`timescale 1ns / 1ps // ============================================================================= // Testbench for fixed-method softmax approximation (Problem 012) // - Signed Q4.4 inputs // - Unsigned Q0.8 outputs // - Exact match against golden vectors generated by generate_golden.py // ============================================================================= module tb_softmax_approx; reg [7:0] x0, x1, x2, x3; wire [7:0] y0, y1, y2, y3; integer i; integer errors; // Golden vectors for 012_softmax_approx // Generated by: python3 generate_golden.py // Algorithm: fixed piecewise softmax from spec.yaml localparam NUM_TESTS = 48; reg [7:0] test_x0 [0:47]; reg [7:0] test_x1 [0:47]; reg [7:0] test_x2 [0:47]; reg [7:0] test_x3 [0:47]; reg [7:0] exp_y0 [0:47]; reg [7:0] exp_y1 [0:47]; reg [7:0] exp_y2 [0:47]; reg [7:0] exp_y3 [0:47]; initial begin // Test 0: all_zero test_x0[0] = 8'd0; test_x1[0] = 8'd0; test_x2[0] = 8'd0; test_x3[0] = 8'd0; exp_y0[0] = 8'd64; exp_y1[0] = 8'd64; exp_y2[0] = 8'd64; exp_y3[0] = 8'd64; // Test 1: all_equal_pos test_x0[1] = 8'd64; test_x1[1] = 8'd64; test_x2[1] = 8'd64; test_x3[1] = 8'd64; exp_y0[1] = 8'd64; exp_y1[1] = 8'd64; exp_y2[1] = 8'd64; exp_y3[1] = 8'd64; // Test 2: all_equal_neg test_x0[2] = 8'd224; test_x1[2] = 8'd224; test_x2[2] = 8'd224; test_x3[2] = 8'd224; exp_y0[2] = 8'd64; exp_y1[2] = 8'd64; exp_y2[2] = 8'd64; exp_y3[2] = 8'd64; // Test 3: single_dominant_pos test_x0[3] = 8'd127; test_x1[3] = 8'd0; test_x2[3] = 8'd0; test_x3[3] = 8'd0; exp_y0[3] = 8'd255; exp_y1[3] = 8'd0; exp_y2[3] = 8'd0; exp_y3[3] = 8'd0; // Test 4: single_dominant_pos_1 test_x0[4] = 8'd0; test_x1[4] = 8'd127; test_x2[4] = 8'd0; test_x3[4] = 8'd0; exp_y0[4] = 8'd0; exp_y1[4] = 8'd255; exp_y2[4] = 8'd0; exp_y3[4] = 8'd0; // Test 5: single_dominant_pos_2 test_x0[5] = 8'd0; test_x1[5] = 8'd0; test_x2[5] = 8'd127; test_x3[5] = 8'd0; exp_y0[5] = 8'd0; exp_y1[5] = 8'd0; exp_y2[5] = 8'd255; exp_y3[5] = 8'd0; // Test 6: single_dominant_pos_3 test_x0[6] = 8'd0; test_x1[6] = 8'd0; test_x2[6] = 8'd0; test_x3[6] = 8'd127; exp_y0[6] = 8'd0; exp_y1[6] = 8'd0; exp_y2[6] = 8'd0; exp_y3[6] = 8'd255; // Test 7: close_cluster_high test_x0[7] = 8'd127; test_x1[7] = 8'd126; test_x2[7] = 8'd125; test_x3[7] = 8'd124; exp_y0[7] = 8'd70; exp_y1[7] = 8'd66; exp_y2[7] = 8'd61; exp_y3[7] = 8'd57; // Test 8: close_cluster_mid test_x0[8] = 8'd16; test_x1[8] = 8'd15; test_x2[8] = 8'd14; test_x3[8] = 8'd13; exp_y0[8] = 8'd70; exp_y1[8] = 8'd66; exp_y2[8] = 8'd61; exp_y3[8] = 8'd57; // Test 9: boundary_exact test_x0[9] = 8'd0; test_x1[9] = 8'd252; test_x2[9] = 