API tools faq
paste
Advertisement
Kitomas

kmixer_instrinsics.txt as of 8-18-23

Kitomas
Aug 18th, 2023
145
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 21.30 KB | None | 0 0
raw download clone embed print report
  1. ----BY EXTENSION----
  2.  
  3.  
  4.  
  5. AVX: {
  6. arithmetic: {
  7. __m256 _mm256_add_ps (__m256 a, __m256 b): add packed f32
  8. __m256 _mm256_div_ps (__m256 a, __m256 b): divide packed f32
  9. __m256 _mm256_mul_ps (__m256 a, __m256 b): multiply packed f32
  10. __m256 _mm256_sub_ps (__m256 a, __m256 b): subtract packed f32
  11.  
  12. __m256 _mm256_hadd_ps (__m256 a, __m256 b): horiz. add packed f32
  13. }
  14. convert: {
  15. __m256i _mm256_cvtps_epi32 (__m256 a): packed f32 to i32 (w/ rounding)
  16. __m256i _mm256_cvttps_epi32 (__m256 a): packed f32 to i32 (w/ truncation)
  17. __m256 _mm256_cvtepi32_ps (__m256i a): packed i32 to f32 (w/ rounding)
  18. }
  19. load: {
  20. __m256 _mm256_load_ps (float const * mem_addr): f32 (32B aligned for dst.mem_addr)
  21. __m256 _mm256_loadu_ps (float const * mem_addr): f32 (no alignment for dst.mem_addr)
  22. __m256i _mm256_load_si256 (__m256i const * mem_addr): int (32B aligned for dst.mem_addr)
  23. __m256i _mm256_loadu_si256 (__m256i const * mem_addr): int (no alignment for dst.mem_addr)
  24. }
  25. store: {
  26. void _mm256_store_ps (float * mem_addr, __m256 a): f32 (32B aligned for mem_addr)
  27. void _mm256_storeu_ps (float * mem_addr, __m256 a): f32 (no alignment for mem_addr)
  28. void _mm256_store_si256 (__m256i * mem_addr, __m256i a): int (32B aligned for mem_addr)
  29. void _mm256_storeu_si256 (__m256i * mem_addr, __m256i a): int (no alignment for mem_addr)
  30. }
  31. set: {
  32. __m256i _mm256_set1_epi8 (char a): set all elements of dst to i8 value
  33. __m256i _mm256_set1_epi16 (short a): set all elements of dst to i16 value
  34. __m256i _mm256_set1_epi32 (int a): set all elements of dst to i32 value
  35. __m256 _mm256_set1_ps (float a): set all elements of dst to f32 value
  36. }
  37. special math: {
  38. __m256 _mm256_max_ps (__m256 a, __m256 b): max() on packed f32
  39. __m256 _mm256_min_ps (__m256 a, __m256 b): min() on packed f32
  40. __m256 _mm256_round_ps (__m256 a, int rounding): round packed f32
  41. }
  42. swizzle: {
  43. __m256 _mm256_unpackhi_ps (__m256 a, __m256 b): unpack & interleave f32 from high 128b of a&b
  44.  
  45. __m256 _mm256_unpacklo_ps (__m256 a, __m256 b): unpack & interleave f32 from low 128b of a&b
  46. }
  47. }
  48. AVX2: {
  49. arithmetic: {
  50. __m256i _mm256_add_epi8 (__m256i a, __m256i b): add packed i8
  51. __m256i _mm256_add_epi16 (__m256i a, __m256i b): add packed i16
  52. __m256i _mm256_add_epi32 (__m256i a, __m256i b): add packed i32
  53.  
  54. __m256i _mm256_adds_epi8 (__m256i a, __m256i b): add packed i8 w/ saturation
  55. __m256i _mm256_adds_epi16 (__m256i a, __m256i b): add packed i16 w/ saturation
  56. __m256i _mm256_adds_epu8 (__m256i a, __m256i b): add packed u8 w/ saturation
  57. __m256i _mm256_adds_epu16 (__m256i a, __m256i b): add packed u16 w/ saturation
  58.  
  59. __m256i _mm256_sub_epi8 (__m256i a, __m256i b): subtract packed i8
  60. __m256i _mm256_sub_epi16 (__m256i a, __m256i b): subtract packed i16
  61. __m256i _mm256_sub_epi32 (__m256i a, __m256i b): subtract packed i32
  62.  
