lint: FA-2 FP16 and FP8

Author: Viviane Potocnik

sw/dnn/flashattention_2/src/flashattention_2_fp16.h

@@ -119,7 +119,6 @@ static inline void flashattention_2_fp16(flashattention_2_layer_t layer) {
 
         // Iterate column blocks of K (corresponding to row blocks of V)
         for (int t_c = 0; t_c < T_c; t_c++) {
-
             // DMA copy K column block (B_c, d) and V row block (B_c, d) to
             // TCDM. Both K and V are stored in (S, d) form in memory
             if (!snrt_is_compute_core()) {
@@ -246,7 +245,6 @@ static inline void flashattention_2_fp16(flashattention_2_layer_t layer) {
             snrt_mcycle();
         }  // end of T_c loop
 
-
         // Rescaling for last t_c iteration
         // O_i = diag(l_i_Tc)^-1 * O_i
         if (snrt_is_compute_core()) {

sw/dnn/flashattention_2/src/flashattention_2_fp8.h

@@ -86,7 +86,7 @@ static inline void flashattention_2_fp8(flashattention_2_layer_t layer) {
 
     // Allocate space for V^t
     char *V_t;
-    if(!baseline) {
+    if (!baseline) {
         V_t = tcdm_ptr;
         tcdm_ptr += B_c * d * sizeof(char);
     }