 0);
- shorts = minsw4(shorts, WORD_VEC(0x00ff));
- stl(pkwb(shorts), pixels);
+ shorts0 = ldq(block);
+ shorts0 = maxsw4(shorts0, 0);
+ shorts0 = minsw4(shorts0, clampmask);
+ stl(pkwb(shorts0), pixels);
- shorts = ldq(block + 4);
- shorts = maxsw4(shorts, 0);
- shorts = minsw4(shorts, WORD_VEC(0x00ff));
- stl(pkwb(shorts), pixels + 4);
+ shorts1 = ldq(block + 4);
+ shorts1 = maxsw4(shorts1, 0);
+ shorts1 = minsw4(shorts1, clampmask);
+ stl(pkwb(shorts1), pixels + 4);
- pixels += line_size;
- block += 8;
+ pixels += line_size;
+ block += 8;
} while (--i);
}
-static void add_pixels_clamped_axp(const DCTELEM *block, UINT8 *pixels,
- int line_size)
+void add_pixels_clamped_mvi(const DCTELEM *block, uint8_t *pixels,
+ int line_size)
{
- int i = 8;
+ int h = 8;
+ /* Keep this function a leaf function by generating the constants
+ manually (mainly for the hack value ;-). */
+ uint64_t clampmask = zap(-1, 0xaa); /* 0x00ff00ff00ff00ff */
+ uint64_t signmask = zap(-1, 0x33);
+ signmask ^= signmask >> 1; /* 0x8000800080008000 */
+
+ ASM_ACCEPT_MVI;
+
do {
- UINT64 shorts;
-
- shorts = ldq(block);
- shorts &= ~WORD_VEC(0x8000); /* clear highest bit to avoid overflow */
- shorts += unpkbw(ldl(pixels));
- shorts &= ~WORD_VEC(0x8000); /* hibit would be set for e. g. -2 + 3 */
- shorts = minuw4(shorts, WORD_VEC(0x4000)); /* set neg. to 0x4000 */
- shorts &= ~WORD_VEC(0x4000); /* ...and zap them */
- shorts = minsw4(shorts, WORD_VEC(0x00ff)); /* clamp to 255 */
- stl(pkwb(shorts), pixels);
-
- /* next 4 */
- shorts = ldq(block + 4);
- shorts &= ~WORD_VEC(0x8000);
- shorts += unpkbw(ldl(pixels + 4));
- shorts &= ~WORD_VEC(0x8000);
- shorts = minuw4(shorts, WORD_VEC(0x4000));
- shorts &= ~WORD_VEC(0x4000);
- shorts = minsw4(shorts, WORD_VEC(0x00ff));
- stl(pkwb(shorts), pixels + 4);
-
- pixels += line_size;
- block += 8;
- } while (--i);
+ uint64_t shorts0, pix0, signs0;
+ uint64_t shorts1, pix1, signs1;
+
+ shorts0 = ldq(block);
+ shorts1 = ldq(block + 4);
+
+ pix0 = unpkbw(ldl(pixels));
+ /* Signed subword add (MMX paddw). */
+ signs0 = shorts0 & signmask;
+ shorts0 &= ~signmask;
+ shorts0 += pix0;
+ shorts0 ^= signs0;
+ /* Clamp. */
+ shorts0 = maxsw4(shorts0, 0);
+ shorts0 = minsw4(shorts0, clampmask);
+
+ /* Next 4. */
+ pix1 = unpkbw(ldl(pixels + 4));
+ signs1 = shorts1 & signmask;
+ shorts1 &= ~signmask;
+ shorts1 += pix1;
+ shorts1 ^= signs1;
+ shorts1 = maxsw4(shorts1, 0);
+ shorts1 = minsw4(shorts1, clampmask);
+
+ stl(pkwb(shorts0), pixels);
+ stl(pkwb(shorts1), pixels + 4);
+
+ pixels += line_size;
+ block += 8;
+ } while (--h);
}
+#endif
/* Average 8 unsigned bytes in parallel: (b1 + b2) >> 1
Since the immediate result could be greater than 255, we do the
@@ -216,8 +244,7 @@ void dsputil_init_alpha(void)
/* amask clears all bits that correspond to present features. */
if (amask(AMASK_MVI) == 0) {
- fprintf(stderr, "MVI extension detected\n");
- put_pixels_clamped = put_pixels_clamped_axp;
- add_pixels_clamped = add_pixels_clamped_axp;
+ put_pixels_clamped = put_pixels_clamped_mvi_asm;
+ add_pixels_clamped = add_pixels_clamped_mvi_asm;
}
}
diff --git a/src/libffmpeg/libavcodec/alpha/mpegvideo_alpha.c b/src/libffmpeg/libavcodec/alpha/mpegvideo_alpha.c
index d0af5e1d3..eb1997eee 100644
--- a/src/libffmpeg/libavcodec/alpha/mpegvideo_alpha.c
+++ b/src/libffmpeg/libavcodec/alpha/mpegvideo_alpha.c
@@ -28,6 +28,9 @@ static void dct_unquantize_h263_axp(MpegEncContext *s,
{
int i, level;
UINT64 qmul, qadd;
+
+ ASM_ACCEPT_MVI;
+
if (s->mb_intra) {
if (n < 4)
block[0] = block[0] * s->y_dc_scale;
generated by cgit v1.2.3 (git 2.25.1) at 2025-07-18 02:18:32 +0000