  63. __m256i _mm256_subs_epi8 (__m256i a, __m256i b): subtract packed i8 w/ saturation
  64. __m256i _mm256_subs_epi16 (__m256i a, __m256i b): subtract packed i16 w/ saturation
  65. __m256i _mm256_subs_epu8 (__m256i a, __m256i b): subtract packed u8 w/ saturation
  66. __m256i _mm256_subs_epu16 (__m256i a, __m256i b): subtract packed u16 w/ saturation
  67.  
  68. __m256i _mm256_hadd_epi16 (__m256i a, __m256i b): horiz. add packed i16 pairs
  69. __m256i _mm256_hadds_epi16 (__m256i a, __m256i b): horiz. add packed i16 pairs w/ saturation
  70. __m256i _mm256_hadd_epi32 (__m256i a, __m256i b): horiz. add packed i32 pairs
  71. }
  72. convert: {
  73. __m256i _mm256_cvtepi8_epi16 (__m128i a): sign-extend packed i8 to i16
  74. __m256i _mm256_cvtepi16_epi32 (__m128i a): sign-extend packed i16 to i32
  75.  
  76. __m256i _mm256_cvtepi8_epi32 (__m128i a): zero-extend packed i8 to i32
  77. __m256i _mm256_cvtepu8_epi16 (__m128i a): zero-extend packed u8 to i16
  78. __m256i _mm256_cvtepu8_epi32 (__m128i a): zero-extend packed u8 to i32
  79. __m256i _mm256_cvtepu16_epi32 (__m128i a): zero-extend packed u16 to i32
  80. }
  81. prob/stat: {
  82. __m256i _mm256_avg_epu8 (__m256i a, __m256i b): average packed u8
  83. __m256i _mm256_avg_epu16 (__m256i a, __m256i b): average packed u16
  84. }
  85. special math: {
  86. __m256i _mm256_max_epi8 (__m256i a, __m256i b): max() on packed i8
  87. __m256i _mm256_max_epi16 (__m256i a, __m256i b): max() on packed i16
  88. __m256i _mm256_max_epi32 (__m256i a, __m256i b): max() on packed i32
  89. __m256i _mm256_max_epu8 (__m256i a, __m256i b): max() on packed u8
  90. __m256i _mm256_max_epu16 (__m256i a, __m256i b): max() on packed u16
  91. __m256i _mm256_max_epu32 (__m256i a, __m256i b): max() on packed u32
  92.  
  93. __m256i _mm256_min_epi8 (__m256i a, __m256i b): min() on packed i8
  94. __m256i _mm256_min_epi16 (__m256i a, __m256i b): min() on packed i16
  95. __m256i _mm256_min_epi32 (__m256i a, __m256i b): min() on packed i32
  96. __m256i _mm256_min_epu8 (__m256i a, __m256i b): min() on packed u8
  97. __m256i _mm256_min_epu16 (__m256i a, __m256i b): min() on packed u16
  98. __m256i _mm256_min_epu32 (__m256i a, __m256i b): min() on packed u32
  99. }
  100. misc: {
  101. __m256i _mm256_packs_epi32 (__m256i a, __m256i b): packed i32 to i16 (w/ signed saturation)
  102. __m256i _mm256_packs_epi16 (__m256i a, __m256i b): packed i16 to i8 (w/ signed saturation)
  103. __m256i _mm256_packus_epi32 (__m256i a, __m256i b): packed i32 to i16 (w/ unsigned saturation)
  104. __m256i _mm256_packus_epi16 (__m256i a, __m256i b): packed i16 to i8 (w/ unsigned saturation)
  105. }
  106. swizzle: {
  107. __m256i _mm256_unpackhi_epi8 (__m256i a, __m256i b): unpack and interleave i8 from high 128b of a&b
  108. __m256i _mm256_unpackhi_epi16 (__m256i a, __m256i b): unpack and interleave i16 from high 128b of a&b
  109. __m256i _mm256_unpackhi_epi32 (__m256i a, __m256i b): unpack and interleave i32 from high 128b of a&b
  110.  
