Mercurial > libavcodec.hg
comparison i386/vp3dsp_sse2.c @ 7883:6f77eda89813 libavcodec
Make a separate macro for writing an entire block from registers to memory
author | conrad |
---|---|
date | Thu, 18 Sep 2008 03:08:13 +0000 |
parents | 1a684e0e6420 |
children | 4077df298ba2 |
comparison
equal
deleted
inserted
replaced
7882:1a684e0e6420 | 7883:6f77eda89813 |
---|---|
119 SHIFT(%%xmm2) /* xmm2 = op2 */ \ | 119 SHIFT(%%xmm2) /* xmm2 = op2 */ \ |
120 "psubsw %%xmm7, %%xmm4 \n\t" /* xmm4 = E - G = E. */ \ | 120 "psubsw %%xmm7, %%xmm4 \n\t" /* xmm4 = E - G = E. */ \ |
121 SHIFT(%%xmm1) /* xmm1 = op1 */ \ | 121 SHIFT(%%xmm1) /* xmm1 = op1 */ \ |
122 "movdqa "I(2)", %%xmm3 \n\t" /* Load D. from I(2) */ \ | 122 "movdqa "I(2)", %%xmm3 \n\t" /* Load D. from I(2) */ \ |
123 "paddsw %%xmm7, %%xmm7 \n\t" /* xmm7 = G + G */ \ | 123 "paddsw %%xmm7, %%xmm7 \n\t" /* xmm7 = G + G */ \ |
124 "movdqa %%xmm2, "O(2)" \n\t" /* Write out op2 */ \ | |
125 "paddsw %%xmm4, %%xmm7 \n\t" /* xmm7 = E + G = G. */ \ | 124 "paddsw %%xmm4, %%xmm7 \n\t" /* xmm7 = E + G = G. */ \ |
126 "movdqa %%xmm1, "O(1)" \n\t" /* Write out op1 */ \ | |
127 "psubsw %%xmm3, %%xmm4 \n\t" /* xmm4 = E. - D. = R4 */ \ | 125 "psubsw %%xmm3, %%xmm4 \n\t" /* xmm4 = E. - D. = R4 */ \ |
128 ADD(%%xmm4) /* Adjust R4 and R3 before shifting */ \ | 126 ADD(%%xmm4) /* Adjust R4 and R3 before shifting */ \ |
129 "paddsw %%xmm3, %%xmm3 \n\t" /* xmm3 = D. + D. */ \ | 127 "paddsw %%xmm3, %%xmm3 \n\t" /* xmm3 = D. + D. */ \ |
130 "paddsw %%xmm4, %%xmm3 \n\t" /* xmm3 = E. + D. = R3 */ \ | 128 "paddsw %%xmm4, %%xmm3 \n\t" /* xmm3 = E. + D. = R3 */ \ |
131 SHIFT(%%xmm4) /* xmm4 = op4 */ \ | 129 SHIFT(%%xmm4) /* xmm4 = op4 */ \ |
133 SHIFT(%%xmm3) /* xmm3 = op3 */ \ | 131 SHIFT(%%xmm3) /* xmm3 = op3 */ \ |
134 ADD(%%xmm6) /* Adjust R6 and R5 before shifting */ \ | 132 ADD(%%xmm6) /* Adjust R6 and R5 before shifting */ \ |
135 "paddsw %%xmm5, %%xmm5 \n\t" /* xmm5 = B.. + B.. */ \ | 133 "paddsw %%xmm5, %%xmm5 \n\t" /* xmm5 = B.. + B.. */ \ |
136 "paddsw %%xmm6, %%xmm5 \n\t" /* xmm5 = F. + B.. = R5 */ \ | 134 "paddsw %%xmm6, %%xmm5 \n\t" /* xmm5 = F. + B.. = R5 */ \ |
137 SHIFT(%%xmm6) /* xmm6 = op6 */ \ | 135 SHIFT(%%xmm6) /* xmm6 = op6 */ \ |
138 "movdqa %%xmm4, "O(4)" \n\t" /* Write out op4 */ \ | |
139 SHIFT(%%xmm5) /* xmm5 = op5 */ \ | 136 SHIFT(%%xmm5) /* xmm5 = op5 */ \ |
140 "movdqa %%xmm3, "O(3)" \n\t" /* Write out op3 */ \ | |
141 "psubsw %%xmm0, %%xmm7 \n\t" /* xmm7 = G. - C. = R7 */ \ | 137 "psubsw %%xmm0, %%xmm7 \n\t" /* xmm7 = G. - C. = R7 */ \ |
