{"family": "elementwise", "bug_class": "mask_leak", "correct_name": "gelu", "buggy_name": "elementwise_nomask", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n\n@triton.jit\ndef gelu(x_ptr, out_ptr, M, N, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        x = tl.load(x_ptr + row * N + cols, mask=mask, other=0.0)\n        y = 0.5 * x * (1.0 + tl.erf(x * 0.7071067811865476))\n        tl.store(out_ptr + row * N + cols, y, mask=mask)", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*N\n# litmus: buffer w_ptr = N\n# litmus: buffer b_ptr = N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n\n@triton.jit\ndef elementwise_scale_bias(x_ptr, w_ptr, b_ptr, out_ptr, M, N, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        x = tl.load(x_ptr + row * N + cols)\n        w = tl.load(w_ptr + cols)\n        b = tl.load(b_ptr + cols)\n        tl.store(out_ptr + row * N + cols, x * w + b)", "why_standard_misses": "N=512 is a multiple of BLOCK=32 so no tail exists -- and even at a ragged shape the spilled write lands in a neighbouring allocation, so the kernel's own output is unchanged and no allclose test can see it", "witness_inputs": "", "witness_dims": "", "numerically_observable": false}
{"family": "reduction", "bug_class": "mask_leak", "correct_name": "row_max", "buggy_name": "row_sum_nomask", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_reduce(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.zeros((BLOCK,), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols)\n        acc += v\n    tl.store(out_ptr + row, tl.sum(acc, axis=0))", "why_standard_misses": "N=512 is a multiple of BLOCK=32, so no tail tile exists", "witness_inputs": "{\"x\": {\"shape\": [3, 33], \"truncated\": false, \"values\": [[0.9323224351204542, -0.13484240850688187, -1.0977711087115172, 1.2093136358526926, -1.5200609774258935, -0.3659055784186133, 0.08753656217963342, -0.673525657000616, -0.31220063548687593, 0.6517512317274256, 0.10188336399475949, 2.1183280668541165, -0.1747104908892027, -0.05302204609851407, 1.1776554100520484, 1.034069668664329, 1.3570641388632136, -0.26345967846302504, 0.6563534008927115, -1.0510361039263771, -0.2951569873791522, -0.6036145420186865, 0.4275727371703606, 1.164091172456542, -0.18349346919422196, -0.06950722813263917, 0.9665602467906304, -0.6179201120759003, 1.307118331059199, -1.809250277245433, -0.24609517424533933, 0.047437465889070246, -0.1726226634799063], [-0.4265367270080289, -0.8613653826306371, -1.884530122726135, -0.842549971820494, 0.4946735479336724, 0.4465705367621375, 1.85655320464552, 0.7426474381606197, 1.6626073177772, -0.7445255020708731, -0.8749235310133816, -0.1406877389606945, 0.4921162579956519, -0.23537017889100362, -0.2329928411829563, 1.0203585485236448, -0.3950961781815714, 1.586702826547813, 0.7790573295349854, 0.6084367506154511, -0.7746609271150169, 0.5145944304272745, 0.30077846461329444, 1.0300494046122595, -0.38908207485805274, 1.1903487622827287, -2.3898260878140896, -0.22780056268989937, 1.780801496873732, 0.052726437928369584, -0.2905865934347286, 1.5681129649560095, -0.34023019162555135], [-0.590480507527205, 0.3952876666332603, 0.41516876928919283, 0.3747274728808734, -0.6381902366431598, -0.4137973997801163, -0.38841359754271754, 0.9259711547147788, -0.5889466811888183, -0.131217053696824, 0.00859743441700256, 0.12076421285922352, 0.23244373318601652, 0.12115874783464392, 0.7007774085680416, -1.1021614558814128, -0.20761115023330756, 1.0637356081910359, 0.4318807786996173, 0.5185060470598175, -0.12765925263970465, 0.0023738622684158404, 0.32685676241280304, 1.574304297758674, -0.497854771468547, 1.1708647780992363, -1.1308104177840286, -0.5076025950103698, 0.3658925302392737, 0.11861084275632296, -0.7123668701917418, 0.15577888425550526, 1.376028723925904]]}, \"w\": {\"shape\": [33], \"truncated\": false, \"values\": [-0.8521657544163167, -0.9913974466747576, -0.6171660403968796, 0.36085432621749663, 0.725788747445546, 0.7048039780615126, 0.856549025221355, 1.1007333725162887, -0.5368462420915631, 0.4897982425989597, 1.9345120072465691, 0.6114975628959227, 2.545833547596705, 0.8446100794186416, -0.4976219513230361, -0.6579083416586656, 0.15435453835917065, -1.2352837415993694, -1.5907987411963338, -1.3607153546827384, 0.4951250366915605, -1.5295864071475327, 0.02331219452117253, 1.5065421776698205, -1.0374513272097987, 0.6412381170410911, 1.4182025956774233, 0.4869832493833102, -1.187126614866286, -0.5977878136525421, 0.5499482947031736, -1.282288774354227, -2.0776904842464647]}, \"b\": {\"shape\": [33], \"truncated\": false, \"values\": [2.061194735710628, -0.5525172695521284, 0.5098670391746202, -1.363457812071284, -0.19746682414381686, 1.1420645413492911, 0.6255783450063146, -0.061071397736976014, -0.8579834503967044, 1.983500446602315, -0.06843040976465128, -1.954830554634629, 1.0475056154390079, -1.611744070743091, -0.17456089960628418, -0.015990131632859814, -0.9816276242759144, 0.18854014565014418, -0.4928272722487404, 0.5059976051141007, 0.8652126974567892, 0.4340038332999275, 0.2040632519731441, -0.5270238182528804, 1.0758099641593462, -0.6375573748121397, 1.716919815122407, 0.08713222254118819, 2.0350573599285773, -0.16991243315971818, -0.20766799832413915, 0.42743737204579724, 0.6971226418809447]}}", "witness_dims": "{\"M\": 3, \"N\": 33}", "numerically_observable": true}
{"family": "softmax", "bug_class": "mask_leak", "correct_name": "softmax", "buggy_name": "softmax_nomask", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n\n@triton.jit\ndef softmax(x_ptr, out_ptr, M, N, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    mx = float('-inf')\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        mx = tl.maximum(mx, tl.max(v, axis=0))\n    denom = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        e = tl.where(mask, tl.exp(v - mx), 0.0)\n        denom += tl.sum(e, axis=0)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        tl.store(out_ptr + row * N + cols, tl.exp(v - mx) / denom, mask=mask)", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n\n@triton.jit\ndef softmax(x_ptr, out_ptr, M, N, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    mx = float('-inf')\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols)\n        mx = tl.maximum(mx, tl.max(v, axis=0))\n    denom = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols)\n        e = tl.where(mask, tl.exp(v - mx), 0.0)\n        denom += tl.sum(e, axis=0)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols)\n        tl.store(out_ptr + row * N + cols, tl.exp(v - mx) / denom)", "why_standard_misses": "N=512 is a multiple of BLOCK=32, so no tail tile exists", "witness_inputs": "{\"x\": {\"shape\": [3, 33], \"truncated\": false, \"values\": [[0.9323224351204542, -0.13484240850688187, -1.0977711087115172, 1.2093136358526926, -1.5200609774258935, -0.3659055784186133, 0.08753656217963342, -0.673525657000616, -0.31220063548687593, 0.6517512317274256, 0.10188336399475949, 2.1183280668541165, -0.1747104908892027, -0.05302204609851407, 1.1776554100520484, 1.034069668664329, 1.3570641388632136, -0.26345967846302504, 0.6563534008927115, -1.0510361039263771, -0.2951569873791522, -0.6036145420186865, 0.4275727371703606, 1.164091172456542, -0.18349346919422196, -0.06950722813263917, 0.9665602467906304, -0.6179201120759003, 1.307118331059199, -1.809250277245433, -0.24609517424533933, 0.047437465889070246, -0.1726226634799063], [-0.4265367270080289, -0.8613653826306371, -1.884530122726135, -0.842549971820494, 0.4946735479336724, 0.4465705367621375, 1.85655320464552, 0.7426474381606197, 1.6626073177772, -0.7445255020708731, -0.8749235310133816, -0.1406877389606945, 0.4921162579956519, -0.23537017889100362, -0.2329928411829563, 1.0203585485236448, -0.3950961781815714, 1.586702826547813, 0.7790573295349854, 0.6084367506154511, -0.7746609271150169, 0.5145944304272745, 0.30077846461329444, 1.0300494046122595, -0.38908207485805274, 1.1903487622827287, -2.3898260878140896, -0.22780056268989937, 1.780801496873732, 0.052726437928369584, -0.2905865934347286, 1.5681129649560095, -0.34023019162555135], [-0.590480507527205, 0.3952876666332603, 0.41516876928919283, 0.3747274728808734, -0.6381902366431598, -0.4137973997801163, -0.38841359754271754, 0.9259711547147788, -0.5889466811888183, -0.131217053696824, 0.00859743441700256, 0.12076421285922352, 0.23244373318601652, 0.12115874783464392, 0.7007774085680416, -1.1021614558814128, -0.20761115023330756, 1.0637356081910359, 0.4318807786996173, 0.5185060470598175, -0.12765925263970465, 0.0023738622684158404, 0.32685676241280304, 1.574304297758674, -0.497854771468547, 1.1708647780992363, -1.1308104177840286, -0.5076025950103698, 0.3658925302392737, 0.11861084275632296, -0.7123668701917418, 0.15577888425550526, 1.376028723925904]]}, \"w\": {\"shape\": [33], \"truncated\": false, \"values\": [-0.8521657544163167, -0.9913974466747576, -0.6171660403968796, 0.36085432621749663, 0.725788747445546, 0.7048039780615126, 0.856549025221355, 1.1007333725162887, -0.5368462420915631, 0.4897982425989597, 1.9345120072465691, 0.6114975628959227, 2.545833547596705, 0.8446100794186416, -0.4976219513230361, -0.6579083416586656, 0.15435453835917065, -1.2352837415993694, -1.5907987411963338, -1.3607153546827384, 0.4951250366915605, -1.5295864071475327, 0.02331219452117253, 1.5065421776698205, -1.0374513272097987, 0.6412381170410911, 1.4182025956774233, 0.4869832493833102, -1.187126614866286, -0.5977878136525421, 0.5499482947031736, -1.282288774354227, -2.0776904842464647]}, \"b\": {\"shape\": [33], \"truncated\": false, \"values\": [2.061194735710628, -0.5525172695521284, 0.5098670391746202, -1.363457812071284, -0.19746682414381686, 1.1420645413492911, 0.6255783450063146, -0.061071397736976014, -0.8579834503967044, 1.983500446602315, -0.06843040976465128, -1.954830554634629, 1.0475056154390079, -1.611744070743091, -0.17456089960628418, -0.015990131632859814, -0.9816276242759144, 0.18854014565014418, -0.4928272722487404, 0.5059976051141007, 0.8652126974567892, 0.4340038332999275, 0.2040632519731441, -0.5270238182528804, 1.0758099641593462, -0.6375573748121397, 1.716919815122407, 0.08713222254118819, 2.0350573599285773, -0.16991243315971818, -0.20766799832413915, 0.42743737204579724, 0.6971226418809447]}}", "witness_dims": "{\"M\": 3, \"N\": 33}", "numerically_observable": true}
