avcodec/x86/idctdsp: Remove obsolete MMX(EXT) functions

x64 always has MMX, MMXEXT, SSE and SSE2 and this means that some functions for MMX, MMXEXT and 3dnow are always overridden by other functions (unless one e.g. explicitly disables SSE2) for x64. So given that the only systems that benefit from these functions are truely ancient 32bit x86s they are removed.

Andreas Rheinhardt committed 4 years ago

Commit bfb28b51 parent b2437a45 files · +10 −111Browse files

Unified · Split

5 files

libavcodec/tests/x86/dct.c +0 −3
libavcodec/x86/idctdsp.asm +4 −75
libavcodec/x86/idctdsp.h +0 −6
libavcodec/x86/idctdsp_init.c +4 −7
libavcodec/x86/simple_idct.asm +2 −20

libavcodec/tests/x86/dct.c+0−3

@@ -65,9 +65,6 @@ static const struct algo fdct_tab_arch[] = { · 64 unmodified lines
65	65	};
66	66
67	67	static const struct algo idct_tab_arch[] = {
68		−#if HAVE_MMX_EXTERNAL
69		− { "SIMPLE-MMX", ff_simple_idct_mmx, FF_IDCT_PERM_SIMPLE, AV_CPU_FLAG_MMX },
70		−#endif
71	68	#if CONFIG_MPEG4_DECODER && HAVE_X86ASM
72	69	#if HAVE_SSE2_EXTERNAL
73	70	{ "XVID-SSE2", ff_xvid_idct_sse2, FF_IDCT_PERM_SSE2, AV_CPU_FLAG_SSE2, 1 },