  111. __m256i _mm256_unpacklo_epi8 (__m256i a, __m256i b): unpack and interleave i8 from low 128b of a&b
  112. __m256i _mm256_unpacklo_epi16 (__m256i a, __m256i b): unpack and interleave i16 from low 128b of a&b
  113. __m256i _mm256_unpacklo_epi32 (__m256i a, __m256i b): unpack and interleave i32 from low 128b of a&b
  114. }
  115. }
  116.  
  117. SSE: {
  118. arithmetic: {
  119. __m128 _mm_add_ps (__m128 a, __m128 b): add packed f32
  120. __m128 _mm_div_ps (__m128 a, __m128 b): divide packed f32
  121. __m128 _mm_mul_ps (__m128 a, __m128 b): multiply packed f32
  122. __m128 _mm_sub_ps (__m128 a, __m128 b): subtract packed f32
  123. }
  124. convert: {
  125. __m128 _mm_cvtpi16_ps (__m64 a): packed i16 to f32
  126. __m128 _mm_cvtpu16_ps (__m64 a): packed u16 to f32
  127. __m64 _mm_cvttps_pi32 (__m128 a): packed f32 to i32 (w/ truncation)
  128. __m64 _mm_cvtt_ps2pi (__m128 a): alias of _mm_cvttps_pi32
  129. }
  130. load: {
  131. __m128 _mm_load_ps1 (float const* mem_addr): basically an f32 set1 but from memory
  132. __m128 _mm_load1_ps (float const* mem_addr): alias of _mm_load1_ps
  133. __m128 _mm_load_ps (float const* mem_addr): f32 (16B alignment for dst.mem_addr)
  134. __m128 _mm_loadu_ps (float const* mem_addr): f32 (no alignment for dst.mem_addr)
  135. }
  136. store: {
  137. void _mm_store_ps (float* mem_addr, __m128 a): f32 (16B alignment for mem_addr)
  138. void _mm_storeu_ps (float* mem_addr, __m128 a): f32 (no alignment for mem_addr)
  139. }
  140. move: {
  141. __m128 _mm_movehl_ps (__m128 a, __m128 b): f32; b's hi 2 to low dst, a's hi 2 to high dst
  142. }
  143. prob/stat: {
  144. __m64 _mm_avg_pu8 (__m64 a, __m64 b): average packed u8
  145. __m64 _mm_avg_pu16 (__m64 a, __m64 b): average packed u16
  146. __m64 _m_pavgb (__m64 a, __m64 b): average packed u8 (alias of _mm_avg_pu8)
  147. __m64 _m_pavgw (__m64 a, __m64 b): average packed u16 (alias of _mm_avg_pu16)
  148. }
  149. set: {
  150. __m128 _mm_set1_ps (float a): set all elements of dst to f32
  151. }
  152. special math: {
  153. __m64 _mm_max_pu8 (__m64 a, __m64 b): max() packed u8
  154. __m64 _mm_max_pi16 (__m64 a, __m64 b): max() packed i16
  155. __m128 _mm_max_ps (__m128 a, __m128 b): max() packed f32
  156.  
  157. __m64 _mm_min_pu8 (__m64 a, __m64 b): min() packed u8
  158. __m64 _mm_min_pi16 (__m64 a, __m64 b): min() packed i16
  159. __m128 _mm_min_ps (__m128 a, __m128 b): min() packed f32
  160.  
  161. __m64 _m_pmaxub (__m64 a, __m64 b): max() packed u8 (alias of _mm_max_pu8)
  162. __m64 _m_pmaxsw (__m64 a, __m64 b): max() packed i16 (alias of _mm_max_pi16)
  163.  
  164. __m64 _m_pminub (__m64 a, __m64 b): min() packed u8 (alias of _mm_min_pu8)
  165. __m64 _m_pminsw (__m64 a, __m64 b): min() packed i16 (alias of _mm_min_pi16)
  166. }
  167. swizzle: {
  168. __m128 _mm_unpackhi_ps (__m128 a, __m128 b): unpack & interleave f32 from high 64b of a&b
  169.  
  170. __m128 _mm_unpacklo_ps (__m128 a, __m128 b): unpack & interleave f32 from low 64b of a&b
  171.  