{"family": "matmul", "bug_class": "mask_leak", "correct_name": "matmul_fp16", "buggy_name": "matmul_no_k_mask", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_M = 32\n# litmus: constexpr BLOCK_N = 32\n# litmus: constexpr BLOCK_K = 32\n# litmus: buffer a_ptr = M*K\n# litmus: buffer b_ptr = K*N\n# litmus: buffer c_ptr = M*N\n# litmus: grid = cdiv(M, BLOCK_M), cdiv(N, BLOCK_N)\n\n@triton.jit\ndef matmul(a_ptr, b_ptr, c_ptr, M, N, K,\n           BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr):\n    pm = tl.program_id(0)\n    pn = tl.program_id(1)\n    offs_m = pm * BLOCK_M + tl.arange(0, BLOCK_M)\n    offs_n = pn * BLOCK_N + tl.arange(0, BLOCK_N)\n    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)\n    for k in range(0, tl.cdiv(K, BLOCK_K)):\n        offs_k = k * BLOCK_K + tl.arange(0, BLOCK_K)\n        a = tl.load(a_ptr + offs_m[:, None] * K + offs_k[None, :],\n                    mask=(offs_m[:, None] < M) & (offs_k[None, :] < K), other=0.0)\n        b = tl.load(b_ptr + offs_k[:, None] * N + offs_n[None, :],\n                    mask=(offs_k[:, None] < K) & (offs_n[None, :] < N), other=0.0)\n        acc += tl.dot(a, b)\n    tl.store(c_ptr + offs_m[:, None] * N + offs_n[None, :], acc,\n             mask=(offs_m[:, None] < M) & (offs_n[None, :] < N))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_M = 32\n# litmus: constexpr BLOCK_N = 32\n# litmus: constexpr BLOCK_K = 32\n# litmus: buffer a_ptr = M*K\n# litmus: buffer b_ptr = K*N\n# litmus: buffer c_ptr = M*N\n# litmus: grid = cdiv(M, BLOCK_M), cdiv(N, BLOCK_N)\n\n@triton.jit\ndef matmul(a_ptr, b_ptr, c_ptr, M, N, K,\n           BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr):\n    pm = tl.program_id(0)\n    pn = tl.program_id(1)\n    offs_m = pm * BLOCK_M + tl.arange(0, BLOCK_M)\n    offs_n = pn * BLOCK_N + tl.arange(0, BLOCK_N)\n    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)\n    for k in range(0, tl.cdiv(K, BLOCK_K)):\n        offs_k = k * BLOCK_K + tl.arange(0, BLOCK_K)\n        a = tl.load(a_ptr + offs_m[:, None] * K + offs_k[None, :],\n                    mask=(offs_m[:, None] < M), other=0.0)\n        b = tl.load(b_ptr + offs_k[:, None] * N + offs_n[None, :],\n                    mask=(offs_n[None, :] < N), other=0.0)\n        acc += tl.dot(a, b)\n    tl.store(c_ptr + offs_m[:, None] * N + offs_n[None, :], acc,\n             mask=(offs_m[:, None] < M) & (offs_n[None, :] < N))", "why_standard_misses": "M, N and K are all multiples of the block sizes", "witness_inputs": "{\"a\": {\"shape\": [33, 33], \"truncated\": false, \"values\": [[-1.0748366301390833, -1.2726386174365862, 0.1279170812576446, 0.8101496187256504, 0.532242471968371, 0.2531135831967665, -1.2944793771672414, 1.3746620788661008, -1.9738913399051794, -0.6994014763584023, -0.5944289753341545, 1.975460784799716, 0.5942356035795224, 0.4592726014176917, 1.9200064333777975, 1.4626689524475038, 0.6886336227340901, -0.8373051552436173, 0.5903247290249255, -1.7060210002109955, -0.08494533486106046, -0.09155786224119905, -1.6551041666133515, -0.24156261582302874, -1.0807420169809403, -0.1364560703488655, 0.01924657578585117, -0.8487360399974262, -0.8943368381441787, 0.5855654882323104, -0.4700808977097314, 0.6976956486467251, 0.09554377221489849], [2.424124859289715, -2.0577280498444392, -0.7348318929554748, 0.0669486349007909, -1.782152595801556, -0.057066546479999855, 1.7374643878363838, 2.58465112772032, -0.527285950380558, 0.06340815093674432, 0.24054662611429561, 0.9815657577063183, -0.23186749388871286, -1.5209171813337368, 0.7450623419382897, 0.6763318006537178, -0.013987201726883001, -0.8225001428703966, 1.4483146133557638, -1.6178667899121562, 1.6125817725130798, 0.1420949267480898, 0.4714139408792385, 0.2920003226883046, -0.5367213006364278, 0.86948631701992, -0.6156907526218024, 0.04422016124784024, 1.1442232651635214, 0.32291977152110407, 1.721239573897935, 1.3335983024974876, -0.48350923037962956], [0.15828118141217742, 0.4546767963403526, 0.18973470409216833, -1.0217597130698246, -0.6532746080439249, -1.3889111631302027, -0.5105212093065897, 0.3178973346811513, -0.7003046583693199, -0.23094787204562528, 0.29958578882093545, 2.371517517908646, 0.21608466819635766, 0.2732875776025374, -0.9552094597237817, 2.2154489952754464, 0.15029399144773994, -0.5498597548633171, 1.3417717130789804, 1.1951930094638925, -0.19768220630750852, 0.07336170830303572, -0.28897902986775836, -0.21987583537399577, -0.6630728013177875, -0.09877853835980617, -2.215108403945119, 0.6079260143664029, -0.4632227358908815, 1.917198746561271, -0.6704699924302057, 0.26936286388528097, 1.824980021043638], [0.25469996378301246, -0.8692683102860979, -0.6239098885684432, -1.4429147427266957, -1.813770583774327, -1.636193499527376, -1.6535262017511647, -1.655307729168907, -0.7504335088181704, -1.220979329277082, 1.2643077276922545, 0.901935903438859, 0.13652426410470436, 0.8390541278916184, 0.07311228691200766, 1.3623190560099745, 0.7571315186705624, 0.18366207419669883, -1.038712704927411, 1.0461070294063728, 1.0204654713511945, -2.469067317508643, 0.5887582136455138, -0.8260471531989283, -0.18015349883332443, 2.052005423242608, 1.993762102385426, -0.10164539147961009, 2.5516026936221876, -0.5698731099093456, 0.10815087221962186, 0.09851682021041043, 0.8731234882974661], [-0.6458602441383268, 0.4294754424372442, 0.0012655174260254344, -1.6503231011841526, -0.5597664902168124, -0.17722921855410054, -0.15447495228816072, 0.7372011759667763, 0.47263647047773727, -0.11016064208602055, -1.624741030771671, 1.763766468974073, 0.5154044636171018, 1.4964133466378162, 0.9829864494883538, -0.10751792170005856, -1.3201498797631799, -1.331838448770617, 0.6268598102553273, 0.008142689078355336, 0.6668973009704623, 0.21551948320209555, -0.8362763123041761, -0.23811369531282606, -0.9465109906812686, -0.2834136173365799, 2.0257758909900265, 0.16463958089424566, 1.1632274725143699, 0.7761718612791615, -0.5354707947207118, -0.7711829523942276, 0.1406736773952149], [1.0474191815860432, -0.5633724911368144, -0.9998257281629955, -0.4984320811322518, 0.6351650984458882, -0.8814630735725756, -0.7833689379809143, 0.01369592267514481, 0.22551058489241238, -1.6046331305377386, -0.6739243077086866, 0.7707950105172949, 0.41434174643488153, 2.0369033218619608, -1.3712158239495011, 1.7904912637321488, 0.6886413400899629, 0.3233147366866604, -1.0109266638146597, -1.012820788920861, -1.4961624890003085, -0.5850652402083891, 0.7608574626970368, 0.09561257953167908, 0.21325808786300776, 1.1483694417957764, -1.9943353700514392, -1.8318155059466745, 0.4282241286664831, -0.4805360712687873, 1.2990952902359012, 0.12568292238510875, -0.13653572012389095], [1.2290938967648666, 1.107844040678581, -1.2601031657514208, -0.4360961303817667, 0.6173525795423591, -0.5756089715598145, -0.912432410432686, 0.08299164627904501, 0.23999816044764982, 0.4501867982068695, 0.3864778687232484, 0.8568089141662596, 0.4201124916609812, 0.22263175981498237, -0.16296637893674606, 0.6462727499073916, 0.889662604956298, 1.6390773940699477, -1.06982379045398, -0.1816502008699326, -1.052379146474454, -0.8195412698514873, -0.7296959173439739, 0.5360503177045354, -1.1918832950253082, 0.13337507128494186, -0.2830798836631551, 1.3323138921223459, 1.892750442494279, 0.13808913522965802, 0.766968813971879, 0.4682236616857068, -0.6170671340918237], [-0.616518050480722, 0.3707416634451356, 0.5889912025660174, -0.3203351671515253, -0.4201636694249066, 1.396865834332008, -1.2909030688787442, -0.44756251751189435, -0.6825727500106771, 0.6766997155466041, -0.20542925357022593, 1.6178213237822126, 0.7842614451810307, -0.10454355456169209, 1.6232807865672816, 0.48127367438581975, -0.8448790200155148, 0.5518795006906163, -0.5179592754361728, -0.4088028100660791, 1.0524532163241374, 1.4125160564714039, 0.5976202892610843, -2.0996287972119596, -0.6103675764825854, 0.6544393610569096, 0.6128202922884814, 0.9031965400881671, -2.3737886525817733, -0.9364596944233238, 0.7175590299995087, -1.2601006669073038, -0.19135536160218788], [-0.3019901467994136, -1.1550087627689742, 0.8257050645140025, -0.4631688728698733, 0.7037620445572461, 0.7791662819130616, 0.4686335701993368, -1.5631066080977158, -0.19632906833608288, -1.20880888731534, -1.5653059172137207, -0.7711104965472202, 0.4383649909714993, 1.032059251108301, 0.9072456742334569, -1.2573503839049758, -0.5300985978206776, 1.201368829888416, 1.8961830868079292, -0.031233862303579687, 1.2561137276855958, 1.7907862864918447, -0.39115693131042106, 1.4605034978617528, 0.4612839567539784, 0.7705521607608105, -0.4041854400616031, -0.45493238880618897, -0.5791138274983113, 0.19478556535765953, 0.4409077616445661, -0.2514047225526951, -1.2679977313057227], [0.6018492383165531, -0.9928700709818674, -1.0306370039501986, -2.504898819985593, 0.16395056036115974, 0.518273580071067, 0.48510071234103047, 1.6146829525406192, -1.9706742433581657, -0.7201162796648135, 1.1958993988561333, -1.5074453301771724, 0.208446563097675, -0.8098435070081967, -0.6127335395400271, 1.1713864190475107, -1.536488216999092, -0.5084233783180347, 0.573800218356771, -1.794326642572397, -0.789741815047863, 0.9072139791457806, 0.4538792215945493, -1.3256996473390714, -0.3127094426274398, 0.6336578815573628, -1.8939564393590405, 0.9102442874510123, -0.5313010488170695, 1.106835808516993, -0.9563655477427024, 1.1710560454333392, -0.045969156231697776], [-1.309360672531933, -0.2627807872499071, -0.6104799971452983, 0.15063489869345795, -0.6343426204689698, 1.9500436519396913, -0.09591071685132702, -0.573635527616759, 0.641235168569275, -0.1543304824158165, 0.8004073928614236, -1.0122361690292165, 0.4733566623789222, 0.7045118369761443, -0.07925756217034499, -0.25653339660490565, 0.6368283542073017, -0.11161720319157051, 2.0768926044360136, 1.2445577463432818, -0.14625576159859524, 0.062235192354476346, 1.1366287822552175, 0.0015266415105569008, -0.8469475535296822, -0.40750237883803436, -0.9192555831048332, 0.936187358274782, -0.027576166013613596, 0.633665944680109, 0.02341687783115202, -2.7030451101757986, -0.778703467746581], [2.585593320082809, -0.6982563796253166, 0.3167110898864558, 0.25620889174743705, 0.6649945789817407, -1.5631398680064235, 0.5683871531719529, -1.1090385031632106, -0.08142826585228607, 0.02077337575989919, 1.5537281325497359, 0.9817687148661224, 1.0295467364789148, 1.029505151904218, -0.606773884840364, -0.5969323045262911, -1.2462332142652512, -0.6362644025413166, -0.35599233676142095, 0.25471311609432834, -2.0134159304071533, 1.8921535364133686, -0.49565384980123256, -1.090216917433605, -0.43021537325846765, 1.0067142613226139, -1.6301604531522291, 0.565713674458791, 0.610117449190936, 1.6690039694716776, 1.6777869720105816, 0.3859935397911247, 0.37813874920823604], [0.727148914809699, -0.5499209436941259, -1.161259856117227, 0.7987930485193193, -0.5833351855653075, -1.022305064613862, -0.6986415234201693, -0.6183251211927205, -1.270719546624503, -0.041199258987136195, -0.3621126077637227, 0.30282480490245844, -0.21799256647275322, 0.6695803897746322, -0.4182616031456354, 1.8831616338324215, 0.0024902665473554107, 0.2071656226218365, -0.2566662153660363, 0.2995664168688906, 2.241173439421694, 0.5820441254648142, 1.339684611115866, 0.4468018235198888, 0.6852925210126992, 0.6025856690144062, 0.4487494014426376, -0.18064154439378477, -0.7384214292713446, 0.38128978503164906, -0.753788204991336, 0.8315790516056355, 0.4017805663783501], [1.8953488167372818, -1.1256137429151043, 1.6837364341023382, -1.402537103849362, 0.34920273168796817, 0.9844082786214087, -1.0422064367714687, 0.718102522261993, 0.6343914810880262, -2.365892057278868, -0.5194616495364589, -0.8460510522189459, -0.11841578104852815, 0.16795739740721816, -0.5401190241872876, 0.38579603566311804, -0.7450425343728937, 1.4633222185406396, -0.28576686023554604, 0.8444993825192197, -0.018487578984498924, -0.39817276297261733, 1.2612259617569503, -0.18077347397582558, 1.050232071739853, -1.342914037323189, -0.7947727416984522, 0.137885042440643, -0.5474578233156856, -0.5495423872172167, -0.7043347084082309, -0.5432143252697955, 0.5247145996398906], [0.4607600356944241, -0.3827974691120077, 0.336706677196896, -0.1944430080273119, 0.3221181664027985, -1.247659498215969, -1.357097926577916, 0.589124732229506, -0.3048576672570029, -0.43340803585047966, -1.8828188879976706, 1.00507530382275, -0.10355927237473617, -1.7442918758805328, 2.214414885861089, 1.2691790556027889, -0.18362387104156622, 0.7412283939735096, -0.0712849264130756, 