#
� GLSL.std.450
main , � � � � � �
GL_GOOGLE_cpp_style_line_directive GL_GOOGLE_include_directive main token_idx gl_WorkGroupID tid gl_LocalInvocationID nthreads slot SM slot_mapping ( block_idx * Pc * key_stride * value_stride * num_heads * head_size * block_size * x , 2 block_offset 7 n ? i I head_idx N head_offset S x_idx Y x_offset ^ src_key_idx e src_value_idx m tgt_key_idx � tgt_value_idx � KC � key_cache � � K � key � � VC � value_cache � � V � value � G G G H H # G G " G ! H * # H * # H * # H * # H * # H * # G * G � H � # G � G � " G � ! G � H � H � # G � G � " G � ! G � H � # G � G � " G � ! G � H � H � # G � G � " G � ! G � !
;
+
;
+ � ; + + " ���� # * + * ; + , + - . + 8 + ; + U + g � � � � � � � ; � � � � � � � � ; � � � � � � � � � � ; � � � � � � � � ; � � + � , � � � 6 � ; ; ; ; ; ( ; 2 ; 7 ; ? ; I ; N ; S ; Y ; ^ ; e ; m ; � A
= > A
= > > = A = > = ! � # $ ! " � & � $ % & � % � � & = ) A . / , - = 0 / � 1 ) 0 > ( 1 = 3 A . 4 , - = 5 4 � 6 3 5 > 2 6 A . 9 , 8 = : 9 A . < , ; = = < � > : = > 7 > = @ > ? @ � A � A � C D � E � E = F ? = G 7 � # H F G � H B C � B = J ? A . K , ; = L K � M J L > I M = O ? A . P , ; = Q P � R O Q > N R = T N A . V , U = W V � X T W > S X = Z N A . [ , U = \ [ � ] Z \ > Y ] = _ A . ` , = a ` � b _ a = c ? � d b c > ^ d = f A . h , g = i h � j f i = k ? � l j k > e l = n ( A . o , 8 = p o � q n p A . r , ; = s r A . t , U = u t � v s u � w q v A . x , - = y x � z w y A . { , U = | { � } z | = ~ I A . , ; = � A . � , U = � � � � � � � � ~ � A . � , - = � � � � � � A . � , U = � � � � � � � � } � = � S A . � , - = � � � � � � A . � , U = � � � � � � � � � � = � 2 A . � , U = � � � � � � � � � � = � Y � � � � > m � = � ( A . � , 8 = � � � � � � A . � , ; = � � � � � � A . � , - = � � � � � � = � I A . � , ; = � � � � � � A . � , - = � � � � � � � � � � = � N A . � , - = � � � � � � � � � � = � 2 � � � � > � � = � m = � ^ A � � � � = � � � A � � � � > � � = � � = � e A � � � � = � � � A � � � � > � � � D � D = � = � ? � � � � > ? � � A � C � 8