142 ADD(%%xmm7) /* Adjust R7 and R0 before shifting */ \ | 138 ADD(%%xmm7) /* Adjust R7 and R0 before shifting */ \ |
143 "paddsw %%xmm0, %%xmm0 \n\t" /* xmm0 = C. + C. */ \ | 139 "paddsw %%xmm0, %%xmm0 \n\t" /* xmm0 = C. + C. */ \ |
144 "paddsw %%xmm7, %%xmm0 \n\t" /* xmm0 = G. + C. */ \ | 140 "paddsw %%xmm7, %%xmm0 \n\t" /* xmm0 = G. + C. */ \ |
145 SHIFT(%%xmm7) /* xmm7 = op7 */ \ | 141 SHIFT(%%xmm7) /* xmm7 = op7 */ \ |
146 "movdqa %%xmm6, "O(6)" \n\t" /* Write out op6 */ \ | 142 SHIFT(%%xmm0) /* xmm0 = op0 */ |
147 SHIFT(%%xmm0) /* xmm0 = op0 */ \ | 143 |
148 "movdqa %%xmm5, "O(5)" \n\t" /* Write out op5 */ \ | 144 #define PUT_BLOCK(r0, r1, r2, r3, r4, r5, r6, r7) \ |
149 "movdqa %%xmm7, "O(7)" \n\t" /* Write out op7 */ \ | 145 "movdqa " #r0 ", " O(0) "\n\t" \ |
150 "movdqa %%xmm0, "O(0)" \n\t" /* Write out op0 */ | 146 "movdqa " #r1 ", " O(1) "\n\t" \ |
151 | 147 "movdqa " #r2 ", " O(2) "\n\t" \ |
148 "movdqa " #r3 ", " O(3) "\n\t" \ | |
149 "movdqa " #r4 ", " O(4) "\n\t" \ | |
150 "movdqa " #r5 ", " O(5) "\n\t" \ | |
151 "movdqa " #r6 ", " O(6) "\n\t" \ | |
152 "movdqa " #r7 ", " O(7) "\n\t" | |
152 | 153 |
153 #define SSE2_Transpose() \ | 154 #define SSE2_Transpose() \ |
154 "movdqa "I(4)", %%xmm4 \n\t" /* xmm4=e7e6e5e4e3e2e1e0 */ \ | 155 "movdqa "I(4)", %%xmm4 \n\t" /* xmm4=e7e6e5e4e3e2e1e0 */ \ |
155 "movdqa "I(5)", %%xmm0 \n\t" /* xmm4=f7f6f5f4f3f2f1f0 */ \ | 156 "movdqa "I(5)", %%xmm0 \n\t" /* xmm4=f7f6f5f4f3f2f1f0 */ \ |
156 "movdqa %%xmm4, %%xmm5 \n\t" /* make a copy */ \ | 157 "movdqa %%xmm4, %%xmm5 \n\t" /* make a copy */ \ |
216 #define O(x) I(x) | 217 #define O(x) I(x) |
217 #define C(x) AV_STRINGIFY(16*(x-1))"(%1)" | 218 #define C(x) AV_STRINGIFY(16*(x-1))"(%1)" |
218 | 219 |
219 asm volatile ( | 220 asm volatile ( |
220 VP3_1D_IDCT_SSE2(NOP, NOP) | 221 VP3_1D_IDCT_SSE2(NOP, NOP) |
222 PUT_BLOCK(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7) | |
221 | 223 |
222 SSE2_Transpose() | 224 SSE2_Transpose() |
223 | 225 |
224 VP3_1D_IDCT_SSE2(ADD8, SHIFT4) | 226 VP3_1D_IDCT_SSE2(ADD8, SHIFT4) |
227 PUT_BLOCK(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm6, %%xmm7) | |
225 :: "r"(input_data), "r"(ff_vp3_idct_data), "m"(ff_pw_8) | 228 :: "r"(input_data), "r"(ff_vp3_idct_data), "m"(ff_pw_8) |
226 ); | 229 ); |
227 } | 230 } |
228 | 231 |
229 void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block) | 232 void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block) |