libavcodec/x86/idctdsp.asm+4−75

@@ -37,47 +37,24 @@ SECTION .text · 36 unmodified lines
37	37	%macro PUT_SIGNED_PIXELS_CLAMPED_HALF 1
38	38	mova m1, [blockq+mmsize*0+%1]
39	39	mova m2, [blockq+mmsize*2+%1]
40		−%if mmsize == 8
41		− mova m3, [blockq+mmsize*4+%1]
42		− mova m4, [blockq+mmsize*6+%1]
43		−%endif
44	40	packsswb m1, [blockq+mmsize*1+%1]
45	41	packsswb m2, [blockq+mmsize*3+%1]
46		−%if mmsize == 8
47		− packsswb m3, [blockq+mmsize*5+%1]
48		− packsswb m4, [blockq+mmsize*7+%1]
49		−%endif
50	42	paddb m1, m0
51	43	paddb m2, m0
52		−%if mmsize == 8
53		− paddb m3, m0
54		− paddb m4, m0
55		− movq [pixelsq+lsizeq*0], m1
56		− movq [pixelsq+lsizeq*1], m2
57		− movq [pixelsq+lsizeq*2], m3
58		− movq [pixelsq+lsize3q ], m4
59		−%else
60	44	movq [pixelsq+lsizeq*0], m1
61	45	movhps [pixelsq+lsizeq*1], m1
62	46	movq [pixelsq+lsizeq*2], m2
63	47	movhps [pixelsq+lsize3q ], m2
64		−%endif
65	48	%endmacro
66	49
67		−%macro PUT_SIGNED_PIXELS_CLAMPED 1
68		−cglobal put_signed_pixels_clamped, 3, 4, %1, block, pixels, lsize, lsize3
	50	+INIT_XMM sse2
	51	+cglobal put_signed_pixels_clamped, 3, 4, 3, block, pixels, lsize, lsize3
69	52	mova m0, [pb_80]
70	53	lea lsize3q, [lsizeq*3]
71	54	PUT_SIGNED_PIXELS_CLAMPED_HALF 0
72	55	lea pixelsq, [pixelsq+lsizeq*4]
73	56	PUT_SIGNED_PIXELS_CLAMPED_HALF 64
74	57	RET
75		−%endmacro
76		−
77		−INIT_MMX mmx
78		−PUT_SIGNED_PIXELS_CLAMPED 0
79		−INIT_XMM sse2
80		−PUT_SIGNED_PIXELS_CLAMPED 3
81	58
82	59	;--------------------------------------------------------------------------
83	60	; void ff_put_pixels_clamped(const int16_t block, uint8_t pixels,
@@ -87,40 +64,21 @@ PUT_SIGNED_PIXELS_CLAMPED 3 · 3 unmodified lines
87	64	%macro PUT_PIXELS_CLAMPED_HALF 1
88	65	mova m0, [blockq+mmsize*0+%1]
89	66	mova m1, [blockq+mmsize*2+%1]
90		−%if mmsize == 8
91		− mova m2, [blockq+mmsize*4+%1]
92		− mova m3, [blockq+mmsize*6+%1]
93		−%endif
94	67	packuswb m0, [blockq+mmsize*1+%1]
95	68	packuswb m1, [blockq+mmsize*3+%1]
96		−%if mmsize == 8
97		− packuswb m2, [blockq+mmsize*5+%1]
98		− packuswb m3, [blockq+mmsize*7+%1]
99		− movq [pixelsq], m0
100		− movq [lsizeq+pixelsq], m1
101		− movq [2*lsizeq+pixelsq], m2
102		− movq [lsize3q+pixelsq], m3
103		−%else
104	69	movq [pixelsq], m0
105	70	movhps [lsizeq+pixelsq], m0
106	71	movq [2*lsizeq+pixelsq], m1
107	72	movhps [lsize3q+pixelsq], m1
108		−%endif
109	73	%endmacro
110	74
111		−%macro PUT_PIXELS_CLAMPED 0
	75	+INIT_XMM sse2
112	76	cglobal put_pixels_clamped, 3, 4, 2, block, pixels, lsize, lsize3
113	77	lea lsize3q, [lsizeq*3]
114	78	PUT_PIXELS_CLAMPED_HALF 0
115	79	lea pixelsq, [pixelsq+lsizeq*4]
116	80	PUT_PIXELS_CLAMPED_HALF 64
117	81	RET
118		−%endmacro
119		−
120		−INIT_MMX mmx
121		−PUT_PIXELS_CLAMPED
122		−INIT_XMM sse2
123		−PUT_PIXELS_CLAMPED
124	82
125	83	;--------------------------------------------------------------------------
126	84	; void ff_add_pixels_clamped(const int16_t block, uint8_t pixels,
@@ -130,41 +88,18 @@ PUT_PIXELS_CLAMPED · 3 unmodified lines
130	88	%macro ADD_PIXELS_CLAMPED 1
131	89	mova m0, [blockq+mmsize*0+%1]
132	90	mova m1, [blockq+mmsize*1+%1]
133		−%if mmsize == 8
134		− mova m5, [blockq+mmsize*2+%1]
135		− mova m6, [blockq+mmsize*3+%1]
136		−%endif
137	91	movq m2, [pixelsq]
138	92	movq m3, [pixelsq+lsizeq]
139		−%if mmsize == 8
140		− mova m7, m2
141		− punpcklbw m2, m4
142		− punpckhbw m7, m4
143		− paddsw m0, m2
144		− paddsw m1, m7
145		− mova m7, m3
146		− punpcklbw m3, m4
147		− punpckhbw m7, m4
148		− paddsw m5, m3
149		− paddsw m6, m7
150		−%else
151	93	punpcklbw m2, m4
152	94	punpcklbw m3, m4
153	95	paddsw m0, m2
154	96	paddsw m1, m3
155		−%endif
156	97	packuswb m0, m1
157		−%if mmsize == 8
158		− packuswb m5, m6
159		− movq [pixelsq], m0
160		− movq [pixelsq+lsizeq], m5
161		−%else
162	98	movq [pixelsq], m0
163	99	movhps [pixelsq+lsizeq], m0
164		−%endif
165	100	%endmacro
166	101
167		−%macro ADD_PIXELS_CLAMPED 0
	102	+INIT_XMM sse2
168	103	cglobal add_pixels_clamped, 3, 3, 5, block, pixels, lsize
169	104	pxor m4, m4
170	105	ADD_PIXELS_CLAMPED 0
@@ -175,9 +110,3 @@ cglobal add_pixels_clamped, 3, 3, 5, block, pixels, lsize · 4 unmodified lines
175	110	lea pixelsq, [pixelsq+lsizeq*2]
176	111	ADD_PIXELS_CLAMPED 96
177	112	RET
178		−%endmacro
179		−
180		−INIT_MMX mmx
181		−ADD_PIXELS_CLAMPED
182		−INIT_XMM sse2
183		−ADD_PIXELS_CLAMPED

libavcodec/x86/idctdsp.h+0−6

@@ -22,16 +22,10 @@ · 21 unmodified lines
22	22	#include <stddef.h>
23	23	#include <stdint.h>
24	24
25		−void ff_add_pixels_clamped_mmx(const int16_t block, uint8_t pixels,
26		− ptrdiff_t line_size);
27	25	void ff_add_pixels_clamped_sse2(const int16_t block, uint8_t pixels,
28	26	ptrdiff_t line_size);
29		−void ff_put_pixels_clamped_mmx(const int16_t block, uint8_t pixels,
30		− ptrdiff_t line_size);
31	27	void ff_put_pixels_clamped_sse2(const int16_t block, uint8_t pixels,
32	28	ptrdiff_t line_size);
33		−void ff_put_signed_pixels_clamped_mmx(const int16_t block, uint8_t pixels,
34		− ptrdiff_t line_size);
35	29	void ff_put_signed_pixels_clamped_sse2(const int16_t block, uint8_t pixels,
36	30	ptrdiff_t line_size);
37	31