8'd248; test_x3[9] = 8'd240; exp_y0[9] = 8'd95; exp_y1[9] = 8'd71; exp_y2[9] = 8'd53; exp_y3[9] = 8'd34; // Test 10: boundary_just_above test_x0[10] = 8'd0; test_x1[10] = 8'd253; test_x2[10] = 8'd249; test_x3[10] = 8'd241; exp_y0[10] = 8'd93; exp_y1[10] = 8'd76; exp_y2[10] = 8'd57; exp_y3[10] = 8'd29; // Test 11: boundary_just_below test_x0[11] = 8'd0; test_x1[11] = 8'd251; test_x2[11] = 8'd247; test_x3[11] = 8'd239; exp_y0[11] = 8'd99; exp_y1[11] = 8'd70; exp_y2[11] = 8'd52; exp_y3[11] = 8'd33; // Test 12: wide_negative_span test_x0[12] = 8'd0; test_x1[12] = 8'd224; test_x2[12] = 8'd192; test_x3[12] = 8'd128; exp_y0[12] = 8'd225; exp_y1[12] = 8'd30; exp_y2[12] = 8'd0; exp_y3[12] = 8'd0; // Test 13: mixed_signed_values test_x0[13] = 8'd32; test_x1[13] = 8'd240; test_x2[13] = 8'd48; test_x3[13] = 8'd224; exp_y0[13] = 8'd67; exp_y1[13] = 8'd0; exp_y2[13] = 8'd188; exp_y3[13] = 8'd0; // Test 14: negative_close_cluster test_x0[14] = 8'd255; test_x1[14] = 8'd254; test_x2[14] = 8'd253; test_x3[14] = 8'd252; exp_y0[14] = 8'd70; exp_y1[14] = 8'd66; exp_y2[14] = 8'd61; exp_y3[14] = 8'd57; // Test 15: descending_positive test_x0[15] = 8'd127; test_x1[15] = 8'd63; test_x2[15] = 8'd31; test_x3[15] = 8'd0; exp_y0[15] = 8'd255; exp_y1[15] = 8'd0; exp_y2[15] = 8'd0; exp_y3[15] = 8'd0; // Test 16: ascending_positive test_x0[16] = 8'd0; test_x1[16] = 8'd16; test_x2[16] = 8'd32; test_x3[16] = 8'd48; exp_y0[16] = 8'd0; exp_y1[16] = 8'd22; exp_y2[16] = 8'd61; exp_y3[16] = 8'd171; // Test 17: ascending_negative test_x0[17] = 8'd128; test_x1[17] = 8'd144; test_x2[17] = 8'd160; test_x3[17] = 8'd176; exp_y0[17] = 8'd0; exp_y1[17] = 8'd22; exp_y2[17] = 8'd61; exp_y3[17] = 8'd171; // Test 18: alternating_extremes test_x0[18] = 8'd127; test_x1[18] = 8'd128; test_x2[18] = 8'd127; test_x3[18] = 8'd128; exp_y0[18] = 8'd128; exp_y1[18] = 8'd0; exp_y2[18] = 8'd128; exp_y3[18] = 8'd0; // Test 19: three_similar_one_low test_x0[19] = 8'd80; test_x1[19] = 8'd79; test_x2[19] = 8'd78; test_x3[19] = 8'd16; exp_y0[19] = 8'd90; exp_y1[19] = 8'd85; exp_y2[19] = 8'd79; exp_y3[19] = 8'd0; // Test 20: three_similar_permuted test_x0[20] = 8'd80; test_x1[20] = 8'd16; test_x2[20] = 8'd79; test_x3[20] = 8'd78; exp_y0[20] = 8'd90; exp_y1[20] = 8'd0; exp_y2[20] = 8'd85; exp_y3[20] = 8'd79; // Test 21: three_similar_permuted2 test_x0[21] = 8'd16; test_x1[21] = 8'd80; test_x2[21] = 8'd79; test_x3[21] = 8'd78; exp_y0[21] = 8'd0; exp_y1[21] = 8'd90; exp_y2[21] = 8'd85; exp_y3[21] = 8'd79; // Test 22: neg_to_zero test_x0[22] = 8'd128; test_x1[22] = 8'd192; test_x2[22] = 8'd224; test_x3[22] = 