  172. __m128 _mm_shuffle_ps (__m128 a, __m128 b, unsigned int imm8): shuffle f32 using control
  173. }
  174. }
  175. SSE2: {
  176. arithmetic: {
  177. __m128i _mm_add_epi8 (__m128i a, __m128i b): add packed i8
  178. __m128i _mm_add_epi16 (__m128i a, __m128i b): add packed i16
  179. __m128i _mm_add_epi32 (__m128i a, __m128i b): add packed i32
  180. __m128i _mm_adds_epi8 (__m128i a, __m128i b): add packed i8 (w/ saturation)
  181. __m128i _mm_adds_epi16 (__m128i a, __m128i b): add packed i16 (w/ saturation)
  182. __m128i _mm_adds_epu8 (__m128i a, __m128i b): add packed u8 (w/ saturation)
  183. __m128i _mm_adds_epu16 (__m128i a, __m128i b): add packed u16 (w/ saturation)
  184.  
  185. __m128i _mm_sub_epi8 (__m128i a, __m128i b): subtract packed i8
  186. __m128i _mm_sub_epi16 (__m128i a, __m128i b): subtract packed i16
  187. __m128i _mm_sub_epi32 (__m128i a, __m128i b): subtract packed i32
  188. __m128i _mm_subs_epi8 (__m128i a, __m128i b): subtract packed i8 (w/ saturation)
  189. __m128i _mm_subs_epi16 (__m128i a, __m128i b): subtract packed i16 (w/ saturation)
  190. __m128i _mm_subs_epu8 (__m128i a, __m128i b): subtract packed u8 (w/ saturation)
  191. __m128i _mm_subs_epu16 (__m128i a, __m128i b): subtract packed u16 (w/ saturation)
  192. }
  193. convert: {
  194. __m128 _mm_cvtepi32_ps (__m128i a): packed i32 to f32
  195. __m128i _mm_cvtps_epi32 (__m128 a): packed f32 to i32 (w/ rounding)
  196. __m128i _mm_cvttps_epi32 (__m128 a): packed f32 to i32 (w/ truncation)
  197. }
  198. load: {
  199. __m128i _mm_load_si128 (__m128i const* mem_addr): 128-bits int (16B aligned for dst.mem_addr)
  200. __m128i _mm_loadu_si128 (__m128i const* mem_addr): 128-bits int (no alignment for dst.mem_addr)
  201. }
  202. store: {
  203. void _mm_store_si128 (__m128i* mem_addr, __m128i a): 128-bits int (16B aligned for mem_addr)
  204. void _mm_storeu_si128 (__m128i* mem_addr, __m128i a): 128-bits int (no alignment for mem_addr)
  205. }
  206. prob/stat: {
  207. __m128i _mm_avg_epu8 (__m128i a, __m128i b): average packed u8
  208. __m128i _mm_avg_epu16 (__m128i a, __m128i b): average packed u16
  209. }
  210. set: {
  211. __m128i _mm_set1_epi8 (char a): set all elements of dst to i8
  212. __m128i _mm_set1_epi16 (short a): set all elements of dst to i16
  213. __m128i _mm_set1_epi32 (int a): set all elements of dst to i32
  214. }
  215. special math: {
  216. __m128i _mm_max_epu8 (__m128i a, __m128i b): max() packed u8
  217. __m128i _mm_max_epi16 (__m128i a, __m128i b): max() packed i16
  218.  
  219. __m128i _mm_min_epu8 (__m128i a, __m128i b): min() packed u8
  220. __m128i _mm_min_epi16 (__m128i a, __m128i b): min() packed i16
  221. }
  222. swizzle: {
  223. __m128i _mm_unpackhi_epi8 (__m128i a, __m128i b): unpack & interleave i8 from high 64b of a&b
  224. __m128i _mm_unpackhi_epi16 (__m128i a, __m128i b): unpack & interleave i16 from high 64b of a&b
  225. __m128i _mm_unpackhi_epi32 (__m128i a, __m128i b): unpack & interleave i32 from high 64b of a&b
  226.  