0.2329317761925135, -0.33699882197109104, -0.920437927386934, 0.13609995943591316, -1.730795024893005, -0.08601980200898618, -1.5176506845764965, 0.08612513291982772, -0.7586501949525397, 0.12518695349241193, -0.5872823101795994, 0.02700990544626414, 0.4039741469475973, -0.4762729184854708], [0.9248050839092286, 1.0767853790953354, 0.016014467140098958, 0.7529917478057112, 0.9330394359201949, 1.2372541373090795, -1.7109612607814995, -1.644005413019705, -0.01784932755613562, -0.10429111535185148, -0.510762402822936, -0.06932723173258168, -0.23583707965632714, -1.3624144714600597, -0.6927679371690024, 0.11520172175217205, 0.4162948639628038, -1.5116969615532918, 1.0839605507555368, -0.2767807507586267, 0.301968121882131, 0.2339618420208407, 0.14071499282254799, -1.0254921530422683, -0.5566745599929677, -0.7433133539911652, -1.2303057200552963, 2.018116162468479, 0.24361962828849087, -0.2777153168727969, 2.708521288018422, -1.0177640956421137, 0.1798838362821318], [-0.5674883178900689, -0.7122279704672001, 0.5187117528587216, 0.9495163602833474, -0.3103374819753708, 1.2443005387588033, -0.2934005209152034, 1.2697800524289287, 1.1223960490950229, 0.039473727427344965, -0.9263989579768589, -0.9114673659636161, -0.05847003708235599, 0.6483249967067181, -1.2494848576612758, -0.2221366568384964, -0.2625066321239175, -0.6751414278863467, 0.14479813375878514, 2.5489157677217116, 0.7238567047831507, -0.6471941815701788, 0.014415181988391984, 1.3269586175601218, 0.5909377474285974, 0.9053544859813075, -0.8248121055409668, -1.067886709872721, 2.018807524462758, 0.9266786759762334, -0.40942224076608574, -0.244191049236937, -0.21901918469196086], [0.7525662848783455, 0.3007640787254965, -0.3839872590646576, 0.6111848090839865, -0.3095543259394971, -1.27856690830152, 0.11121187432480126, -1.462040537285202, 0.7027390214981993, 1.3240116954540746, 0.6856944526973499, 1.0645627497449819, 0.1788028469091856, -0.45784801120777274, -1.628965659213334, 0.2930330819330415, 1.1582512158446963, -0.8499224975133973, 0.48591289211347605, -1.6560027451065213, -1.019563716250691, -1.309214469588788, 0.06738579740906854, -1.021826176738002, -1.3529132881513013, -0.925468473480843, 0.7465015968350156, -0.7959674106432428, -0.19239580341203621, 2.2182600162469654, 0.09083217940090747, 0.9484282026137046, -0.45433177277228376], [0.8875762508898719, -0.7004957685921276, 0.9324580700571918, -0.17610141083006378, -1.6662060091137019, 0.6295108112027791, 0.8524162624146004, -0.7626824655696538, -0.6720006704264035, 0.22917397308659168, -1.541661292769876, 0.2666143427911581, -0.16982689581033547, -1.8286075374211768, -0.8237812414461999, 0.3958135038886779, -0.48753585067003613, 2.2868922688275837, 0.19625021561542205, -0.8989097130680286, -0.831019006237124, 1.0708515206550726, 0.192428636107656, 1.6948138177443415, 0.7654760991502103, -0.9501539286662322, -1.7802287726769637, 0.29913435188784465, -0.09034802547666863, 1.3416827985766917, -0.6449364117754112, -1.41408120977416, -0.10971473823545637], [-1.5168439160615248, 1.8158589160744592, -1.2737238470711885, 2.732730296561201, -0.4959958415510384, 1.5130507132006286, 0.8149382237234661, 0.18997974905326845, 1.9387875687156841, -0.9607711277164701, -1.0948260061266721, -0.4562271092475804, 1.6841463619111654, 0.7217596755771999, -1.4255493492214735, 0.020795416919687903, -0.7746020969592795, -0.056108025774657455, -0.8899246137290763, 1.133220868855304, 1.1211936905751563, 0.42279319155426415, 0.7742397036679807, -1.4327256402702722, 1.9904983094377577, -0.9272066259425062, 1.2692462750883184, 1.809531053915948, 0.5424932117034644, 0.6955298469737857, 0.23424317120530044, 1.68925478589919, -0.31508917245523854], [1.9543367554747684, 0.647384643920268, 0.8171817106934013, 1.0444774192749744, 0.7850711273244785, -0.7413378419776598, -1.1852861362866174, -1.370213460943296, 0.10419065505910663, 1.2064906525563213, -0.2647148582089821, -1.0358851812861192, 0.03051543486451495, 0.8829911643504218, -0.7876043419926217, 0.08622229571246542, 0.2706941445464281, 0.1819481501984991, -0.03739513717674145, 0.3706331710070257, 0.33530592168653206, 0.6806384597180317, -0.7911744521605723, -0.7310625987758355, -1.165776580431568, -1.1423905600972666, 0.2991087611431728, -0.2643162971402625, -0.882822199363337, -0.516733210653849, 1.64453756303801, 0.05221957748729904, -0.04714724951283961], [-0.040337397317870664, -2.5395005620448345, -0.3511094652369483, -0.8771426084527687, 1.537807186729314, 0.24658427931329027, -0.0430051135685586, 0.8061696734739736, -0.6841749539579569, -0.6697416785397673, -0.04722312248707703, 0.1921120518091983, -0.39737518441057357, 0.29312494963285746, 1.216558903949619, -0.9957153604648467, 0.3866694145094916, 0.09837778201001071, -0.9207702276240856, 2.0024447116068553, -1.9263542715096664, -0.910864128651696, 1.7076098970311708, 0.7859308065456844, -1.1526449252623132, -1.2166440664762281, -0.9507104336271389, 0.5737097112536609, 1.390312822081186, 0.3797126167981325, 0.14398788664966122, -0.7716718877585234, -1.4064931784127261], [-1.5489629758807553, -0.35597632507832905, -0.6145719641033343, -0.28616421632125416, -0.48929079228821926, -1.792308610641631, -1.072941721008388, -1.326428532589712, 0.39612875276662896, 0.6052359982573042, 1.6095328538516824, 0.4319647346013911, -0.4991788966287252, -1.0332312600375304, 0.4977769757910534, -0.16776920348751484, 0.08376569724497247, 0.8262848084110642, -1.509552906097504, -0.9800776270139737, 1.1031819865672814, 0.603089444863038, 1.295334579104453, -0.6292047957263366, 0.03770267848462141, 1.6465768041241104, 1.198817170333842, 0.767478505934231, 0.310761942077447, 1.0955963411847525, 0.75086528355277, -0.7201503340229437, 0.8434816231881453], [0.14405952764879623, -0.2769007158833734, 0.5744980357657726, 0.8418216024361523, -0.8378956476483826, 0.05747787314644454, -1.0637723772467764, -0.9538869681163533, -0.2246172911318381, -1.4179888398726268, 1.0254529553087106, -0.4183087747316908, -0.6437456109342159, -0.2991833476351894, -0.611337068626623, 1.100669595875057, -1.3286535926165617, -0.8026008487682667, -0.24138759654327363, 0.6012076870926035, -0.9192645897506296, 1.185803091757265, 0.4454788679759251, -1.0895750499653487, 0.35679626725252167, 2.797692700688354, -2.339238457729132, 0.9981749759553485, -0.043727703053862, -0.3401096681903651, 0.006099773874422041, 1.0514335265945394, -0.22412277007897816], [-0.8926217786078103, -0.6560303350539332, 1.5219906451176946, -0.6705654886252416, -0.4880880773759162, -0.03098999632605924, 0.6092853110135678, -0.4809504499156947, -0.4178073003684377, -0.9495880452682257, -0.8245744197360441, -1.2257683879709194, -0.44252151827268316, -0.18110085291387124, -0.46688057995794296, -0.3075728297211508, -1.6530666584215739, 0.9181870239460506, -0.6870086034818195, -0.39431462191541145, -0.32260311331223507, -0.28029894689077767, -1.8532812370199463, -0.03036695373556515, 0.08919792038589668, -0.052709265705512716, -0.7008042661765826, -1.3112420579117845, 1.1483539975845545, -1.07403966741994, 0.12205486611177539, -2.6119989774074113, -0.11472444564684829], [0.9374096768107922, -0.7526612888891635, -1.6527742315000327, -0.24681050637191354, 0.41236720221909107, -1.0222377853239768, -0.9871032197970401, 0.9165762608694048, 2.6944042386316003, -1.7112526288004541, 0.7826192392798526, -0.03832231577804053, 1.3187399263868322, 1.5088907225671229, 0.28857320771365463, 0.04798969359520662, 0.23697005730771267, -1.270306195191877, 0.08715402633204872, -0.3077708288801561, 1.3099558181238007, -1.4176929366687463, -0.12457197910996864, -0.20877813885129304, 0.9545716812110783, -1.3892003991187682, 0.05345318458698971, -0.1677944891384727, -1.674402360301741, 1.0977977743740404, -0.055426569664788344, 1.523992610463968, 0.3415410317551333], [0.5520231538689215, 0.7664459482097885, 0.6063347262625605, 0.2465532513347509, -1.1741800288764115, -0.4190113202892622, -1.417561021488402, 0.19839848458852044, 0.52503938704429, 0.27918958780715764, -2.7054778450434425, 0.4196065590517392, -1.4990329645591358, -0.4978387116089464, -1.4639014067378648, 0.2927451982141157, -2.3023770068414025, 0.12410229907161052, -0.3374581159728249, -1.4590870597331111, 0.399601594809199, 0.9902586345059418, 0.04941559765667704, 0.522124245918224, 2.163175271417502, -0.4724213955508632, 0.17808630609251616, -0.08531882690214737, -0.9992349816683587, 0.48268709528685805, 1.0708750686844641, 1.2794344393078998, -1.6222441957027776], [0.12754187014325738, 2.2086445325211117, 1.2621475629802574, 0.9134071116587861, 0.12185038010938509, 1.4597757602629793, -0.08029283981674279, 1.0011375691213968, 1.2445226306532635, -0.5419775860840962, 1.3311485838547021, 0.837842382982009, 0.4652882408208582, -0.70472409150045, 1.7384551970064117, -0.9428182798219845, 1.3212764261160457, -0.04921347151931463, -0.11068554694676991, -0.027323576074941024, 1.1355249187077703, 0.8157552732801526, -0.6276695825736446, 0.7355823268484449, 0.26601884240911294, 0.7188890712420369, 0.9379846012961494, -0.8260346051024529, 0.3461617640153806, 0.8179154225090091, -0.147289026792162, -1.2601554208241383, 0.01809678333803274], [-0.06375786097998147, -0.8390700348835936, -1.3245280854597166, -0.4679003824024528, -0.6139815851894986, -1.289838342027645, 1.4387528438497472, -0.9000851365829051, -0.33211963602587086, 0.06819792119739375, -0.18615087829454477, -0.7222779694880593, -0.6483676351065557, 0.8904256102837863, -0.20312893681227343, 0.38367236423461426, 0.6518047375875223, -1.191173109313937, 0.7524288440680994, 1.136936374647746, 2.249105398983923, -0.23515858159763534, 0.9936944796377606, -0.010227412661511165, 1.3357336668076976, -0.3814871198241611, -0.6635150132319877, -0.4788071320511866, 0.5888694528063445, -0.5118466787500822, 0.4891667669636512, -0.5347683432727965, -0.8248880669030172], [-1.1304455282411252, -1.2286654400088093, -0.5806288354269539, 0.2220632980915186, 1.271396980163033, -0.42309251795622016, -0.22536296856709426, -0.05146497904436276, 1.3656083593914523, -1.044550345136785, 0.5628488247323555, 0.3271887946270686, 0.24313991334184018, -0.7582144385871367, 1.509720613242719, 1.144138381087701, 0.6328526190992081, -1.5024915532201293, 1.2537838155557557, -0.5240066860879065, 0.8923890878308385, -1.142547913697989, -1.0338204264216677, -0.141679375897209, 0.20961163926893675, 0.34103365521183693, -0.805038266672717, 0.9034971107741879, -0.09748346153725423, 1.0037939079879983, 2.0623856975608796, 0.7723440962555211, -0.5571119138320387], [0.6227150402726764, -0.9100118434221568, 1.1063647120651288, -1.6696027483343518, -1.3440496969040183, -0.5604863409395231, 1.5408538849542048, 0.4795657026073415, 0.22041160417821173, -0.9341396784398609, 0.3702468043008141, 1.186042194673424, 1.571725887619264, 0.688338207252726, 0.3000498672035875, 1.822699897486137, 1.5249512329938444, 0.5444060436487851, -0.7348492558281796, 0.06240241915957827, -0.6632126873018073, 1.7803526775213954, -1.1424218669787694, -0.1776790386472152, 0.907352034052068, 0.19566265997832552, 0.0618577971320697, -1.0598869018623365, 0.9390642102888942, 1.0224081510491438, 0.5859682269084047, 1.1158289225914493, 0.6888517885215606], [-0.7428320739847264, -0.29400730036103534, 0.2937624435149127, -0.33474286481329224, -0.26655702539610066, -0.19855515010554636, -1.2297819883431398, -0.056214043393229884, 0.9752186556413777, 0.03606564256422318, -1.4980920766173995, 0.26409700460292607, 