cast.ll 8.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254
  1. ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=knl | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX512
  2. ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=core-avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX2
  3. ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-AVX
  4. target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
  5. target triple = "x86_64-apple-macosx10.8.0"
  6. define i32 @add(i32 %arg) {
  7. ; CHECK-LABEL: for function 'add'
  8. ; -- Same size registeres --
  9. ;CHECK: cost of 1 {{.*}} zext
  10. %A = zext <4 x i1> undef to <4 x i32>
  11. ;CHECK: cost of 2 {{.*}} sext
  12. %B = sext <4 x i1> undef to <4 x i32>
  13. ;CHECK: cost of 0 {{.*}} trunc
  14. %C = trunc <4 x i32> undef to <4 x i1>
  15. ; -- Different size registers --
  16. ;CHECK-NOT: cost of 1 {{.*}} zext
  17. %D = zext <8 x i1> undef to <8 x i32>
  18. ;CHECK-NOT: cost of 2 {{.*}} sext
  19. %E = sext <8 x i1> undef to <8 x i32>
  20. ;CHECK-NOT: cost of 2 {{.*}} trunc
  21. %F = trunc <8 x i32> undef to <8 x i1>
  22. ; -- scalars --
  23. ;CHECK: cost of 1 {{.*}} zext
  24. %G = zext i1 undef to i32
  25. ;CHECK: cost of 0 {{.*}} trunc
  26. %H = trunc i32 undef to i1
  27. ;CHECK: cost of 0 {{.*}} ret
  28. ret i32 undef
  29. }
  30. define i32 @zext_sext(<8 x i1> %in) {
  31. ; CHECK-AVX2-LABEL: for function 'zext_sext'
  32. ; CHECK-AVX-LABEL: for function 'zext_sext'
  33. ;CHECK-AVX2: cost of 3 {{.*}} zext
  34. ;CHECK-AVX: cost of 4 {{.*}} zext
  35. %Z = zext <8 x i1> %in to <8 x i32>
  36. ;CHECK-AVX2: cost of 3 {{.*}} sext
  37. ;CHECK-AVX: cost of 7 {{.*}} sext
  38. %S = sext <8 x i1> %in to <8 x i32>
  39. ;CHECK-AVX2: cost of 1 {{.*}} zext
  40. ;CHECK-AVX: cost of 4 {{.*}} zext
  41. %A1 = zext <16 x i8> undef to <16 x i16>
  42. ;CHECK-AVX2: cost of 1 {{.*}} sext
  43. ;CHECK-AVX: cost of 4 {{.*}} sext
  44. %A2 = sext <16 x i8> undef to <16 x i16>
  45. ;CHECK-AVX2: cost of 1 {{.*}} sext
  46. ;CHECK-AVX: cost of 4 {{.*}} sext
  47. %A = sext <8 x i16> undef to <8 x i32>
  48. ;CHECK-AVX2: cost of 1 {{.*}} zext
  49. ;CHECK-AVX: cost of 4 {{.*}} zext
  50. %B = zext <8 x i16> undef to <8 x i32>
  51. ;CHECK-AVX2: cost of 1 {{.*}} sext
  52. ;CHECK-AVX: cost of 4 {{.*}} sext
  53. %C = sext <4 x i32> undef to <4 x i64>
  54. ;CHECK-AVX2: cost of 3 {{.*}} zext
  55. ;CHECK-AVX: cost of 4 {{.*}} zext
  56. %C.v8i8.z = zext <8 x i8> undef to <8 x i32>
  57. ;CHECK-AVX2: cost of 3 {{.*}} sext
  58. ;CHECK-AVX: cost of 7 {{.*}} sext
  59. %C.v8i8.s = sext <8 x i8> undef to <8 x i32>
  60. ;CHECK-AVX2: cost of 3 {{.*}} zext
  61. ;CHECK-AVX: cost of 3 {{.*}} zext
  62. %C.v4i16.z = zext <4 x i16> undef to <4 x i64>
  63. ;CHECK-AVX2: cost of 3 {{.*}} sext
  64. ;CHECK-AVX: cost of 6 {{.*}} sext
  65. %C.v4i16.s = sext <4 x i16> undef to <4 x i64>
  66. ;CHECK-AVX2: cost of 3 {{.*}} zext
  67. ;CHECK-AVX: cost of 4 {{.*}} zext
  68. %C.v4i8.z = zext <4 x i8> undef to <4 x i64>