  227. __m128i _mm_unpacklo_epi8 (__m128i a, __m128i b): unpack & interleave i8 from low 64b of a&b
  228. __m128i _mm_unpacklo_epi16 (__m128i a, __m128i b): unpack & interleave i16 from low 64b of a&b
  229. __m128i _mm_unpacklo_epi32 (__m128i a, __m128i b): unpack & interleave i32 from low 64b of a&b
  230. }
  231. }
  232. SSE3: {
  233. arithmetic: {
  234. __m128 _mm_hadd_ps (__m128 a, __m128 b): horiz. add packed f32 pairs
  235. }
  236. }
  237. SSE41: {
  238. convert: {
  239. __m128i _mm_cvtepi8_epi16 (__m128i a): sign-extend packed i8 to i16
  240. __m128i _mm_cvtepi8_epi32 (__m128i a): sign-extend packed i8 to i32
  241. __m128i _mm_cvtepi16_epi32 (__m128i a): sign-extend packed i16 to i32
  242.  
  243. __m128i _mm_cvtepu8_epi16 (__m128i a): zero-extend packed u8 to i16
  244. __m128i _mm_cvtepu8_epi32 (__m128i a): zero-extend packed u8 to i32
  245. __m128i _mm_cvtepu16_epi32 (__m128i a): zero-extend packed u16 to i32
  246. }
  247. special math: {
  248. __m128i _mm_max_epi8 (__m128i a, __m128i b): max() packed i8
  249. __m128i _mm_max_epi32 (__m128i a, __m128i b): max() packed i32
  250. __m128i _mm_max_epu16 (__m128i a, __m128i b): max() packed u16
  251. __m128i _mm_max_epu32 (__m128i a, __m128i b): max() packed u32
  252.  
  253. __m128i _mm_min_epi8 (__m128i a, __m128i b): min() packed i8
  254. __m128i _mm_min_epi32 (__m128i a, __m128i b): min() packed i32
  255. __m128i _mm_min_epu16 (__m128i a, __m128i b): min() packed u16
  256. __m128i _mm_min_epu32 (__m128i a, __m128i b): min() packed u32
  257.  
  258. __m128 _mm_round_ps (__m128 a, int rounding): round packed f32
  259. }
  260. }
  261.  
  262.  
  263.  
  264. ----BY OPERATION----
  265.  
  266.  
  267.  
  268. add: {
  269. SSE, __m128 ->__m128 : f32 (_mm_add_ps)
  270. SSE2, __m128i->__m128i: i8 (_mm_add_epi8)
  271. SSE2, __m128i->__m128i: i16 (_mm_add_epi16)
  272. SSE2, __m128i->__m128i: i32 (_mm_add_epi32)
  273. AVX, __m256 ->__m256 : f32 (_mm256_add_ps)
  274. AVX2, __m256i->__m256i: i8 (_mm256_add_epi8)
  275. AVX2, __m256i->__m256i: i16 (_mm256_add_epi16)
  276. AVX2, __m256i->__m256i: i32 (_mm256_add_epi32)
  277. }
  278.  
  279. divide: {
  280. SSE, __m128 ->__m128 : f32 (_mm_div_ps)
  281. AVX, __m256->__m256: f32 (_mm256_div_ps)
  282. }
  283.  
  284. multiply: {
  285. SSE, __m128 ->__m128 : f32 (_mm_mul_ps)
  286. AVX, __m256->__m256: f32 (_mm256_mul_ps)
  287. }
  288.  
  289. subtract: {
  290. SSE, __m128 ->__m128 : f32 (_mm_sub_ps)
  291. SSE2, __m128i->__m128i: i8 (_mm_sub_epi8)
  292. SSE2, __m128i->__m128i: i16 (_mm_sub_epi16)
  293. SSE2, __m128i->__m128i: i32 (_mm_sub_epi32)
  294. AVX, __m256 ->__m256 : f32 (_mm256_sub_ps)
  295. AVX2, __m256i->__m256i: i8 (_mm256_sub_epi8)
  296. AVX2, __m256i->__m256i: i16 (_mm256_sub_epi16)
  297. AVX2, __m256i->__m256i: i32 (_mm256_sub_epi32)
  298. }
  299.  
  300.  