0.1770955887724215, -0.4878558705332571, 0.17885468194798707, -0.16605094173322124, -1.6246346991714384, 0.04581956795505132, -0.44932987389933204, 0.7734320845251811, -1.8774606814287298, -1.1485103140113198, -0.7060360229930224, -1.0797162496155337, 1.1310396767517261, 0.7675019727765372, 0.19287964714032635, -0.781741627959355, 1.4992668249940928, 0.0276390012973259, 0.3452822903589055, 1.4158850578202862, 0.4932576116170105], [-0.8581733599878674, 0.24060742371946894, 0.027797259573077587, 0.44068026064347676, 1.2410978490312738, 0.9923563293807862, -0.9525886815125113, 1.3348221348049776, -0.4438950876324884, -1.9510358382460657, -0.537300118301392, 0.5130589751571805, -0.17930160314203328, -0.602107610657398, 1.7997525482135, 0.6587580918082151, 0.6952194845505386, 0.587944096382174, 0.48667570856847747, 1.1058500403380604, 3.588684128164277, 0.7504373309661289, -0.25329860201070814, -1.0882575778202093, -1.2675690465835407, -0.24228417601065205, 0.12811670521671312, 1.1340371244589005, -0.6902099639493352, -0.24452653847414138, -1.4835772645302918, 0.6671040647884331, -1.354999506865279]]}, \"b\": {\"shape\": [33, 33], \"truncated\": false, \"values\": [[-0.07483235454237296, -0.0818218919817741, -0.7935416868324545, -0.9710936362283307, 0.8561508906926462, 1.2953086470177129, 0.23225915368403532, 1.5969971209584786, 0.8869862784227109, -0.6619593862325588, -0.5103138968603929, -0.6303346950505029, -0.5602464318434077, -1.2897745567084922, 0.10504861660236597, 0.6088448641256743, -0.6389231291185852, -0.3805134702781167, 0.3582411981755579, 0.08894330901889816, 0.11023705179773677, 0.12754136909153876, 0.27917225907024773, -0.033832222225957775, -1.0343685162424017, -0.00978890223592246, -0.5807202539487225, -2.213482923993315, -0.7111049688450147, -0.035033112837271306, -1.0227716920272767, 0.763033542531783, 0.7696230727380865], [-0.7836648886138587, -0.8441556734766444, -0.29157250127050555, -1.8330768201030336, -0.5942146732199788, -0.3047066628200817, -0.512433508059187, 0.5156470373510614, 1.031604341777868, -0.16261673802574952, -0.7034328188895436, -0.550029038408011, 0.5425024692262436, -0.47991896549879776, -0.5385456422281114, 0.578569099020054, -0.3531914601228287, 0.41228481356085867, -2.053956307485005, -0.27138378066624447, 0.8275377019735158, 0.041973150825935834, 0.3697407849632848, 0.22623599918621334, 1.104642723320442, 1.0472699439326096, -0.0008797599790089461, 0.4137474291003376, -0.3755602763388962, -2.205436889462535, 0.6329280838760162, -0.15545184655438127, 0.6985752988740442], [0.3406227071469347, 1.4463475491484536, -2.5392699951460713, -1.0498751334132856, -0.5029728894206151, 1.0543488102603948, -0.5666330819727194, 0.18452449235204718, -0.012084168785318358, -1.2734557943964, -0.4673585691538282, 1.0064336376925977, -0.8412140716101162, -0.06320618227837921, 0.5605564489366386, 0.21166963240966702, 0.3745929440545413, -0.16876353606887384, 0.9296839372987517, -0.26142452990919524, -0.5321667935090543, -0.17636566244163665, -1.122109908370447, 0.03017413070635436, -0.6388880326678548, -0.23344719965385416, 0.9654689726949286, -0.5950364039390987, 0.6283606369554883, -0.8981628425876134, -0.3654810986743329, 0.4374392206097443, -1.7604540926446575], [0.7129459163171992, 1.2103115134198914, 0.7654544200295162, -0.9687690566639245, 0.23263748238727439, 0.2890480487937045, -0.1989474947545353, -0.5867745195516114, 1.9085273902692497, -0.148644577557412, -0.8637059416249993, 0.3270607202033256, 0.5273279424389372, 0.030059655288149526, -2.468957089731159, -0.07346219763234976, 0.4782640161241025, -1.402641166026416, -0.9263006150715659, -0.9899199203120455, -2.0928359412271837, 0.7605339367859557, -1.7896902581473533, -0.5903166427803308, 1.302974222762018, 0.3891644298841686, 1.3925693702436612, -1.6585999888667966, -0.8994773372620484, 1.3717128708679611, -0.12362692001653419, -0.11451505232314145, 0.30692958061497766], [0.07931892314681079, 1.6232142932277118, -1.090379864363699, 1.0245538845014037, 0.5696197270532789, 0.025098608784477887, 0.5153806751898232, 0.9059803471331453, -0.11991914117215809, 1.9953443594254947, 1.8420851347089455, 1.6057324817366179, -0.5865507081866521, 1.1821921992618818, -1.6386701144060318, 0.538455947045966, -0.23502366825557414, 0.27600220706419526, 1.4383234630551254, -1.7119680919804514, 0.548017397422855, -0.15553527999209033, 2.48045952765816, 0.7594959748389879, -0.29358323593859753, -0.3350255607253568, -1.3016226963876387, -0.017167737936016503, 0.0801962208021078, -0.6721351566091042, 1.3143732838513358, 1.168616623683571, -0.36447020235632405], [-1.0291787563460009, -2.0868875077308946, 0.35625266252966686, -0.16895539518430927, 1.3254892837126346, -2.1271912918840106, -1.5209383254061528, -0.43116064906069723, 0.42639788745095586, -1.2876453356369004, 1.9536712835701853, 1.5060592562289794, -0.2994431928554976, -1.3935696028387576, -1.3493823072695503, 0.46544677051106853, -0.251977715514646, 0.2986269929439797, -0.4202151154141806, -0.48132714317054565, -0.022780180279449922, -0.07839162821093898, -0.46996753499167077, 0.20432250210315045, -0.38923586675878996, 0.019820628430937497, 0.11870029432677731, -0.09078867098566966, 0.5480077335542941, 0.8826251256257921, 0.22433524940920477, 0.4422916460478181, -0.5179872151533877], [1.4789832580853794, -0.9612363749251033, 0.5998659145669827, 1.4919926126139218, -2.5190059490448387, -0.6106143262496031, 1.0519116025421817, 0.037458377720569494, -0.8201979708617955, 0.19428504689639084, 0.1900020900165003, -1.1801853393781228, -0.17945719842826996, -0.1957100240080421, 0.5194340349007248, 0.5317543845856555, -1.71489624048642, -1.0612064815828963, -0.8172614029637898, 0.8597252955205753, 0.5835870843632908, -0.33228763045417403, -0.00847752742987874, -1.5005221072038575, 0.8260207645821855, -0.2294331369918814, 0.2707654786499221, -1.732409868929778, -2.4784380599914195, 0.519526378523438, 1.5411538259576705, 0.22196414026258052, 1.5258581467302796], [-1.2908677200256538, -0.3796014044185053, -1.3485918522780178, -0.406467279958206, -0.5804632684146168, -1.6809252298092934, -0.4755175832702584, 0.6199285350334878, 1.412651252187678, 0.49141110958919404, -0.1129002011584416, 0.9490432627944841, -0.5907648622067031, -1.519527699874343, -0.8292653542312239, -0.47556253440487545, -1.0268366935598727, -0.3160527040486611, 0.40231921840461027, 0.061789972344806425, -0.5412375079240669, 1.5862760158710518, 0.6401272157795647, 0.2067384845301319, 0.9330716890621804, 1.9286327628721425, -0.192269554552158, -1.725870152770647, -0.03857214338963035, 1.793018411021461, -0.7559004020846755, 0.628437648823874, 0.8789089596543188], [1.1284938962620843, -0.8157328956102361, -0.7690363444700574, 0.11616930285633252, -1.4470183648951438, 0.08330863524881632, 1.3082857471029319, 1.0043465871265218, 0.49076352475718654, -0.27812247123672884, 0.9275854669400203, 0.3059520635790701, -1.0447216174044103, 1.9420108882114422, -1.2133717329547764, 0.6394678798803897, 1.4598421772436698, -1.039685081407858, 0.16390913965465265, 1.7355250239149644, 0.09504797758500975, -0.4550124413979754, 0.22387345462057512, 0.4131527124726539, 0.3579203694254046, -0.04429239386701592, -1.3227589865716052, -0.4223366830623341, 0.3553792455922079, 0.6538957058949499, 0.21995251813050118, 1.6512898400369724, 1.658995809180784], [-0.5402960263765969, 0.02177192869312365, -0.23529765549159692, -0.4816720320498833, -1.0600989995851076, -0.09118687311886069, -0.4158394244717688, -0.26601453004915154, 0.6699558668681361, -0.8659168829619214, 1.583453780338715, -0.5221965011841629, -1.2520468952538346, -1.0754196353570586, 1.5435603482733662, -0.6808315302237075, 1.2291971117834568, 0.7097512482046644, -0.36198443467838937, -0.3293687501862838, -1.8833522674680538, 0.6215100175021587, -1.002601723884583, -0.6724878528459699, 0.9918403260706862, 1.9484234539587224, 0.21430628677371985, 0.15772591413795883, -0.9036142843407274, -0.8896372159422593, 0.6036336438171023, -0.3422422166941913, 0.3264144244090985], [-0.12978868417583603, 0.9393382786591656, -0.4285332948584352, -3.0809325252745245, -0.4908532952122323, 0.24353317642017713, 0.445105254460519, -0.5003737973553734, -1.0399351631605895, -0.35645999683440743, -0.06392669057348561, -0.9625071721470271, -0.26907215806110935, 1.1174260285676352, 0.9497146749287936, 0.7751755874403174, 0.39259654723923715, 1.4004753895711717, 0.43449045202625763, 0.6113682762754901, 0.14588730829664262, -0.16569112399634106, 0.12507129234826653, -0.6260620564464323, 1.2520480960830638, 0.4140392390871062, 1.224595547155823, -0.17136332984690222, -0.42586284415834935, 1.1132159876916783, -1.5051533478255006, -0.8939106844660724, 0.08978977355734566], [0.8778305002530795, 0.06266774791167196, -0.6257412127975734, -1.7133956682746838, -2.216238663971115, -0.8788280281161002, -1.6844549351963987, -0.9515915539518947, -0.016021938341850523, -1.2249137485879702, 0.5553577745644822, 0.3480274719491204, -1.2580218482662895, -0.3124695036111332, 0.8997983976302891, -1.0803608193956455, 0.9244674443898508, 0.1432159082774759, 1.1020619308422652, -1.4814354329272463, -0.5402017679322737, -1.4238037599668003, -0.2488288735534823, 0.9918607223757433, -1.6980014823200391, 0.043417631085191664, -1.0545544630042434, 1.5398138475333303, 1.9033839922102036, -0.511384607464535, -0.991659949086372, 1.4023428432088552, 0.36134958660236244], [-1.6916777308309532, 0.383271009524246, -1.0625823745193739, 0.5906619795766875, 1.119845881129846, -0.5591216818198134, -0.3647780655213838, -0.055593217198522046, 0.023528796369898533, -1.1634621911894343, 0.6086211696174952, 1.242204554677818, 1.2884519743254343, -0.684115729605091, -1.4113796475391596, 0.2728081085451476, 1.243616796748014, 0.26016067136403453, -0.7615568032597841, 1.3191834761565728, -0.47936761075735773, -0.04479652746174671, 0.43802976781453606, -0.4930410546596817, -0.4012913045411972, -0.48432559779124335, -1.0843641353657651, -0.6906514675884551, 1.1159725030503453, -0.4546732259064923, 2.167887049257035, 1.405788809748897, 0.33082002009189376], [-0.6399382336840662, 0.8221251609002174, -1.4218367726201808, -0.17221765192345412, -0.4827825404961051, 0.1308677779173294, 0.2084701936103003, -0.621733343176822, -0.8504992817121269, -0.14262530088416328, 0.2512955626232276, -1.0320071537284317, -0.03666955273135014, 0.6215551589376913, 1.6610809585907624, -0.006948430777327355, 0.1218977135448095, 1.775765166235498, -1.3681984430567622, 0.016325489478408527, 1.2423053175400824, -0.13625874029965726, -0.18566803590018635, 1.0161703077976783, -0.48967584989792995, 0.15093462145474495, 0.9456253397258091, -1.296485309307345, -0.324927864212485, -1.4144060880402323, -1.4213441849297053, 0.9253701949515383, -0.19624956286210452], [-0.7353480540551117, 0.5889046371923171, -1.287293640934617, -0.2331285812361381, -1.1291615234369803, -0.4542201975496317, 0.24473350648355183, 0.0933972144630859, 1.6934292897327, -0.8369579009156178, 0.4844777134570532, -1.3782385767485095, 1.2043060809200232, 0.6211811605427119, 0.9164265590674083, -0.10729133858964873, 0.3995217950335715, -0.17287106598480012, -0.5729944203079494, -1.1456575210329543, -1.080426902110817, 0.2418470929860758, 0.7319520648728054, 1.5757335556224392, -0.4984001820216582, 0.9941244787376275, -1.0313190342328056, 1.4606960079960614, -0.18452404273271705, -0.5056751482047533, 1.9276721034411577, 0.6179250570122987, 