* Add a missing ! in write_output_regular().

 * Use while loops instead of do while, since loop_next checks for zero before
   decrementing, not after.
Regular FFT seems to work now, though there seem to be quite some rounding errors.
Testing a larger FFT is hard, since decoding the results is still manual.

diff --git a/FFT.mc b/FFT.mc
index 626b0dc51508f41d2eff5804bed7b8ac8f2537a7..6688275a9f32d7a7341ead36d3859efe18578e57 100644 (file)
--- a/FFT.mc
+++ b/FFT.mc
@@ -52,7 +52,7 @@ INLINE void write_output_regular(struct mems m, struct bf_out res, bool second_h
        add_offset(m.output_b_re, 2);\r
        add_offset(m.output_b_im, 2);\r
        \r
-       if (second_half) {\r
+       if (!second_half) {\r
                write_mem(m.output_a_re, res.a_re);\r
                write_mem(m.output_a_im, res.a_im);\r
                write_mem(m.output_b_re, res.b_re);\r
@@ -187,9 +187,9 @@ INLINE void do_half_regular_stage(struct mems m, bool stage_odd, bool second_hal
 \r
        /* Now, do half a single stage. That means N_t / 4 cycles. Since we do 2\r
         * cycles on every iteration, plus one before and after the loop,\r
-        * we will loop N_t / 8 - 1 times. */\r
+        * we will loop N_t / 8 - 1 times. We add an extra - 1 because this is a do while loop... */\r
        init_loop(LC2, (PARAM_N_t / 8) - 1);\r
-       do {\r
+       while (loop_next(LC2)) {\r
                /* Write outputs of previous cycle */\r
                write_output_regular(m, out, second_half);\r
 \r
@@ -204,7 +204,7 @@ INLINE void do_half_regular_stage(struct mems m, bool stage_odd, bool second_hal
                /* Even cycle */\r
                in = read_input_regular(m, EVEN_CYCLE, second_half);\r
                out = butterfly(in);\r
-       } while (loop_next(LC2));\r
+       }\r
        \r
        /* Write outputs of previous cycle */\r
        write_output_regular(m, out, second_half);\r
@@ -256,7 +256,7 @@ void run() {
        /* We need to do n_t regular stages. Since we do two stages each\r
         * iteration, we'll do n_t / 2 iterations. */\r
        init_loop(LC1, (PARAM_n_t / 2));\r
-       do {\r
+       while (loop_next(LC1)) {\r
                m = init_mem_mapping(EVEN_STAGE);\r
                init_input_addresses_regular(m, EVEN_STAGE);\r
                /* do_half_regular_stage will init output addresses */\r
@@ -271,5 +271,5 @@ void run() {
                do_half_regular_stage(m, ODD_STAGE, FIRST_HALF);\r
                do_half_regular_stage(m, ODD_STAGE, SECOND_HALF);\r
                stage++;\r
-       } while (loop_next(LC1));\r
+       }\r
 }\r