  301. add w/ saturation: {
  302. SSE2, __m128i->__m128i: i8 (_mm_adds_epi8)
  303. SSE2, __m128i->__m128i: i16 (_mm_adds_epi16)
  304. SSE2, __m128i->__m128i: u8 (_mm_adds_epu8)
  305. SSE2, __m128i->__m128i: u16 (_mm_adds_epu16)
  306. AVX2, __m256i->__m256i: i8 (_mm256_adds_epi8)
  307. AVX2, __m256i->__m256i: i16 (_mm256_adds_epi16)
  308. AVX2, __m256i->__m256i: u8 (_mm256_adds_epu8)
  309. AVX2, __m256i->__m256i: u16 (_mm256_adds_epu16)
  310. }
  311.  
  312. subtract w/ saturation: {
  313. SSE2, __m128i->__m128i: i8 (_mm_subs_epi8)
  314. SSE2, __m128i->__m128i: i16 (_mm_subs_epi16)
  315. SSE2, __m128i->__m128i: u8 (_mm_subs_epu8)
  316. SSE2, __m128i->__m128i: u16 (_mm_subs_epu16)
  317. AVX2, __m256i->__m256i: i8 (_mm256_subs_epi8)
  318. AVX2, __m256i->__m256i: i16 (_mm256_subs_epi16)
  319. AVX2, __m256i->__m256i: u8 (_mm256_subs_epu8)
  320. AVX2, __m256i->__m256i: u16 (_mm256_subs_epu16)
  321. }
  322.  
  323. horiz. add: {
  324. SSE3, __m128 ->__m128 : f32 (_mm_hadd_ps)
  325. AVX, __m256 ->__m256 : f32 (_mm256_hadd_ps)
  326. AVX2, __m256i->__m256i: i16 (_mm256_hadd_epi16)
  327. AVX2, __m256i->__m256i: i32 (_mm256_hadd_epi32)
  328. }
  329.  
  330. horiz. add w/ saturation: {
  331. AVX2, __m256i->__m256i: i16 (_mm256_hadds_epi16)
  332. }
  333.  
  334.  
  335. convert: {
  336. rounding or n/a: {
  337. SSE, __m64 ->__m128 : i16 to f32 (_mm_cvtpi16_ps)
  338. SSE, __m64 ->__m128 : u16 to f32 (_mm_cvtpu16_ps)
  339. SSE2, __m128i->__m128 : i32 to f32 (_mm_cvtepi32_ps)
  340. SSE2, __m128 ->__m128i: f32 to i32 (_mm_cvtps_epi32)
  341. AVX, __m256 ->__m256i: f32 to i32 (_mm256_cvtps_epi32)
  342. AVX, __m256i->__m256 : i32 to f32 (_mm256_cvtepi32_ps)
  343. }
  344. truncation: {
  345. SSE, __m128->__m64 : f32 to i32 (_mm_cvttps_pi32)
  346. SSE2, __m128->__m128i: f32 to i32 (_mm_cvttps_epi32)
  347. AVX, __m256->__m256i: f32 to i32 (_mm256_cvttps_epi32)
  348. }
  349. signed saturation: {
  350. }
  351. }
  352.  
  353.  
  354. load (alignment is 16B for m128, 32B for m256): {
  355. SSE, float*->__m128 : f32 (_mm_load_ps)
  356. SSE2, __m128i*->__m128i: int (_mm_load_si128)
  357. AVX, float*->__m256 : f32 (_mm256_load_ps)
  358. AVX, __m256i*->__m256i: int (_mm256_load_si256)
  359. }
  360.  
  361. load unaligned: {
  362. SSE, float*->__m128 : f32 (_mm_load1_ps) (acts like like set1)
  363. SSE, float*->__m128 : f32 (_mm_loadu_ps)
  364. SSE2, __m128i*->__m128i: int (_mm_loadu_si128)
  365. AVX, float*->__m256 : f32 (_mm256_loadu_ps)
  366. AVX, __m256i*->__m256i: int (_mm256_loadu_si256)
  367. }
  368.  
  369.  
  370. store (alignment is 16B for m128, 32B for m256): {
  371. SSE, __m128 ->float* : f32 (_mm_store_ps)
  372. SSE2, __m128i->__m128i*: int (_mm_store_si128)
  373. AVX, __m256 ->float* : f32 (_mm256_store_ps)
  374. AVX, __m256i->__m256i*: int (_mm256_store_si256)
  375. }
  376.  