8'd0; exp_y0[22] = 8'd0; exp_y1[22] = 8'd0; exp_y2[22] = 8'd30; exp_y3[22] = 8'd225; // Test 23: mixed_seed_like test_x0[23] = 8'd8; test_x1[23] = 8'd89; test_x2[23] = 8'd52; test_x3[23] = 8'd129; exp_y0[23] = 8'd0; exp_y1[23] = 8'd234; exp_y2[23] = 8'd21; exp_y3[23] = 8'd0; // Test 24: random_00 test_x0[24] = 8'd213; test_x1[24] = 8'd5; test_x2[24] = 8'd152; test_x3[24] = 8'd188; exp_y0[24] = 8'd0; exp_y1[24] = 8'd255; exp_y2[24] = 8'd0; exp_y3[24] = 8'd0; // Test 25: random_01 test_x0[25] = 8'd99; test_x1[25] = 8'd138; test_x2[25] = 8'd223; test_x3[25] = 8'd82; exp_y0[25] = 8'd191; exp_y1[25] = 8'd0; exp_y2[25] = 8'd0; exp_y3[25] = 8'd64; // Test 26: random_02 test_x0[26] = 8'd191; test_x1[26] = 8'd63; test_x2[26] = 8'd221; test_x3[26] = 8'd133; exp_y0[26] = 8'd0; exp_y1[26] = 8'd255; exp_y2[26] = 8'd0; exp_y3[26] = 8'd0; // Test 27: random_03 test_x0[27] = 8'd89; test_x1[27] = 8'd95; test_x2[27] = 8'd181; test_x3[27] = 8'd46; exp_y0[27] = 8'd101; exp_y1[27] = 8'd154; exp_y2[27] = 8'd0; exp_y3[27] = 8'd0; // Test 28: random_04 test_x0[28] = 8'd211; test_x1[28] = 8'd85; test_x2[28] = 8'd75; test_x3[28] = 8'd105; exp_y0[28] = 8'd0; exp_y1[28] = 8'd52; exp_y2[28] = 8'd8; exp_y3[28] = 8'd194; // Test 29: random_05 test_x0[29] = 8'd39; test_x1[29] = 8'd97; test_x2[29] = 8'd174; test_x3[29] = 8'd164; exp_y0[29] = 8'd0; exp_y1[29] = 8'd255; exp_y2[29] = 8'd0; exp_y3[29] = 8'd0; // Test 30: random_06 test_x0[30] = 8'd12; test_x1[30] = 8'd234; test_x2[30] = 8'd173; test_x3[30] = 8'd13; exp_y0[30] = 8'd117; exp_y1[30] = 8'd13; exp_y2[30] = 8'd0; exp_y3[30] = 8'd125; // Test 31: random_07 test_x0[31] = 8'd212; test_x1[31] = 8'd1; test_x2[31] = 8'd3; test_x3[31] = 8'd83; exp_y0[31] = 8'd0; exp_y1[31] = 8'd0; exp_y2[31] = 8'd0; exp_y3[31] = 8'd255; // Test 32: random_08 test_x0[32] = 8'd91; test_x1[32] = 8'd146; test_x2[32] = 8'd50; test_x3[32] = 8'd212; exp_y0[32] = 8'd242; exp_y1[32] = 8'd0; exp_y2[32] = 8'd13; exp_y3[32] = 8'd0; // Test 33: random_09 test_x0[33] = 8'd54; test_x1[33] = 8'd93; test_x2[33] = 8'd192; test_x3[33] = 8'd119; exp_y0[33] = 8'd0; exp_y1[33] = 8'd30; exp_y2[33] = 8'd0; exp_y3[33] = 8'd225; // Test 34: random_10 test_x0[34] = 8'd168; test_x1[34] = 8'd20; test_x2[34] = 8'd118; test_x3[34] = 8'd61; exp_y0[34] = 8'd0; exp_y1[34] = 8'd0; exp_y2[34] = 8'd255; exp_y3[34] = 8'd0; // Test 35: random_11 test_x0[35] = 8'd255; test_x1[35] = 8'd110; test_x2[35] = 8'd95; test_x3[35] = 8'd131; exp_y0[35] = 8'd0; exp_y1[35] = 8'd194; exp_y2[35] = 8'd61; exp_y3[35] = 8'd0; // Test 36: random_12 test_x0[36] = 8'd222; test_x1[36] = 8'd66; test_x2[36] = 