  69. ;CHECK-AVX2: cost of 3 {{.*}} sext
  70. ;CHECK-AVX: cost of 6 {{.*}} sext
  71. %C.v4i8.s = sext <4 x i8> undef to <4 x i64>
  72. ;CHECK-AVX2: cost of 1 {{.*}} zext
  73. ;CHECK-AVX: cost of 4 {{.*}} zext
  74. %D = zext <4 x i32> undef to <4 x i64>
  75. ;CHECK-AVX512: cost of 3 {{.*}} %D1 = zext
  76. %D1 = zext <16 x i32> undef to <16 x i64>
  77. ;CHECK-AVX512: cost of 3 {{.*}} %D2 = sext
  78. %D2 = sext <16 x i32> undef to <16 x i64>
  79. ;CHECK-AVX512: cost of 1 {{.*}} %D3 = zext
  80. %D3 = zext <16 x i16> undef to <16 x i32>
  81. ;CHECK-AVX512: cost of 1 {{.*}} %D4 = zext
  82. %D4 = zext <16 x i8> undef to <16 x i32>
  83. ;CHECK-AVX512: cost of 2 {{.*}} %D5 = zext
  84. %D5 = zext <16 x i1> undef to <16 x i32>
  85. ;CHECK-AVX2: cost of 2 {{.*}} trunc
  86. ;CHECK-AVX: cost of 4 {{.*}} trunc
  87. %E = trunc <4 x i64> undef to <4 x i32>
  88. ;CHECK-AVX2: cost of 2 {{.*}} trunc
  89. ;CHECK-AVX: cost of 5 {{.*}} trunc
  90. %F = trunc <8 x i32> undef to <8 x i16>
  91. ;CHECK-AVX2: cost of 4 {{.*}} trunc
  92. ;CHECK-AVX: cost of 4 {{.*}} trunc
  93. %F1 = trunc <16 x i16> undef to <16 x i8>
  94. ;CHECK-AVX2: cost of 2 {{.*}} trunc
  95. ;CHECK-AVX: cost of 4 {{.*}} trunc
  96. %F2 = trunc <8 x i32> undef to <8 x i8>
  97. ;CHECK-AVX2: cost of 2 {{.*}} trunc
  98. ;CHECK-AVX: cost of 4 {{.*}} trunc
  99. %F3 = trunc <4 x i64> undef to <4 x i8>
  100. ;CHECK-AVX2: cost of 4 {{.*}} trunc
  101. ;CHECK-AVX: cost of 9 {{.*}} trunc
  102. ;CHECK_AVX512: cost of 1 {{.*}} G = trunc
  103. %G = trunc <8 x i64> undef to <8 x i32>
  104. ;CHECK-AVX512: cost of 4 {{.*}} %G1 = trunc
  105. %G1 = trunc <16 x i64> undef to <16 x i32>
  106. ret i32 undef
  107. }
  108. define i32 @masks8(<8 x i1> %in) {
  109. ; CHECK-AVX2-LABEL: for function 'masks8'
  110. ; CHECK-AVX-LABEL: for function 'masks8'
  111. ;CHECK-AVX2: cost of 3 {{.*}} zext
  112. ;CHECK-AVX: cost of 4 {{.*}} zext
  113. %Z = zext <8 x i1> %in to <8 x i32>
  114. ;CHECK-AVX2: cost of 3 {{.*}} sext
  115. ;CHECK-AVX: cost of 7 {{.*}} sext
  116. %S = sext <8 x i1> %in to <8 x i32>
  117. ret i32 undef
  118. }
  119. define i32 @masks4(<4 x i1> %in) {
  120. ; CHECK-AVX2-LABEL: for function 'masks4'
  121. ; CHECK-AVX-LABEL: for function 'masks4'
  122. ;CHECK-AVX2: cost of 3 {{.*}} zext
  123. ;CHECK-AVX: cost of 4 {{.*}} zext
  124. %Z = zext <4 x i1> %in to <4 x i64>
  125. ;CHECK-AVX2: cost of 3 {{.*}} sext
  126. ;CHECK-AVX: cost of 6 {{.*}} sext
  127. %S = sext <4 x i1> %in to <4 x i64>
  128. ret i32 undef
  129. }
  130. define void @sitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
  131. ; CHECK-LABEL: for function 'sitofp4'
  132. ; CHECK: cost of 3 {{.*}} sitofp
  133. %A1 = sitofp <4 x i1> %a to <4 x float>
  134. ; CHECK: cost of 3 {{.*}} sitofp
  135. %A2 = sitofp <4 x i1> %a to <4 x double>
  136. ; CHECK: cost of 3 {{.*}} sitofp
  137. %B1 = sitofp <4 x i8> %b to <4 x float>
  138. ; CHECK: cost of 3 {{.*}} sitofp
  139. %B2 = sitofp <4 x i8> %b to <4 x double>
  140. ; CHECK: cost of 3 {{.*}} sitofp
