CM7: Simplify Dilithium 769 NTT code

example.py

@@ -1814,6 +1814,7 @@ def core(self, slothy):
         slothy.config.constraints.stalls_first_attempt = 32
 
         r = slothy.config.reserved_regs
+        r.add("r1")
         r = r.union(f"s{i}" for i in range(31)) # reserve FPR
         slothy.config.reserved_regs = r
 
@@ -1825,13 +1826,12 @@ def core(self, slothy):
         slothy.config.variable_size = True
         slothy.config.split_heuristic = True
         slothy.config.timeout = 360 # Not more than 2min per step
-        slothy.config.split_heuristic_factor = 1
         slothy.config.visualize_expected_performance = False
-        slothy.config.split_heuristic_factor = 4
+        slothy.config.split_heuristic_factor = 5
         slothy.config.split_heuristic_stepsize = 0.15
-        slothy.optimize_loop("layer1234_loop")
+        slothy.optimize_loop("layer1234_loop", forced_loop_type=Arch_Armv7M.BranchLoop)
         slothy.config.split_heuristic_optimize_seam = 6
-        slothy.optimize_loop("layer1234_loop")
+        slothy.optimize_loop("layer1234_loop", forced_loop_type=Arch_Armv7M.BranchLoop)
 
         slothy.config.outputs = ["r14"]
 

examples/naive/armv7m/ntt_769_dilithium.s

@@ -146,7 +146,6 @@ small_ntt_asm_769:
 	// s24: tmp
         // s25: twiddle_ptr
 	vmov s24, tmp
-        vmov s25, twiddle_ptr
 	layer1234_loop:
 		// load a1, a3, ..., a15
 		vmov s23, poly
@@ -251,10 +250,10 @@ small_ntt_asm_769:
 		uadd16 tmp, poly0, poly1
 		usub16 twiddle1, poly0, poly1
 		str.w twiddle1, [poly, #offset]
-		str.w tmp, [poly], #4 // @slothy:core
+		str.w tmp, [poly], #4 // @slothy:core // @slothy:before=cmp
 
 	vmov tmp, s24
-	cmp.w poly, tmp
+	cmp.w poly, tmp // @slothy:id=cmp
 	bne.w layer1234_loop
 
 	sub.w poly, #8*strincr
@@ -266,7 +265,6 @@ small_ntt_asm_769:
 
 	add.w tmp, poly, #strincr2*16
 	vmov s13, tmp
-        vmov twiddle_ptr, s25
     layer567_loop:
 		vmov s23, poly
 		load poly, poly0, poly1, poly2, poly3, #0, #distance2/4, #2*distance2/4, #3*distance2/4