8'd21; test_x3[36] = 8'd84; exp_y0[36] = 8'd0; exp_y1[36] = 8'd61; exp_y2[36] = 8'd0; exp_y3[36] = 8'd194; // Test 37: random_13 test_x0[37] = 8'd154; test_x1[37] = 8'd4; test_x2[37] = 8'd76; test_x3[37] = 8'd117; exp_y0[37] = 8'd0; exp_y1[37] = 8'd0; exp_y2[37] = 8'd13; exp_y3[37] = 8'd242; // Test 38: random_14 test_x0[38] = 8'd179; test_x1[38] = 8'd175; test_x2[38] = 8'd80; test_x3[38] = 8'd197; exp_y0[38] = 8'd0; exp_y1[38] = 8'd0; exp_y2[38] = 8'd255; exp_y3[38] = 8'd0; // Test 39: random_15 test_x0[39] = 8'd165; test_x1[39] = 8'd6; test_x2[39] = 8'd25; test_x3[39] = 8'd31; exp_y0[39] = 8'd0; exp_y1[39] = 8'd22; exp_y2[39] = 8'd92; exp_y3[39] = 8'd140; // Test 40: random_16 test_x0[40] = 8'd227; test_x1[40] = 8'd40; test_x2[40] = 8'd198; test_x3[40] = 8'd91; exp_y0[40] = 8'd0; exp_y1[40] = 8'd0; exp_y2[40] = 8'd0; exp_y3[40] = 8'd255; // Test 41: random_17 test_x0[41] = 8'd5; test_x1[41] = 8'd188; test_x2[41] = 8'd110; test_x3[41] = 8'd62; exp_y0[41] = 8'd0; exp_y1[41] = 8'd0; exp_y2[41] = 8'd255; exp_y3[41] = 8'd0; // Test 42: random_18 test_x0[42] = 8'd246; test_x1[42] = 8'd31; test_x2[42] = 8'd7; test_x3[42] = 8'd251; exp_y0[42] = 8'd10; exp_y1[42] = 8'd192; exp_y2[42] = 8'd34; exp_y3[42] = 8'd18; // Test 43: random_19 test_x0[43] = 8'd166; test_x1[43] = 8'd215; test_x2[43] = 8'd103; test_x3[43] = 8'd186; exp_y0[43] = 8'd0; exp_y1[43] = 8'd0; exp_y2[43] = 8'd255; exp_y3[43] = 8'd0; // Test 44: random_20 test_x0[44] = 8'd61; test_x1[44] = 8'd76; test_x2[44] = 8'd40; test_x3[44] = 8'd253; exp_y0[44] = 8'd57; exp_y1[44] = 8'd180; exp_y2[44] = 8'd17; exp_y3[44] = 8'd0; // Test 45: random_21 test_x0[45] = 8'd84; test_x1[45] = 8'd171; test_x2[45] = 8'd118; test_x3[45] = 8'd104; exp_y0[45] = 8'd19; exp_y1[45] = 8'd0; exp_y2[45] = 8'd174; exp_y3[45] = 8'd61; // Test 46: random_22 test_x0[46] = 8'd83; test_x1[46] = 8'd83; test_x2[46] = 8'd147; test_x3[46] = 8'd76; exp_y0[46] = 8'd98; exp_y1[46] = 8'd98; exp_y2[46] = 8'd0; exp_y3[46] = 8'd59; // Test 47: random_23 test_x0[47] = 8'd46; test_x1[47] = 8'd1; test_x2[47] = 8'd75; test_x3[47] = 8'd1; exp_y0[47] = 8'd16; exp_y1[47] = 8'd0; exp_y2[47] = 8'd240; exp_y3[47] = 8'd0; end // Summary: // 0. all_zero x=[ 0.00, 0.00, 0.00, 0.00] -> y=[0.250,0.250,0.250,0.250] sum=256 // 1. all_equal_pos x=[ 4.00, 4.00, 4.00, 4.00] -> y=[0.250,0.250,0.250,0.250] sum=256 // 2. all_equal_neg x=[ -2.00, -2.00, -2.00, -2.00] -> y=[0.250,0.250,0.250,0.250] sum=256 // 3. single_dominant_pos x=[ 7.94, 0.00, 0.00, 0.00] -> y=[0.996,0.000,0.000,0.000] sum=255 // 4. single_dominant_pos_1 x=[ 0.00, 7.94, 0.00, 