0.16385131299021444], [0.8863527324548172, -0.34342161399294796, -0.09490005824491812, 0.3822582198947298, 1.9990592283065474, 1.101431820016636, 0.23339016491845344, -0.1355501812684996, 0.5909065630436978, -0.07810974089803717, -0.21241654121291756, -0.8347073242548612, 0.6093403297531014, -0.786701717206692, -0.3480917892639549, 0.8713056664110936, -1.0474710578458364, 1.5017001469336877, 0.6282428793459366, 0.5289342244290455, -1.0487662210941608, -1.0089718498898672, 0.35821926966834194, 1.3030371143347688, -1.2091806941079826, -0.5420853996972036, -0.6864352552834531, 0.2576330411277016, 2.091581448636091, -1.2700258209932607, -2.5324843317680257, 2.112413624500102, -0.9664996084529615], [1.4725099389613097, 1.3541163812413488, 1.61161317401283, 0.0050733806162459715, -0.4769765397627714, 0.4988772169729922, 0.1562597521827973, -1.3217336142609926, 1.0210998324473268, 0.42955293417928264, -0.8026881444379483, 1.7402643353732523, 0.48001062413350415, -0.2970546314424407, 1.7767262389864682, 0.20138839208689832, -0.2994290850740613, -0.45064278722967127, -0.7504957555626104, 1.7953646050677075, -0.6736564333941646, -0.9806485263785693, 0.6009253020304886, -1.1859267553904491, -0.1843668651824083, -1.0996292877385971, -0.030703961209194993, -1.2659080945240302, -0.8415534677231704, 0.032333976680263, -0.1808246458803584, 0.45948888437429863, 1.3446922205554064], [1.3951891046841847, -0.9614791366525781, -1.9053046911621758, 2.3825440730176464, -0.9176816166943018, 0.03382448385123731, -0.8409617844585153, 0.21724913485271988, -1.8900686358888983, 0.6618072544485217, -1.5402990688396914, -1.1362184824010375, -1.2055158864408044, 0.40080400582676845, -0.9928995408170669, -1.0425915994542598, -0.3139647846775428, -0.6554075337399632, 1.3212982173788137, -1.953031240076931, 0.042073994597950984, -0.48648708514238737, -0.5526301156174038, 0.8968092360447815, 0.4739157327852433, 0.9244697151989478, 0.716520654962315, 1.0045307527160718, 1.2483003609872934, -0.6213729639454517, -0.6464303307041362, -1.3107758211429485, -1.155161998114943], [-1.1392397507192717, -0.8387418109397434, -0.7753718094172652, -0.15076243700788383, -0.8411813410147664, 0.4542992110836568, 0.18853144411957312, -0.0709133717076783, 1.5519829042014786, 0.5226703040096146, 1.3084248377253294, -0.25945106851615835, -0.046541061270621846, 0.017464817918956498, 1.117630950003374, -1.2868244596268936, -0.08758413255957471, 1.390123211848914, -0.3897695841737716, -1.0959870854444427, -0.4537674793690715, 0.3911942118562052, -0.5526908821880998, -0.31772612296521163, 1.7507076735400362, 0.8224816136645269, -0.34847138363941876, -0.06154306572112132, 1.4945445488119073, 0.1674801518196847, 1.819374974215514, -0.7920917965230101, -1.2311153637996954], [-0.8655702363034342, 0.8004799486356303, 0.5665318087874819, 0.145395966901056, 0.6045797748185597, 0.9219366076493639, -0.07971179792962263, -2.407232666367152, 0.8908887671115661, 0.17641741170904643, 0.38997703782967175, 1.2939543842529768, -1.3445260146392608, -1.419793229270162, -0.6453356201405096, -0.5657165322899605, 1.0779461500997607, -1.7156739607948128, -0.30379373989668895, 0.30066809112689247, 0.3338225428173482, -0.0737232314512341, 0.9829785626536136, 0.24481761926783388, -0.5681842653636401, -0.1221487377364702, -0.15874933997963048, -0.2740190043670028, 0.2154632988707264, 1.4421220365480707, 0.2485124083938476, -0.6262835245307712, 0.14731883038892335], [-0.8012312315604347, -0.05180613084775089, -0.9957616665915425, 0.8933911092876785, 0.046550244654527374, 1.2959802505004727, 0.2136999828613242, 0.8022489035492546, 0.28717271038384656, -0.25364821324482467, -0.24666675636336374, -0.9314051717874073, -1.2858045845285748, 0.7000188427296913, 0.6602411282536481, -1.5612961231086622, 0.14768519786252354, -1.7227465110078728, 0.41834256236749173, -1.3231132024073617, -1.5340597098294022, 0.197155312235981, 0.11247411118040121, -0.04601157723194831, -1.7443447042902636, -0.6359042038004133, 0.4838023806378361, -2.2333924711120936, 0.5047263802043016, 0.9810257496898854, -0.7694114318918822, -0.06258070599860072, 0.5862131800958873], [-2.284613253354679, 2.24647927964823, -1.4584237598306398, -0.36497517992158607, 0.3277908954797056, -1.223953976716361, 1.169554142754974, -0.7278032827148361, -0.04211195128373977, -0.08498801929196897, 1.40773199136336, -0.07364383482684234, 1.387809500339442, -0.11902148866425019, 1.3847092850348373, -0.09047939578911202, -0.7247398621379565, 1.433828430721718, -1.5736495989205745, -1.6827791712044986, 0.15034391266640124, -0.5699377568844317, -0.21319132908665428, 0.14082800957942404, -0.023034982590695453, -0.7961089648030919, -1.2075719007703347, 0.9751498603245721, 0.21542377875830912, -0.8440680944516712, -1.5974002546430335, -0.9605130249080008, -0.33623996507392473], [1.5720040442267365, 1.1313997730620617, -1.0918448787629038, 1.0573303270293757, -1.9986244151167045, 1.9632311411765593, 1.205136712697065, 0.35524378483691105, 0.5657036108655569, -0.36994394731156693, 0.2118002400898013, -0.8189260950680816, -1.5106954616965955, -0.11376317469459209, -0.05271571514640277, 1.1760969929597047, -0.9302304387516012, -1.6875543696106707, -1.5008616352355546, 1.0465540185673483, 1.935631101748327, 1.744089211322113, 1.0338386807537447, 0.07912852245893191, -1.5680251432337708, 0.49463420144195164, 0.2582797349348785, 0.4437839592089884, 2.9736453603425645, 0.7142443632071779, -0.029667711466703592, -0.07045459867726206, -1.668591489928309], [-1.825738804708136, -0.20928083001776182, 0.08379942546629734, -0.9947121976261558, -2.7983345690000836, 0.5923276070763966, 0.4762977609686943, -0.7653387212839206, 1.8160689724787509, -1.4505549295788334, 0.4476150776155769, 0.8002859134593144, 0.5994650794979298, 0.5810331995084516, 0.8388483566696605, 0.6338168054145441, 0.05188749223038297, 0.46978701216016194, 0.6736326293547912, 0.0017313800502386568, 0.1396758474752417, -0.4277328177803785, 0.9078249571368148, 0.4643347669273009, 1.0047667966014848, -0.9628233157676889, -0.6186546288288534, -0.18158553686054157, -1.2625998976921864, -0.618340653648418, 0.2351700067599433, 0.3846997168559226, 0.7991415842524527], [-1.3670727677570798, -0.11857318862511433, -1.546195778352691, 0.7063282259008217, -0.8387980310719249, -0.3121835001425729, 1.1236886921231835, -0.1579036291193209, -0.032331871596959064, -0.8508978054349292, 1.1230734236392672, 0.8694528676366677, 0.08799040416999131, 0.6012208856345357, 1.3722071250456929, 2.3598862563917398, -0.8056488098101875, -0.6043516148369612, -1.6140541391189114, 1.0498749779179524, -0.19938011799020727, -1.1868214797482914, -0.3595177140762201, -0.3245272405454665, 0.2570845274661971, -1.371005150650357, -0.3762244693180108, -0.3479683630327635, -0.5564319927588331, -0.3159157570069575, -0.1274508409612201, -0.19112336734942406, -0.6098430518436285], [-0.03417212539651545, -1.611956699732807, 0.29839741070556364, 0.49154160084869986, 0.4604430549950527, 0.6150079487766489, 0.9340559700267766, -1.85344866606792, 0.1617433859412663, -0.922934912184165, 0.8341150444918889, -1.3374677845713248, -0.599156208924033, -0.118993134373346, 0.5881220057307589, 0.4523732381548191, -0.4125087812056498, 1.1850827795984133, -0.9883745558086681, 1.7435086064361933, 1.0186848663484065, 0.950599739963088, 1.962326813541374, 0.5480325652162259, 1.4297929407044718, 0.2272551110712657, -1.7758575087007022, 0.38266517149415163, -0.7091660474803414, -0.3787219540227604, -1.917008753241169, -0.8772647267513993, 1.4458257080027297], [0.6627878090130102, -0.8622453245256245, -0.6003797899527903, 0.9925853940591883, -0.07247260184038536, 0.2807041977016741, 0.4969450283822336, -0.4904609418270982, -0.14015191643789465, 1.0195882981989681, -0.03514240266294441, 0.2363891145401504, 0.16198657518906906, 1.9015134918435208, 1.5591903772281916, -0.14125876966370293, 0.8999150713802109, 0.486229485034689, 0.6494549615633087, -0.016883138007788408, 0.09901601626720236, -0.5362918988958474, 0.7944444620073492, -0.25392455716509843, -1.2310746421691736, -0.8211403623796307, -2.2045058670744133, -0.01810959743934909, 2.1303074445907253, -0.6270850081516626, -0.46821559355429715, 1.300959660055022, 0.6931765983384196], [0.2619781215358181, 0.09384924540729124, 1.3602435685913894, -1.8991422845804025, 0.6021851005454494, 2.1175544851187387, -0.7560914281490503, -1.2253331692075384, -1.5783116576194465, -1.5074458743638397, 0.7377994835158869, 1.1474702574798354, 0.0873672585338384, -0.3925111372157923, -0.4621673403172461, -1.0288913733294294, 0.09640319425432174, 0.13641117423869373, -2.0922111491561766, 0.44616404793710385, 1.3229100528098034, 0.5332015016967651, 0.6561147595094544, -2.196109624036218, -0.3819168878445796, 0.2628825495746469, -0.6966489695359506, -0.9296144512699136, -0.46102330633544697, 1.8099135025639788, -0.4125410970742381, -0.14605377919410076, 0.639525662486428], [-0.353089534915011, 0.050218911925948205, -0.7822883962615191, -1.9804553800204088, 0.8251715548278555, -0.6440749121853679, 0.11844669398447112, 0.8654695869298157, -0.42026196519846387, 0.13229030691963145, -0.022452846433411767, -0.3910493623027078, -1.5206645537355274, -1.2285203243459206, 0.5748315429709654, -0.7452605810311763, -0.2326136973903766, -0.4162541041359512, 0.7339643741431489, -1.3451732603219617, -0.941606481196887, 0.2277373973304034, 2.434364262142867, 0.5600151994273596, 1.8441670948888182, -0.6332537585664547, 0.8599980170417958, 0.5713534047935209, 0.7027133899815589, -0.1881275553049068, 1.0264528896644531, -1.983880631317786, 0.20855334577683513], [-1.2897880601928444, 0.4969928018902153, 0.7659148641189563, -0.919663127568596, 1.0736883148639542, 0.8545993769918717, 0.3780639473389147, 0.10625119038207999, 0.18698336679258054, 0.18452851834624237, 0.18396230341827796, -1.0787887331387913, 2.2401069135177245, -0.8628161262114218, 1.5028723725311774, -2.442480789527209, 0.19271087517251353, 0.32959974562469224, -0.8911671634832217, 1.065370251770907, 1.3262576857020933, -0.37382338377326896, 0.47761131887274233, 1.229329563949083, 0.03720579656491025, 0.19280406251832147, -0.94697543643724, -0.43517456620231804, -0.7099063548688473, 0.2043532544495121, 0.7229170467636317, -0.09128919332142722, 0.30671642732778426], [-0.85714982142524, 1.166398613421453, 0.48775394481063505, -0.13886677319389, -0.6600383066500718, 0.9029695014824611, 0.07427856031894989, 0.17942884220774877, 0.3136181949277942, 1.0361388561394387, 0.3540815261548012, 1.4260379054317078, -0.7869210585056919, 0.022408990182032098, 1.0690672895099986, 1.2733836496049118, 0.5770437266166565, -1.3098919595143017, -0.42647921866613514, 1.724278266150499, -1.925785791099792, 0.4476333318686097, 0.04159902506147449, -0.03137087406433594, -0.6253209757198802, -1.0799310768835757, 2.724981743921066, 0.9178955123585951, 0.017773011082040584, -0.555321443715121, 0.31576979015318146, -1.6856752277353773, -0.1413273336198211], [-0.09800199580853954, -1.7147702478405884, 1.0471146879568116, -1.6446654915951926, -0.744348471336984, 0.41984862394573824, -0.6376467642300864, -1.381118132649884, 0.5906683769820318, -0.6294897341034307, 0.20597820848501122, -0.40360768260943813, 0.10178221019816133, -0.3606833867154866, -0.6625561247738918, 2.8036065829770176, -0.3178235625242121, -0.42333616502457005, -0.10284104573688058, 1.7654035763064484, -0.07664736858184522, 0.13441640171256003, 1.5114704880483079, 0.9320128273059292, 1.5996937335536572, 1.1256501221218385, -0.6877900667366624, 0.6520484111106634, -1.85640538761479, -0.40352580978477254, 1.9494216888263285, -0.14912186835429422, 0.39771370736393563], [0.21344690481284462, 1.735041420818353, 1.0900652800053543, -0.367762176251474, -0.5720224235332447, -0.3689741891256407, -0.6378304745594062, -1.0396458004665465, 1.2513416397199013, 0.5745821931002821, -0.5370212699515421, 1.8893185023597503, 1.894173726997504, -0.43969060098155033, 0.2198851718635022, 2.2529762620202978, 1.3057699175343978, -0.11002860512966929, 2.1947533842780684, -0.5295446140047319, -0.4575163626661797, -1.6446271296895891, -1.246873664706361, -0.014599278109414825, 1.36415245963347, 0.9453094934285363, -1.175727504844197, 0.030440411809441088, -0.7898461457340464, -0.04668934176492838, 1.1177111494939667, 0.9138549743424085, 0.13727671669746389]]}}", "witness_dims": "{\"M\": 33, \"N\": 33, \"K\": 33}", "numerically_observable": true}