libavcodec/x86/idctdsp_init.c+4−7

@@ -63,28 +63,24 @@ av_cold void ff_idctdsp_init_x86(IDCTDSPContext c, AVCodecContext avctx, · 62 unmodified lines
63	63	{
64	64	int cpu_flags = av_get_cpu_flags();
65	65
	66	+#if ARCH_X86_32
66	67	if (EXTERNAL_MMX(cpu_flags)) {
67		− c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
68		− c->put_pixels_clamped = ff_put_pixels_clamped_mmx;
69		− c->add_pixels_clamped = ff_add_pixels_clamped_mmx;
70		−
71	68	if (!high_bit_depth &&
72	69	avctx->lowres == 0 &&
73	70	(avctx->idct_algo == FF_IDCT_AUTO \|\|
74	71	avctx->idct_algo == FF_IDCT_SIMPLEAUTO \|\|
75	72	avctx->idct_algo == FF_IDCT_SIMPLEMMX)) {
76		− c->idct_put = ff_simple_idct_put_mmx;
77		− c->idct_add = ff_simple_idct_add_mmx;
78	73	c->idct = ff_simple_idct_mmx;
79		− c->perm_type = FF_IDCT_PERM_SIMPLE;
80	74	}
81	75	}
	76	+#endif
82	77
83	78	if (EXTERNAL_SSE2(cpu_flags)) {
84	79	c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
85	80	c->put_pixels_clamped = ff_put_pixels_clamped_sse2;
86	81	c->add_pixels_clamped = ff_add_pixels_clamped_sse2;
87	82
	83	+#if ARCH_X86_32
88	84	if (!high_bit_depth &&
89	85	avctx->lowres == 0 &&
90	86	(avctx->idct_algo == FF_IDCT_AUTO \|\|
@@ -94,6 +90,7 @@ av_cold void ff_idctdsp_init_x86(IDCTDSPContext c, AVCodecContext avctx, · 3 unmodified lines
94	90	c->idct_add = ff_simple_idct_add_sse2;
95	91	c->perm_type = FF_IDCT_PERM_SIMPLE;
96	92	}
	93	+#endif
97	94
98	95	if (ARCH_X86_64 &&
99	96	!high_bit_depth &&

libavcodec/x86/simple_idct.asm+2−20

@@ -25,6 +25,7 @@ · 24 unmodified lines
25	25
26	26	%include "libavutil/x86/x86util.asm"
27	27
	28	+%if ARCH_X86_32
28	29	SECTION_RODATA
29	30
30	31	cextern pb_80
@@ -846,26 +847,6 @@ cglobal simple_idct, 1, 2, 8, 128, block, t0 · 815 unmodified lines
846	847	IDCT
847	848	RET
848	849
849		−cglobal simple_idct_put, 3, 5, 8, 128, pixels, lsize, block, lsize3, t0
850		− IDCT
851		− lea lsize3q, [lsizeq*3]
852		− PUT_PIXELS_CLAMPED_HALF 0
853		− lea pixelsq, [pixelsq+lsizeq*4]
854		− PUT_PIXELS_CLAMPED_HALF 64
855		−RET
856		−
857		−cglobal simple_idct_add, 3, 4, 8, 128, pixels, lsize, block, t0
858		− IDCT
859		− pxor m4, m4
860		− ADD_PIXELS_CLAMPED 0
861		− lea pixelsq, [pixelsq+lsizeq*2]
862		− ADD_PIXELS_CLAMPED 32
863		− lea pixelsq, [pixelsq+lsizeq*2]
864		− ADD_PIXELS_CLAMPED 64
865		− lea pixelsq, [pixelsq+lsizeq*2]
866		− ADD_PIXELS_CLAMPED 96
867		−RET
868		−
869	850	INIT_XMM sse2
870	851
871	852	cglobal simple_idct_put, 3, 5, 8, 128, pixels, lsize, block, lsize3, t0
@@ -887,3 +868,4 @@ cglobal simple_idct_add, 3, 4, 8, 128, pixels, lsize, block, t0 · 15 unmodified lines
887	868	lea pixelsq, [pixelsq+lsizeq*2]
888	869	ADD_PIXELS_CLAMPED 96
889	870	RET
	871	+%endif

avcodec/x86/idctdsp: Remove obsolete MMX(EXT) functions

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code