  377. store unaligned: {
  378. SSE, __m128 ->float* : f32 (_mm_storeu_ps)
  379. SSE2, __m128i->__m128i*: int (_mm_storeu_si128)
  380. AVX, __m256 ->float* : f32 (_mm256_storeu_ps)
  381. AVX, __m256i->__m256i*: int (_mm256_storeu_si256)
  382. }
  383.  
  384.  
  385. movehl: {
  386. SSE, __m128->__m128: f32 (_mm_movehl_ps)
  387. }
  388.  
  389.  
  390. shuffle: {
  391. SSE, __m128->__m128: f32 using control (_mm_shuffle_ps)
  392. }
  393.  
  394.  
  395. set1: {
  396. SSE, float->__m128 : f32 (_mm_set1_ps)
  397. SSE2, char ->__m128i: i8 (_mm_set1_epi8)
  398. SSE2, short->__m128i: i16 (_mm_set1_epi16)
  399. SSE2, int ->__m128i: i32 (_mm_set1_epi32)
  400. AVX, char ->__m256i: i8 (_mm256_set1_epi8)
  401. AVX, short->__m256i: i16 (_mm256_set1_epi16)
  402. AVX, int ->__m256i: i32 (_mm256_set1_epi32)
  403. AVX, float->__m256 : f32 (_mm256_set1_ps)
  404. }
  405.  
  406.  
  407. unpack high: {
  408. SSE, __m128 ->__m128 : f32 (_mm_unpackhi_ps)
  409. SSE2, __m128i->__m128i: i8 (_mm_unpackhi_epi8)
  410. SSE2, __m128i->__m128i: i16 (_mm_unpackhi_epi16)
  411. SSE2, __m128i->__m128i: i32 (_mm_unpackhi_epi32)
  412. AVX, __m256 ->__m256 : f32 (_mm256_unpackhi_ps)
  413. AVX2, __m256i->__m256i: i8 (_mm256_unpackhi_epi8)
  414. AVX2, __m256i->__m256i: i16 (_mm256_unpackhi_epi16)
  415. AVX2, __m256i->__m256i: i32 (_mm256_unpackhi_epi32)
  416. }
  417.  
  418. unpack low: {
  419. SSE, __m128 ->__m128 : f32 (_mm_unpacklo_ps)
  420. SSE2, __m128i->__m128i: i8 (_mm_unpacklo_epi8)
  421. SSE2, __m128i->__m128i: i16 (_mm_unpacklo_epi16)
  422. SSE2, __m128i->__m128i: i32 (_mm_unpacklo_epi32)
  423. AVX, __m256 ->__m256 : f32 (_mm256_unpacklo_ps)
  424. AVX2, __m256i->__m256i: i8 (_mm256_unpacklo_epi8)
  425. AVX2, __m256i->__m256i: i16 (_mm256_unpacklo_epi16)
  426. AVX2, __m256i->__m256i: i32 (_mm256_unpacklo_epi32)
  427. }
  428.  
  429. pack w/ signed saturation: {
  430. AVX2, __m256i->__m256i: i32 to i16 (_mm256_packs_epi32)
  431. AVX2, __m256i->__m256i: i16 to i8 (_mm256_packs_epi16)
  432. }
  433.  
  434. pack w/ unsigned saturation: {
  435. AVX2, __m256i->__m256i: i32 to i16 (_mm256_packus_epi32)
  436. AVX2, __m256i->__m256i: i16 to i8 (_mm256_packus_epi16)
  437. }
  438.  
  439.  
  440. sign-extend: {
  441. SSE41, __m128i->__m128i: i8 to i16 (_mm_cvtepi8_epi16)
  442. SSE41, __m128i->__m128i: i8 to i32 (_mm_cvtepi8_epi32)
  443. SSE41, __m128i->__m128i: i16 to i32 (_mm_cvtepi16_epi32)
  444. AVX2, __m128i->__m256i: i8 to i16 (_mm256_cvtepi8_epi16)
  445. AVX2, __m128i->__m256i: i16 to i32 (_mm256_cvtepi16_epi32)
  446. }
  447.  