{"family": "reduction", "bug_class": "reduction_init", "correct_name": "row_max", "buggy_name": "row_sum_init_one", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_reduce(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), 1.0 / BLOCK, dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        acc += v\n    tl.store(out_ptr + row, tl.sum(acc, axis=0))", "why_standard_misses": "a constant offset of 1.0 on a sum of ~256 uniform values is a 0.4% relative error, inside rtol=1e-2", "witness_inputs": "{\"x\": {\"shape\": [2, 32], \"truncated\": false, \"values\": [[1.3640195228050165e-05, -6.271425885369812e-05, 0.0002491574809406205, 4.958468028771155e-05, 0.001182665742342288, 0.0014049343623371935, 5.121767282302068e-06, -0.00027448581677559354, 0.00045090785195024214, -0.0010611754068181793, 0.0002702059525341101, -0.001000725175431826, 0.0005041299711134938, -0.0003497537729459638, 0.0004967331047528937, -0.00047277363991025516, -7.922958692114135e-05, 0.0013191584031693128, 0.0003477229170723744, -7.730866072443808e-05, -4.958810116292692e-05, 6.485567860690553e-05, -0.002025719620739827, 0.0004039208180734784, 0.00044457628328337806, 0.0009811785846381032, 0.0010578945588044664, -0.0018627104973384374, 0.0008895246143854409, 0.00032740331792915683, -0.0004647056323105564, 0.0005614830119226252], [0.0012756537197246719, 0.0007158818858830026, 0.0004933970166185707, 0.0006022845828675233, -0.0004302227042122171, 0.0009995145255388249, 0.0009254095175517151, -0.0007178369390394483, -0.0007755243754113115, -0.00011557759572248035, -0.0016417642269099734, 0.0016325640307554702, -0.0002199943346906009, -0.001030192348086365, -0.0006461228569510426, -0.000692532961738709, 0.00015531876272188036, -0.0003304799701007843, -0.0006666306986131324, 0.0014200960263577014, -0.0014494440734879918, 0.0011756114084950967, 0.0018483454305422832, 0.001789578369698995, -0.0017074980697834332, -0.0012779575959290205, -6.48978779081474e-05, 0.0006457781353429509, -0.0010309056457177574, 0.0009208684300705492, -0.00030750395847272587, 0.0009297243955343462]]}}", "witness_dims": "{\"M\": 2, \"N\": 32}", "numerically_observable": true}
{"family": "reduction", "bug_class": "reduction_init", "correct_name": "row_max", "buggy_name": "row_max_init_zero", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), 0.0, dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "why_standard_misses": "torch.rand() is non-negative, so max(0, row) == max(row) always -- the identity is never exercised", "witness_inputs": "{\"x\": {\"shape\": [2, 32], \"truncated\": false, \"values\": [[-2.0217323059579484, -1.1205637636917152, -2.4247685045640193, -1.3973503794981748, -1.0511792194528577, -2.486817219473164, -2.486142150540828, -1.566083548064621, -1.5546883794496305, -1.303046759826813, -1.4304403953391798, -1.2822758667527667, -2.1291706144278866, -2.7110253839185936, -1.294009285080363, -3.1339114562511314, -1.5766551375058877, -1.4499467547478797, -1.7094037478637614, -2.1437643346652493, -2.7100458066200233, -1.698761531933239, -2.99509693477206, -1.777741274600238, -3.1047936013272235, -2.3049636196778547, -1.0933945724315703, -1.0752201571891646, -1.2676891660465932, -1.4146107860999728, -1.407063513780138, -1.0324062101766636], [-1.61025512968625, -1.7852003100816285, -1.371095688674866, -2.2324970247754026, -1.5462081747220422, -2.665862883109307, -1.1493544092193535, -1.1906553036190992, -1.5528986201304757, -2.1076910529255386, -1.9343039493562955, -1.8581897683705608, -1.934123505302985, -1.8163791946381114, -1.7391517859774557, -1.0479485477850976, -1.7482854674162127, -1.455742074216712, -1.0671163891755486, -2.922170470854413, -1.4489532197286126, -2.239162174849211, -1.412313933496983, -1.6100899198125727, -1.4934579112430093, -2.086065903614463, -2.7749586395748223, -1.3072493309869202, -1.39073549385943, -1.0741024461317035, -1.1889274637642353, -1.4368708771941097]]}}", "witness_dims": "{\"M\": 2, \"N\": 32}", "numerically_observable": true}
{"family": "matmul", "bug_class": "reduction_init", "correct_name": "matmul_fp16", "buggy_name": "matmul_acc_init_one", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_M = 32\n# litmus: constexpr BLOCK_N = 32\n# litmus: constexpr BLOCK_K = 32\n# litmus: buffer a_ptr = M*K\n# litmus: buffer b_ptr = K*N\n# litmus: buffer c_ptr = M*N\n# litmus: grid = cdiv(M, BLOCK_M), cdiv(N, BLOCK_N)\n\n@triton.jit\ndef matmul(a_ptr, b_ptr, c_ptr, M, N, K,\n           BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr):\n    pm = tl.program_id(0)\n    pn = tl.program_id(1)\n    offs_m = pm * BLOCK_M + tl.arange(0, BLOCK_M)\n    offs_n = pn * BLOCK_N + tl.arange(0, BLOCK_N)\n    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)\n    for k in range(0, tl.cdiv(K, BLOCK_K)):\n        offs_k = k * BLOCK_K + tl.arange(0, BLOCK_K)\n        a = tl.load(a_ptr + offs_m[:, None] * K + offs_k[None, :],\n                    mask=(offs_m[:, None] < M) & (offs_k[None, :] < K), other=0.0)\n        b = tl.load(b_ptr + offs_k[:, None] * N + offs_n[None, :],\n                    mask=(offs_k[:, None] < K) & (offs_n[None, :] < N), other=0.0)\n        acc += tl.dot(a, b)\n    tl.store(c_ptr + offs_m[:, None] * N + offs_n[None, :], acc,\n             mask=(offs_m[:, None] < M) & (offs_n[None, :] < N))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_M = 32\n# litmus: constexpr BLOCK_N = 32\n# litmus: constexpr BLOCK_K = 32\n# litmus: buffer a_ptr = M*K\n# litmus: buffer b_ptr = K*N\n# litmus: buffer c_ptr = M*N\n# litmus: grid = cdiv(M, BLOCK_M), cdiv(N, BLOCK_N)\n\n@triton.jit\ndef matmul(a_ptr, b_ptr, c_ptr, M, N, K,\n           BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr):\n    pm = tl.program_id(0)\n    pn = tl.program_id(1)\n    offs_m = pm * BLOCK_M + tl.arange(0, BLOCK_M)\n    offs_n = pn * BLOCK_N + tl.arange(0, BLOCK_N)\n    acc = tl.full((BLOCK_M, BLOCK_N), 1.0, dtype=tl.float32)\n    for k in range(0, tl.cdiv(K, BLOCK_K)):\n        offs_k = k * BLOCK_K + tl.arange(0, BLOCK_K)\n        a = tl.load(a_ptr + offs_m[:, None] * K + offs_k[None, :],\n                    mask=(offs_m[:, None] < M) & (offs_k[None, :] < K), other=0.0)\n        b = tl.load(b_ptr + offs_k[:, None] * N + offs_n[None, :],\n                    mask=(offs_k[:, None] < K) & (offs_n[None, :] < N), other=0.0)\n        acc += tl.dot(a, b)\n    tl.store(c_ptr + offs_m[:, None] * N + offs_n[None, :], acc,\n             mask=(offs_m[:, None] < M) & (offs_n[None, :] < N))", "why_standard_misses": "K=512 with uniform inputs gives entries of order 128, so a +1.0 offset is under the 1e-2 relative tolerance", "witness_inputs": "{\"a\": {\"shape\": [4, 4], \"truncated\": false, \"values\": [[0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0]]}, \"b\": {\"shape\": [4, 4], \"truncated\": false, \"values\": [[0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0]]}}", "witness_dims": "{\"M\": 4, \"N\": 4, \"K\": 4}", "numerically_observable": true}
{"family": "softmax", "bug_class": "no_max_subtract", "correct_name": "softmax", "buggy_name": "softmax_nomaxsub", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n\n@triton.jit\ndef softmax(x_ptr, out_ptr, M, N, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    mx = float('-inf')\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        mx = tl.maximum(mx, tl.max(v, axis=0))\n    denom = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        e = tl.where(mask, tl.exp(v - mx), 0.0)\n        denom += tl.sum(e, axis=0)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        tl.store(out_ptr + row * N + cols, tl.exp(v - mx) / denom, mask=mask)", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n\n@triton.jit\ndef softmax(x_ptr, out_ptr, M, N, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    mx = float('-inf')\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        mx = tl.maximum(mx, tl.max(v, axis=0))\n    denom = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        e = tl.where(mask, tl.exp(v), 0.0)\n        denom += tl.sum(e, axis=0)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=float('-inf'))\n        tl.store(out_ptr + row * N + cols, tl.exp(v) / denom, mask=mask)", "why_standard_misses": "torch.rand() lives in [0,1), where exp(x) never approaches the fp32 range limit", "witness_inputs": "{\"x\": {\"shape\": [2, 32], \"truncated\": false, \"values\": [[200.0, -1.6317065930677275, 1.3217959639836427, 0.08268651793293419, -1.176789075556173, 1.15634948500117, 0.9812034560390803, -1.450232629943399, 1.1097434560609962, -0.6118264608379209, -0.9788509639537006, 0.7335103590365222, -0.7359528732349158, -1.4723083533772292, -0.604586815202591, -0.44243931825931104, -0.041550302195256426, -0.813399947059634, 0.19988849963101607, 0.3058523569251492, -0.04195743276306992, -0.04419802690076879, -2.587726367372583, 0.19537124370042144, -1.0253621593455575, -1.6537885696206223, -0.1875999459546322, -1.7198022770090309, 0.12154065984870106, 1.7510917005767161, 1.158081947244671, -0.3811254992638406], [-1.3765545085718065, 0.14027541418961761, -0.3941151914000429, 0.817545064242547, -0.8611085439572549, -0.7492967467774145, -0.6231931514664533, 1.128681662043641, 1.0538378248948594, -0.25387131621034104, 0.7305169397227304, -0.5579763257127764, 0.24420166211985114, -1.0391281322064165, 0.16542835332157446, -0.3082462580767681, 0.2759840359971707, -0.14531797911144112, -0.38163557791523345, 1.1977392915072063, -0.6649148066985354, -1.5714543754661177, 0.36372506345541805, -0.917267325744919, -0.3717683584339326, 0.3854861868459564, -0.5968691433875065, -0.6578391856405483, -0.020393888025675006, 1.2847599735410689, -1.542826915713378, 0.587250268186104]]}}", "witness_dims": "{\"M\": 2, \"N\": 32}", "numerically_observable": true}