0.00] -> y=[0.000,0.996,0.000,0.000] sum=255 // 5. single_dominant_pos_2 x=[ 0.00, 0.00, 7.94, 0.00] -> y=[0.000,0.000,0.996,0.000] sum=255 // 6. single_dominant_pos_3 x=[ 0.00, 0.00, 0.00, 7.94] -> y=[0.000,0.000,0.000,0.996] sum=255 // 7. close_cluster_high x=[ 7.94, 7.88, 7.81, 7.75] -> y=[0.273,0.258,0.238,0.223] sum=254 // 8. close_cluster_mid x=[ 1.00, 0.94, 0.88, 0.81] -> y=[0.273,0.258,0.238,0.223] sum=254 // 9. boundary_exact x=[ 0.00, -0.25, -0.50, -1.00] -> y=[0.371,0.277,0.207,0.133] sum=253 // 10. boundary_just_above x=[ 0.00, -0.19, -0.44, -0.94] -> y=[0.363,0.297,0.223,0.113] sum=255 // 11. boundary_just_below x=[ 0.00, -0.31, -0.56, -1.06] -> y=[0.387,0.273,0.203,0.129] sum=254 // 12. wide_negative_span x=[ 0.00, -2.00, -4.00, -8.00] -> y=[0.879,0.117,0.000,0.000] sum=255 // 13. mixed_signed_values x=[ 2.00, -1.00, 3.00, -2.00] -> y=[0.262,0.000,0.734,0.000] sum=255 // 14. negative_close_cluster x=[ -0.06, -0.12, -0.19, -0.25] -> y=[0.273,0.258,0.238,0.223] sum=254 // 15. descending_positive x=[ 7.94, 3.94, 1.94, 0.00] -> y=[0.996,0.000,0.000,0.000] sum=255 // 16. ascending_positive x=[ 0.00, 1.00, 2.00, 3.00] -> y=[0.000,0.086,0.238,0.668] sum=254 // 17. ascending_negative x=[ -8.00, -7.00, -6.00, -5.00] -> y=[0.000,0.086,0.238,0.668] sum=254 // 18. alternating_extremes x=[ 7.94, -8.00, 7.94, -8.00] -> y=[0.500,0.000,0.500,0.000] sum=256 // 19. three_similar_one_low x=[ 5.00, 4.94, 4.88, 1.00] -> y=[0.352,0.332,0.309,0.000] sum=254 // 20. three_similar_permuted x=[ 5.00, 1.00, 4.94, 4.88] -> y=[0.352,0.000,0.332,0.309] sum=254 // 21. three_similar_permuted2 x=[ 1.00, 5.00, 4.94, 4.88] -> y=[0.000,0.352,0.332,0.309] sum=254 // 22. neg_to_zero x=[ -8.00, -4.00, -2.00, 0.00] -> y=[0.000,0.000,0.117,0.879] sum=255 // 23. mixed_seed_like x=[ 0.50, 5.56, 3.25, -7.94] -> y=[0.000,0.914,0.082,0.000] sum=255 // 24. random_00 x=[ -2.69, 0.31, -6.50, -4.25] -> y=[0.000,0.996,0.000,0.000] sum=255 // 25. random_01 x=[ 6.19, -7.38, -2.06, 5.12] -> y=[0.746,0.000,0.000,0.250] sum=255 // 26. random_02 x=[ -4.06, 3.94, -2.19, -7.69] -> y=[0.000,0.996,0.000,0.000] sum=255 // 27. random_03 x=[ 5.56, 5.94, -4.69, 2.88] -> y=[0.395,0.602,0.000,0.000] sum=255 // 28. random_04 x=[ -2.81, 5.31, 4.69, 6.56] -> y=[0.000,0.203,0.031,0.758] sum=254 // 29. random_05 x=[ 2.44, 6.06, -5.12, -5.75] -> y=[0.000,0.996,0.000,0.000] sum=255 // 30. random_06 x=[ 0.75, -1.38, -5.19, 0.81] -> y=[0.457,0.051,0.000,0.488] sum=255 // 31. random_07 x=[ -2.75, 0.06, 0.19, 5.19] -> y=[0.000,0.000,0.000,0.996] sum=255 // 32. random_08 x=[ 5.69, -6.88, 3.12, -2.75] -> y=[0.945,0.000,0.051,0.000] sum=255 // 33. random_09 x=[ 3.38, 5.81, -4.00, 7.44] -> y=[0.000,0.117,0.000,0.879] sum=255 // 34. random_10 x=[ -5.50, 1.25, 7.38, 3.81] -> y=[0.000,0.000,0.996,0.000] sum=255 // 35. random_11 x=[ -0.06, 6.88, 5.94, -7.81] -> y=[0.000,0.758,0.238,0.000] sum=255 // 36. random_12 x=[ -2.12, 4.12, 1.31, 5.25] -> y=[0.000,0.238,0.000,0.758] sum=255 // 37. random_13 x=[ -6.38, 0.25, 4.75, 7.31] -> y=[0.000,0.000,0.051,0.945] sum=255 // 38. random_14 x=[ -4.81, -5.06, 5.00, -3.69] -> y=[0.000,0.000,0.996,0.000] sum=255 // 39. random_15 x=[ -5.69, 0.38, 1.56, 1.94] -> y=[0.000,0.086,0.359,0.547] sum=254 // 40. random_16 x=[ -1.81, 2.50, -3.62, 5.69] -> y=[0.000,0.000,0.000,0.996] sum=255 // 41. random_17 x=[ 0.31, -4.25, 6.88, 3.88] -> y=[0.000,0.000,0.996,0.000] sum=255 // 42. random_18 x=[ -0.62, 1.94, 0.44, -0.31] -> y=[0.039,0.750,0.133,0.070] sum=254 // 43. random_19 x=[ -5.62, -2.56, 6.44, -4.38] -> y=[0.000,0.000,0.996,0.000] sum=255 // 44. random_20 x=[ 3.81, 4.75, 2.50, -0.19] -> y=[0.223,0.703,0.066,0.000] sum=254 // 45. random_21 x=[ 5.25, -5.31, 7.38, 6.50] -> y=[0.074,0.000,0.680,0.238] sum=254 // 46. random_22 x=[ 5.19, 5.19, -6.81, 4.75] -> y=[0.383,0.383,0.000,0.230] sum=255 // 47. random_23 x=[ 2.88, 0.06, 4.69, 0.06] -> y=[0.062,0.000,0.938,0.000] sum=256 softmax_approx uut ( .x0(x0), .x1(x1), .x2(x2), .x3(x3), .y0(y0), .y1(y1), .y2(y2), .y3(y3) ); initial begin errors = 0; x0 = 0; x1 = 0; x2 = 0; x3 = 0; $display("==========================================="); $display(" Softmax Approximation Testbench (Fixed)"); $display(" Tests: %0d", NUM_TESTS); $display("==========================================="); #1; for (i = 0; i < NUM_TESTS; i = i + 1) begin x0 = test_x0[i]; x1 = test_x1[i]; x2 = test_x2[i]; x3 = test_x3[i]; #1; if (y0 !== exp_y0[i]) begin $display("ERROR [Test %0d]: y0 expected=%0d got=%0d", i, exp_y0[i], y0); errors = errors + 1; end if (y1 !== exp_y1[i]) begin $display("ERROR [Test %0d]: y1 expected=%0d got=%0d", i, exp_y1[i], y1); errors = errors + 1; end if (y2 !== exp_y2[i]) begin $display("ERROR [Test %0d]: y2 expected=%0d got=%0d", i, exp_y2[i], y2); errors = errors + 1; end if (y3 !== exp_y3[i]) begin $display("ERROR [Test %0d]: y3 expected=%0d got=%0d", i, exp_y3[i], y3); errors = errors + 1; end end $display(""); $display("==========================================="); $display(" Tests Run: %0d", NUM_TESTS); $display("==========================================="); if (errors == 0) begin $display("TEST_RESULT: PASS"); end else begin $display("TEST_RESULT: FAIL (%0d errors)", errors); end $finish; end endmodule