  141. %C1 = sitofp <4 x i16> %c to <4 x float>
  142. ; CHECK: cost of 3 {{.*}} sitofp
  143. %C2 = sitofp <4 x i16> %c to <4 x double>
  144. ; CHECK: cost of 1 {{.*}} sitofp
  145. %D1 = sitofp <4 x i32> %d to <4 x float>
  146. ; CHECK: cost of 1 {{.*}} sitofp
  147. %D2 = sitofp <4 x i32> %d to <4 x double>
  148. ret void
  149. }
  150. define void @sitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
  151. ; CHECK-LABEL: for function 'sitofp8'
  152. ; CHECK: cost of 8 {{.*}} sitofp
  153. %A1 = sitofp <8 x i1> %a to <8 x float>
  154. ; CHECK: cost of 8 {{.*}} sitofp
  155. %B1 = sitofp <8 x i8> %b to <8 x float>
  156. ; CHECK: cost of 5 {{.*}} sitofp
  157. %C1 = sitofp <8 x i16> %c to <8 x float>
  158. ; CHECK: cost of 1 {{.*}} sitofp
  159. %D1 = sitofp <8 x i32> %d to <8 x float>
  160. ret void
  161. }
  162. define void @uitofp4(<4 x i1> %a, <4 x i8> %b, <4 x i16> %c, <4 x i32> %d) {
  163. ; CHECK-LABEL: for function 'uitofp4'
  164. ; CHECK: cost of 7 {{.*}} uitofp
  165. %A1 = uitofp <4 x i1> %a to <4 x float>
  166. ; CHECK: cost of 7 {{.*}} uitofp
  167. %A2 = uitofp <4 x i1> %a to <4 x double>
  168. ; CHECK: cost of 2 {{.*}} uitofp
  169. %B1 = uitofp <4 x i8> %b to <4 x float>
  170. ; CHECK: cost of 2 {{.*}} uitofp
  171. %B2 = uitofp <4 x i8> %b to <4 x double>
  172. ; CHECK: cost of 2 {{.*}} uitofp
  173. %C1 = uitofp <4 x i16> %c to <4 x float>
  174. ; CHECK: cost of 2 {{.*}} uitofp
  175. %C2 = uitofp <4 x i16> %c to <4 x double>
  176. ; CHECK: cost of 6 {{.*}} uitofp
  177. %D1 = uitofp <4 x i32> %d to <4 x float>
  178. ; CHECK: cost of 6 {{.*}} uitofp
  179. %D2 = uitofp <4 x i32> %d to <4 x double>
  180. ret void
  181. }
  182. define void @uitofp8(<8 x i1> %a, <8 x i8> %b, <8 x i16> %c, <8 x i32> %d) {
  183. ; CHECK-LABEL: for function 'uitofp8'
  184. ; CHECK: cost of 6 {{.*}} uitofp
  185. %A1 = uitofp <8 x i1> %a to <8 x float>
  186. ; CHECK: cost of 5 {{.*}} uitofp
  187. %B1 = uitofp <8 x i8> %b to <8 x float>
  188. ; CHECK: cost of 5 {{.*}} uitofp
  189. %C1 = uitofp <8 x i16> %c to <8 x float>
  190. ; CHECK-AVX2: cost of 8 {{.*}} uitofp
  191. ; CHECK-AVX512: cost of 8 {{.*}} uitofp
  192. ; CHECK-AVX: cost of 9 {{.*}} uitofp
  193. %D1 = uitofp <8 x i32> %d to <8 x float>
  194. ret void
  195. }
  196. define void @fp_conv(<8 x float> %a, <16 x float>%b) {
  197. ;CHECK-LABEL: for function 'fp_conv'
  198. ; CHECK-AVX512: cost of 1 {{.*}} fpext
  199. %A1 = fpext <8 x float> %a to <8 x double>
  200. ; CHECK-AVX512: cost of 3 {{.*}} fpext
  201. %A2 = fpext <16 x float> %b to <16 x double>
  202. ; CHECK-AVX2: cost of 3 {{.*}} %A3 = fpext
  203. ; CHECK-AVX512: cost of 1 {{.*}} %A3 = fpext
  204. %A3 = fpext <8 x float> %a to <8 x double>
  205. ; CHECK-AVX2: cost of 3 {{.*}} %A4 = fptrunc
  206. ; CHECK-AVX512: cost of 1 {{.*}} %A4 = fptrunc
  207. %A4 = fptrunc <8 x double> undef to <8 x float>
  208. ; CHECK-AVX512: cost of 3 {{.*}} %A5 = fptrunc
  209. %A5 = fptrunc <16 x double> undef to <16 x float>
  210. ret void
  211. }