{"family": "attention", "bug_class": "no_max_subtract", "correct_name": "attention", "buggy_name": "attention_nomaxsub", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_D = 32\n# litmus: buffer q_ptr = M*D\n# litmus: buffer k_ptr = L*D\n# litmus: buffer v_ptr = L*D\n# litmus: buffer out_ptr = M*D\n# litmus: grid = M\n\n@triton.jit\ndef attention(q_ptr, k_ptr, v_ptr, out_ptr, M, L, D, BLOCK_D: tl.constexpr):\n    i = tl.program_id(0)\n    ds = tl.arange(0, BLOCK_D)\n    dmask = ds < D\n    q = tl.load(q_ptr + i * D + ds, mask=dmask, other=0.0)\n    mx = float('-inf')\n    for j in range(0, L):\n        kv = tl.load(k_ptr + j * D + ds, mask=dmask, other=0.0)\n        mx = tl.maximum(mx, tl.sum(q * kv, axis=0) * 0.17677669529663687)\n    denom = 0.0\n    acc = tl.zeros((BLOCK_D,), dtype=tl.float32)\n    for j in range(0, L):\n        kv = tl.load(k_ptr + j * D + ds, mask=dmask, other=0.0)\n        s = tl.sum(q * kv, axis=0) * 0.17677669529663687\n        p = tl.exp(s - mx)\n        denom += p\n        vv = tl.load(v_ptr + j * D + ds, mask=dmask, other=0.0)\n        acc += p * vv\n    tl.store(out_ptr + i * D + ds, acc / denom, mask=dmask)", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_D = 32\n# litmus: buffer q_ptr = M*D\n# litmus: buffer k_ptr = L*D\n# litmus: buffer v_ptr = L*D\n# litmus: buffer out_ptr = M*D\n# litmus: grid = M\n\n@triton.jit\ndef attention(q_ptr, k_ptr, v_ptr, out_ptr, M, L, D, BLOCK_D: tl.constexpr):\n    i = tl.program_id(0)\n    ds = tl.arange(0, BLOCK_D)\n    dmask = ds < D\n    q = tl.load(q_ptr + i * D + ds, mask=dmask, other=0.0)\n    mx = float('-inf')\n    for j in range(0, L):\n        kv = tl.load(k_ptr + j * D + ds, mask=dmask, other=0.0)\n        mx = tl.maximum(mx, tl.sum(q * kv, axis=0) * 0.17677669529663687)\n    denom = 0.0\n    acc = tl.zeros((BLOCK_D,), dtype=tl.float32)\n    for j in range(0, L):\n        kv = tl.load(k_ptr + j * D + ds, mask=dmask, other=0.0)\n        s = tl.sum(q * kv, axis=0) * 0.17677669529663687\n        p = tl.exp(s)\n        denom += p\n        vv = tl.load(v_ptr + j * D + ds, mask=dmask, other=0.0)\n        acc += p * vv\n    tl.store(out_ptr + i * D + ds, acc / denom, mask=dmask)", "why_standard_misses": "unit-scale q and k keep the scores near zero", "witness_inputs": "{\"q\": {\"shape\": [4, 32], \"truncated\": false, \"values\": [[-24.986394524353937, -58.8677056935254, 136.8091412817628, 98.4846584556703, -141.90936613026366, 57.92291122033178, -26.18228618310603, -18.036587790902377, 16.136338992120105, -92.00511661049477, -84.77698096431314, -47.58644104490596, -55.3344933411368, -110.75195710686633, 58.72403365363463, -7.678707495321565, 19.621273694943955, 68.61785854032756, 13.265282652687244, -81.6651539972112, -37.51340573530533, 73.98368329339534, -3.6012334248103977, -98.59129066102571, -70.94465505225645, -91.02448884967129, 24.48513289579856, -59.89963206838889, -9.491301838151886, -56.28462612711182, 33.057364662579545, -78.87498386481066], [0.7603457924110417, -1.4235649196649909, 1.8251697438591643, -0.8484181808137538, 0.15072182086931216, -1.5644236303945342, -0.8872151351855841, -0.9517765934917211, 0.3809132015775415, 0.6497887065077564, -0.9206247750795987, -0.5557406210302869, -0.18672496566134883, 0.24104470608242015, 0.21181402277050884, -1.1164477437517717, -1.1091258820582612, -1.4024939110803962, -1.226473905585101, -0.6310819590899202, -0.5437140997268878, 1.259438518637766, -0.559208642033918, -1.1688648951064953, -1.4420963945910792, 0.6925850421105483, -0.6804225027339043, -0.8779043128765279, -0.7346977969042491, -0.8639369071373086, -0.03779598533986895, -0.25537141711837286], [0.2649708084090205, -2.1373291567952792, 0.5861359612274133, -1.421073087743797, 0.3269590328175011, 1.0659562497135184, -0.8039736964573706, -0.43793415973643435, -1.24881004660128, -0.3650068388623692, -0.1289422089856872, -1.107195682512445, 0.8471980861748758, 0.933797953510037, 0.8928231187242696, -1.3381356505669022, -1.6159780778425195, 0.26088210640947745, 0.26971229850529943, -2.0994800111132763, 0.30426089868344725, -0.463839033238975, 0.5088637300342912, 0.9603122688492125, -0.593680581277238, 1.420657955072825, -0.42040762732254267, 1.1906091350418941, -1.0543299450538701, 1.650660731035215, -0.32546366952533695, -1.0474784038906861], [0.40118640950618467, 0.22770673668621158, -0.35056472262059707, 0.8373914634399333, 0.3668320651859748, 0.770682266703516, -0.6243363568069182, 0.22592741851337003, 0.7142243671762141, -0.29046887466067045, -0.4764145301799316, 2.520489614671167, -0.6298934185846036, 0.0969678632429039, 1.0484526758690385, -0.710399711098437, 0.7448473023136876, 0.44612570829093595, -0.7045211449892591, 1.3782220632370192, 0.90215614226089, -0.6514550608835349, 0.6206756709506865, -1.9652702657469363, 1.676988985645428, 0.27967347470774023, 0.11349995690894138, 1.189430201701724, 0.21280167329496474, 0.4873671206023092, -2.2812129594394195, 3.636292685855558]]}, \"k\": {\"shape\": [8, 32], \"truncated\": false, \"values\": [[0.4777419556066451, 0.4042890083836756, 0.04773548056489078, -1.402599737484425, 0.5553528516280092, 0.6772586253464523, 1.5710473919454524, -0.33452706984948644, 0.8466651078026631, -0.4449443729325096, -0.04014177977747255, -0.4431682452993693, -0.10765313709431074, 0.3775775571675236, -1.7667690591576213, -0.5331934329694791, 0.7227549675496802, -0.31878111257044744, 1.2609294764636598, -1.2327362330674099, 0.5240962473831902, -1.4331261861709734, -0.15137432045554577, -0.39802347509515723, 1.0668490244540043, -0.07999522317559259, -0.8829967365176751, -0.658834901240982, 0.3640542630021909, -1.2714810512411474, -0.5352552266430427, 0.048536848487985465], [-1.4392217937862857, 0.6802006022124918, 0.39393401095936664, -0.9408639600006308, 1.2434210491691648, 0.11578727618242052, -1.024996216835058, 1.6785474810818035, -0.6106414046512388, -1.2999984850490414, -0.26420758532475286, -0.415265763365291, -0.5535552404460754, -0.20841650074337004, 1.5651573190662456, 2.086334238083272, -0.474083967654543, -1.084661526373959, 0.6560042827071383, -1.3871901718516624, -0.24283430416131804, -0.5108138951518846, 1.3304444073456776, -1.690224965386932, 0.7669747824009857, 2.020321486026368, -0.6403550804885751, 0.07848739824758677, 1.2346342311189071, -1.1215764018128271, 2.864029198131766, 0.185371503485548], [-1.2170691240063904, 1.270200011791255, 0.2302809786796516, -0.247185970471007, -1.3966270823858395, -0.10745742969224735, -0.748890764486357, -0.1226930069499708, -0.3155083368316857, 1.1531968223971576, -0.3845168338764116, -1.2145451731615091, -0.3265745082915637, -0.297107802451031, -1.2612832594560903, -1.386283498817448, 0.5306670728656099, -0.8155271533061873, 0.4890005627294077, -0.9502798472274838, 0.7301645309977686, 0.6224800745915897, -1.2867086283199212, 0.5138341606214654, -2.552093806823091, -1.0080191723570076, -0.39055292781233625, -0.741886018488434, -0.8529472130830144, -0.2518307511493278, -0.4976086481580281, -0.7105414981113436], [-0.23199134518757833, 1.58406962925026, -1.2075263631967672, -2.086120111770107, -0.4665614825807965, -0.7821058364427632, -0.6104153500093171, -0.11824147374893895, -0.06659066399625767, -1.9914657062718975, 1.121508811465096, -1.0762581165197216, 0.5846522720846818, -0.3741773358979624, 0.5960900193699876, -1.1486848350835923, 0.16519354545207207, 0.2462072651451211, -0.3036373124596814, -0.37011176880107854, -0.612294542776691, -1.0790505111418114, -1.1251990926149618, 0.17761980694924792, 1.0662762360261357, 1.446964176857466, -0.928845296726794, 0.9673015190774826, -0.9105497721268476, -0.3358871813514904, 2.1364877825833184, -1.6860375606859468], [-0.6877931325016207, -0.7668355093032386, 0.6847281915495345, 0.17871132094583678, -0.9648397449080639, -0.3078455722060709, 0.8165521568367407, 1.22543975980128, -0.15497323848934097, -0.4571330342691272, -1.1802147076928964, 2.450421342500583, -0.9304727504862892, 1.0580080642672225, 0.04380563917916904, -1.0006324707494672, 1.1546462850149444, 1.6904103407583042, -0.20750828229544827, 0.24972925447563038, -0.38541661237172586, 1.3598311000717385, 0.28369760428812646, -0.4216304215161149, -0.390584519077707, 0.8710717625721719, 1.0143009051646135, 0.7743014555068429, 0.06627517394649306, 1.191560985578381, -0.8085981624553943, -1.283573606104171], [-0.8900772387605761, -1.2302478946267006, -0.10779671540260102, -0.44838637620014665, 0.26967416328636146, -1.3164717555132495, -0.09251526907130718, -0.32478637897954665, 1.3168642211446964, 1.8046367534475531, -1.3792822644187785, 1.4099500607492563, -0.3603420392943861, -0.423783077258572, 0.9231418525109198, -0.7329148726111155, 0.6575500606730241, 0.7157212735502372, -2.3013052481022362, -0.8417132466963989, 0.9257218197719388, -0.9852517725332616, -2.541569988066157, 0.660984217472775, 1.2162830035534065, -0.15962622075822094, 0.35108933678952436, 0.18291744250990447, 1.286143760374327, -0.7297051853488841, 1.6461359012022563, 2.2988507544636], [0.457606840957094, -1.5172136407439718, -1.28360503563569, -0.4624412062004748, 1.220961127831674, -0.33205002855876703, 0.37605021082859313, 0.1842997770462265, -0.2721791722475282, 0.8738702762899054, 1.037668078466411, 0.6621091313754143, 0.6079967624369228, 0.233802881874962, 0.5058030616907334, 0.6805202709654355, -1.1651098871058663, 0.2801285474439017, -1.2705865118839685, -0.05280297226467606, -0.35729309265559034, 0.5281089412902853, -0.3599562446280008, -0.20734872960897693, -0.33504065200972055, -0.4054985328257788, 0.7165321215193505, -0.22148409212264614, 0.25783144748571074, -0.9952954318984449, 2.15884556545617, -0.22598100847870456], [-1.2251816564130225, -1.31614462426175, -1.1268457916840322, -0.8383861326915769, 0.8680647544269458, 0.5565860474135593, 1.0370603205959381, 1.021127708683925, 0.8303778522612429, -0.43589211800235095, -1.4308491099808471, 0.639161248568881, -0.6489076187097301, 0.8100740214658577, 1.5191187111498572, 0.7737469069614498, 1.1675900973147646, 0.3918526917394072, 0.508635885948293, -0.06777679011679204, -0.655069525923777, -1.709993462379751, -0.5875608159909218, 2.1728926643319255, -0.15782936664329084, 1.1836913026748335, 2.1417562645096653, -0.5287324189142008, -0.18670352607646992, 1.1372276147934381, 0.07905602332850292, -1.0546948167582941]]}, \"v\": {\"shape\": [8, 32], \"truncated\": false, \"values\": [[-0.9871834738970214, -0.8353664223222018, 1.2568116022687381, -1.3195644131868456, -0.08708693554974885, -0.36795096877090483, 0.23873776693923549, -1.3891903872381737, -0.5296779404636476, -0.23832051681026423, -0.9749775762632797, -2.019731962961505, 1.3643927976372212, -0.3005655519229637, 0.04692059007780266, -1.3994623465451406, -0.26452791613590815, 0.6529610488531583, -1.1123169926677172, 1.5891707724599657, 1.5850990996063903, 1.49442886700177, -0.370205325994101, -0.47986388437929345, 0.8980539617189371, 0.11776391601792285, -0.10326196329481263, 0.6083436318160872, 0.2856179469619062, 0.42489009949628626, 0.07993575284043423, 0.9715880578621124], [-1.1986918048969228, 0.2178502789780811, -0.20464048000583546, -0.13741424588059534, 0.7500787568813186, -0.6386207947331053, 1.7244932091190477, 0.29552258441013846, 0.9166737428542806, 0.9927860350337918, 1.1747698824887267, 0.18505649138817523, -1.3998766301833554, 0.00451838096427443, 1.6128970513368481, -0.4242866472894959, 0.20670578601752582, -0.7756296695568882, -0.5935417949501957, 1.2714137995729191, -1.1022036127028885, -0.6587032757584876, -0.1155695321111165, -0.7174421226298585, -0.898008879959124, 1.6107221236381615, 0.5508072599521848, 1.3101017575414085, -0.23519542756337028, 0.1919095865606655, -0.3286743464674651, -0.2229314701586826], [0.871068191925532, -0.11547090310742421, -0.9314105126350516, 1.0836465635869388, 2.5132927352085064, -0.8446754421730703, -0.4711068593558731, -0.2654471281155153, -0.18076391862519944, 0.2745198218315799, 0.3141273244725492, -0.2698559539235096, 0.41502153311996265, -0.4369726449493081, -2.0696389983762926, 0.44440214596642613, 0.14286303561757874, -0.6132393438218997, 0.5757802278319596, -1.67998503896808, 