  448. zero-extend: {
  449. SSE41, __m128i->__m128i: u8 to i16 (_mm_cvtepu8_epi16)
  450. SSE41, __m128i->__m128i: u8 to i32 (_mm_cvtepu8_epi32)
  451. SSE41, __m128i->__m128i: u16 to i32 (_mm_cvtepu16_epi32)
  452. AVX2, __m128i->__m256i: i8 to i32 (_mm256_cvtepi8_epi32)
  453. AVX2, __m128i->__m256i: u8 to i16 (_mm256_cvtepu8_epi16)
  454. AVX2, __m128i->__m256i: u8 to i32 (_mm256_cvtepu8_epi32)
  455. AVX2, __m128i->__m256i: u16 to i32 (_mm256_cvtepu16_epi32)
  456. }
  457.  
  458.  
  459. average: {
  460. SSE, __m64 ->__m64 : u8 (_mm_avg_pu8)
  461. SSE, __m64 ->__m64 : u16 (_mm_avg_pu16)
  462. SSE2, __m128i->__m128i: u8 (_mm_avg_epu8)
  463. SSE2, __m128i->__m128i: u16 (_mm_avg_epu16)
  464. AVX2, __m256i->__m256i: u8 (_mm256_avg_epu8)
  465. AVX2, __m256i->__m256i: u16 (_mm256_avg_epu16)
  466. }
  467.  
  468. max: {
  469. SSE, __m64 ->__m64 : u8 (_mm_max_pu8)
  470. SSE, __m64 ->__m64 : i16 (_mm_max_pi16)
  471. SSE, __m128 ->__m128 : f32 (_mm_max_ps)
  472. SSE2, __m128i->__m128i: u8 (_mm_max_epu8)
  473. SSE2, __m128i->__m128i: i16 (_mm_max_epi16)
  474. SSE41, __m128i->__m128i: i8 (_mm_max_epi8)
  475. SSE41, __m128i->__m128i: i32 (_mm_max_epi32)
  476. SSE41, __m128i->__m128i: u16 (_mm_max_epu16)
  477. SSE41, __m128i->__m128i: u32 (_mm_max_epu32)
  478. AVX, __m256 ->__m256 : f32 (_mm256_max_ps)
  479. AVX2, __m256i->__m256i: i8 (_mm256_max_epi8)
  480. AVX2, __m256i->__m256i: i16 (_mm256_max_epi16)
  481. AVX2, __m256i->__m256i: i32 (_mm256_max_epi32)
  482. AVX2, __m256i->__m256i: u8 (_mm256_max_epu8)
  483. AVX2, __m256i->__m256i: u16 (_mm256_max_epu16)
  484. AVX2, __m256i->__m256i: u32 (_mm256_max_epu32)
  485. }
  486.  
  487. min: {
  488. SSE, __m64 ->__m64 : u8 (_mm_min_pu8)
  489. SSE, __m64 ->__m64 : i16 (_mm_min_pi16)
  490. SSE, __m128 ->__m128 : f32 (_mm_min_ps)
  491. SSE2, __m128i->__m128i: u8 (_mm_min_epu8)
  492. SSE2, __m128i->__m128i: i16 (_mm_min_epi16)
  493. SSE41, __m128i->__m128i: i8 (_mm_min_epi8)
  494. SSE41, __m128i->__m128i: i32 (_mm_min_epi32)
  495. SSE41, __m128i->__m128i: u16 (_mm_min_epu16)
  496. SSE41, __m128i->__m128i: u32 (_mm_min_epu32)
  497. AVX, __m256 ->__m256 : f32 (_mm256_min_ps)
  498. AVX2, __m256i->__m256i: i8 (_mm256_min_epi8)
  499. AVX2, __m256i->__m256i: i16 (_mm256_min_epi16)
  500. AVX2, __m256i->__m256i: i32 (_mm256_min_epi32)
  501. AVX2, __m256i->__m256i: u8 (_mm256_min_epu8)
  502. AVX2, __m256i->__m256i: u16 (_mm256_min_epu16)
  503. AVX2, __m256i->__m256i: u32 (_mm256_min_epu32)
  504. }
  505.  
  506. round: {
  507. SSE41, __m128->__m128: f32 (_mm_round_ps)
  508. AVX, __m256->__m256: f32 (_mm256_round_ps)
  509. }
  510.  
  511. (reminder: allow mono streams to have stereo volume applied to them)
Advertisement
Add Comment
Please, Sign In to add comment
Advertisement
Public Pastes
Advertisement
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!