1.1028212547261307, -0.4530651954317401, -0.2372658352638346, 1.2288308791713443, -1.7052127632164724, 0.2856008228316298, 1.3022092845940048, 0.034983730318305435, -0.3688451097749284, -0.01948036062310023, -0.5155289888642857, 1.0174087742979008], [0.6825537076831522, 0.7583351570205299, 0.2750346381467359, -0.262131079962963, 0.22455743923027635, 1.0355943123160056, -0.04403588160441519, 1.2363354520633418, 0.5643066264511769, 1.2354788287331375, 0.4407563960261554, -0.3911827513207477, -0.4488581391664718, -1.0263158989758694, 0.8478239868162505, -2.1126500498761938, 0.8236233022742843, 1.2095840092825945, 0.4811762043881711, -0.8571031004012034, -0.15487321188888165, -0.03868710358509449, 0.5081937004824254, 0.9312044959891596, 1.5092369305251723, 0.9349249110161736, 0.39118845130880525, 1.3455254724152041, -0.3685849287493525, 0.5364809806140909, -0.4127503450515353, -0.3963260529404407], [0.0864394848004402, -0.781546274239926, -0.4826761417303327, 0.41371033990382244, -2.2239097884482817, -0.2520476552198445, -0.13188241941887102, -0.8132156205073131, 1.2755224837879704, -1.0766347361499102, -0.5762956459322262, 0.4029510206202444, -0.3227964232519255, -1.1307970334380144, 0.3362641280059089, 0.39621063875723095, -0.31262662044108275, -0.0006027574781092043, -1.1714150446631024, 0.5553173548387074, -0.26300725224560423, 1.1448696701308214, 0.6942440334499579, 0.03550335470115636, 2.3119086546857934, -0.2762049987422564, -0.06360889250948695, -1.6661431594292735, -0.5374136237191696, 0.3844016512296301, 2.2064152870228777, -0.5736742859826971], [-0.857770933490045, -0.18983739638068223, 0.6227820996336637, 2.0373716456791486, 3.0834733848485283, 0.5907824294400266, -0.27914845191586934, 0.2851128979362026, 0.6586692682568324, -0.2981139097940719, 1.357729168307185, -0.8036293162960297, 0.03055558021814941, -1.7724054105115061, -1.7445276055672283, 0.7530287745032027, 0.7434243053315246, -0.6123115437863614, 0.026951138610223666, 1.4827758166018623, -0.6930812003655916, -0.18804797895065184, 0.24485871140851256, -0.282358936685658, -0.2222373554926694, -1.1650130720803562, -1.3618942319492875, -0.39966198462682373, 0.599979440431969, 2.8449464655935524, 0.26324602224726734, 0.9928265308616021], [1.5225359201787156, -1.1572650226930903, 0.9210369477307075, -0.9464334345231947, -0.8318289808207194, -0.438146839105129, -0.044039984801569856, 0.9038658611141022, -1.424446594106753, -0.7093306211795107, -0.09063497929323405, -0.24329834038317788, 0.7235950542165019, 0.21206206520140308, -0.924371864029957, -0.5967211909151534, -0.41399886494038723, -1.6821932834041846, 0.3479542834808341, 1.2471016294648472, -1.4984121179324597, 0.1641126022785863, -0.4092388622019256, -1.4426116631272192, 0.4223444689546868, -0.16204785679401176, -0.1627081608634382, 3.1600393765755164, -0.31669857438322085, 2.1321581234157603, 0.3435896038779786, 1.1451405615754295], [-0.5891965141714894, 0.14053127519022385, -0.6742063852330489, -0.1839616670877003, -1.1943775763224338, -0.3859205951537836, 1.4306697800155763, 1.0347478781151422, -0.3693646716643041, -0.6047768648148218, 0.29178955388914746, -1.6760604598443096, 0.7558866940130254, -0.5176802388615706, 0.3720540644527708, -0.41538237304297126, 0.28684392766185857, 1.057987971489257, 0.9747300306155051, -0.8987738972121889, 0.12603534237467212, 0.44565480999464746, -0.2962286501499274, -1.102244343800442, 1.2304366585319988, -1.2227791886365296, 0.36637584936449175, 0.15593799547305698, -1.7834074293411806, 0.4350024318829489, 0.7612688469355076, 1.118936499061391]]}}", "witness_dims": "{\"M\": 4, \"L\": 8, \"D\": 32}", "numerically_observable": true}
{"family": "reduction", "bug_class": "race", "correct_name": "row_max", "buggy_name": "row_sum_race", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = 1\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_sum_race(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.zeros((BLOCK,), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        acc += v\n    prev = tl.load(out_ptr)\n    tl.store(out_ptr, prev + tl.sum(acc, axis=0))", "why_standard_misses": "on real hardware the lost update is intermittent; a single allclose run sees it only sometimes", "witness_inputs": "{\"x\": {\"shape\": [4, 32], \"truncated\": false, \"values\": [[0.9323224351204542, -0.13484240850688187, -1.0977711087115172, 1.2093136358526926, -1.5200609774258935, -0.3659055784186133, 0.08753656217963342, -0.673525657000616, -0.31220063548687593, 0.6517512317274256, 0.10188336399475949, 2.1183280668541165, -0.1747104908892027, -0.05302204609851407, 1.1776554100520484, 1.034069668664329, 1.3570641388632136, -0.26345967846302504, 0.6563534008927115, -1.0510361039263771, -0.2951569873791522, -0.6036145420186865, 0.4275727371703606, 1.164091172456542, -0.18349346919422196, -0.06950722813263917, 0.9665602467906304, -0.6179201120759003, 1.307118331059199, -1.809250277245433, -0.24609517424533933, 0.047437465889070246], [-0.1726226634799063, -0.4265367270080289, -0.8613653826306371, -1.884530122726135, -0.842549971820494, 0.4946735479336724, 0.4465705367621375, 1.85655320464552, 0.7426474381606197, 1.6626073177772, -0.7445255020708731, -0.8749235310133816, -0.1406877389606945, 0.4921162579956519, -0.23537017889100362, -0.2329928411829563, 1.0203585485236448, -0.3950961781815714, 1.586702826547813, 0.7790573295349854, 0.6084367506154511, -0.7746609271150169, 0.5145944304272745, 0.30077846461329444, 1.0300494046122595, -0.38908207485805274, 1.1903487622827287, -2.3898260878140896, -0.22780056268989937, 1.780801496873732, 0.052726437928369584, -0.2905865934347286], [1.5681129649560095, -0.34023019162555135, -0.590480507527205, 0.3952876666332603, 0.41516876928919283, 0.3747274728808734, -0.6381902366431598, -0.4137973997801163, -0.38841359754271754, 0.9259711547147788, -0.5889466811888183, -0.131217053696824, 0.00859743441700256, 0.12076421285922352, 0.23244373318601652, 0.12115874783464392, 0.7007774085680416, -1.1021614558814128, -0.20761115023330756, 1.0637356081910359, 0.4318807786996173, 0.5185060470598175, -0.12765925263970465, 0.0023738622684158404, 0.32685676241280304, 1.574304297758674, -0.497854771468547, 1.1708647780992363, -1.1308104177840286, -0.5076025950103698, 0.3658925302392737, 0.11861084275632296], [-0.7123668701917418, 0.15577888425550526, 1.376028723925904, -0.8521657544163167, -0.9913974466747576, -0.6171660403968796, 0.36085432621749663, 0.725788747445546, 0.7048039780615126, 0.856549025221355, 1.1007333725162887, -0.5368462420915631, 0.4897982425989597, 1.9345120072465691, 0.6114975628959227, 2.545833547596705, 0.8446100794186416, -0.4976219513230361, -0.6579083416586656, 0.15435453835917065, -1.2352837415993694, -1.5907987411963338, -1.3607153546827384, 0.4951250366915605, -1.5295864071475327, 0.02331219452117253, 1.5065421776698205, -1.0374513272097987, 0.6412381170410911, 1.4182025956774233, 0.4869832493833102, -1.187126614866286]]}, \"w\": {\"shape\": [32], \"truncated\": false, \"values\": [-0.5977878136525421, 0.5499482947031736, -1.282288774354227, -2.0776904842464647, 2.061194735710628, -0.5525172695521284, 0.5098670391746202, -1.363457812071284, -0.19746682414381686, 1.1420645413492911, 0.6255783450063146, -0.061071397736976014, -0.8579834503967044, 1.983500446602315, -0.06843040976465128, -1.954830554634629, 1.0475056154390079, -1.611744070743091, -0.17456089960628418, -0.015990131632859814, -0.9816276242759144, 0.18854014565014418, -0.4928272722487404, 0.5059976051141007, 0.8652126974567892, 0.4340038332999275, 0.2040632519731441, -0.5270238182528804, 1.0758099641593462, -0.6375573748121397, 1.716919815122407, 0.08713222254118819]}, \"b\": {\"shape\": [32], \"truncated\": false, \"values\": [2.0350573599285773, -0.16991243315971818, -0.20766799832413915, 0.42743737204579724, 0.6971226418809447, 1.4539900715632597, 0.1849208603294952, 0.701146603857423, 0.6025765208327771, -0.8487943832250671, 0.4448961180210981, -1.4012743322129773, -0.6757868015068886, -0.223349621952516, 1.38410819746911, 0.04124791968879348, 1.10614562019248, -0.10658197180618174, 0.8128409688341192, -1.3263767409501788, 0.32066904725723755, -0.5203186214933955, 0.6040014181673045, -0.5102548016749502, -1.4045173362952923, -1.155640456085217, 0.6852226947571901, -0.5076807514417746, 0.2555611847109906, -1.8056237695142763, 0.33995212125650787, -0.012663061533244438]}}", "witness_dims": "{\"M\": 4, \"N\": 32}", "numerically_observable": true}
{"family": "layernorm", "bug_class": "race", "correct_name": "layernorm", "buggy_name": "layernorm_race_stats", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer w_ptr = N\n# litmus: buffer b_ptr = N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef layernorm(x_ptr, w_ptr, b_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    s1 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        s1 += tl.sum(v, axis=0)\n    mean = s1 / N\n    s2 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        d = tl.where(mask, v - mean, 0.0)\n        s2 += tl.sum(d * d, axis=0)\n    var = s2 / N\n    rstd = 1.0 / tl.sqrt(var + 1e-5)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        w = tl.load(w_ptr + cols, mask=mask, other=0.0)\n        b = tl.load(b_ptr + cols, mask=mask, other=0.0)\n        tl.store(out_ptr + row * N + cols, (v - mean) * rstd * w + b, mask=mask)", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer w_ptr = N\n# litmus: buffer b_ptr = N\n# litmus: buffer out_ptr = M*N\n# litmus: buffer stats_ptr = 2\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef layernorm_race(x_ptr, w_ptr, b_ptr, out_ptr, stats_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    s1 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        s1 += tl.sum(v, axis=0)\n    mean = s1 / N\n    s2 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        d = tl.where(mask, v - mean, 0.0)\n        s2 += tl.sum(d * d, axis=0)\n    var = s2 / N\n    tl.store(stats_ptr + 0, mean)\n    tl.store(stats_ptr + 1, var)\n    mean = tl.load(stats_ptr + 0)\n    var = tl.load(stats_ptr + 1)\n    rstd = 1.0 / tl.sqrt(var + 1e-5)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        w = tl.load(w_ptr + cols, mask=mask, other=0.0)\n        b = tl.load(b_ptr + cols, mask=mask, other=0.0)\n        tl.store(out_ptr + row * N + cols, (v - mean) * rstd * w + b, mask=mask)", "why_standard_misses": "rows of i.i.d. noise have near-identical statistics, so reading another row's mean and variance barely changes the output", "witness_inputs": "{\"x\": {\"shape\": [4, 32], \"truncated\": false, \"values\": [[-0.7111066620607831, -0.18152747526685256, -0.5776135416552802, -0.1570444641441432, 1.023144014263097, -0.6279538720442022, -0.5224700162677491, 1.964888119715187, -2.007477339053849, -0.6355140596028792, 0.8284850190307785, -0.5158542499703735, -0.491047728186838, -1.7519032811239368, -1.914127349116291, 1.1443453419182168, 1.6512646699514324, 0.29142250799284164, -0.7206785692962974, -1.5410245016803001, -0.1884183002862263, 0.2567322687087859, -1.899208717879165, 0.4581405355649837, -0.46190139474424924, 1.405363499223668, 0.5235088198805894, 0.7185388800840798, -0.21640862091288507, -0.5772274719226567, -0.8631238077255025, 1.335149077344922], [0.5828965688072903, 0.02813924091965331, -0.1406443203806588, -0.16346815064781, 1.0143379657456404, 1.322055361650601, 1.5209658797118788, -1.7404112286296933, 0.22381979115666684, -1.7604352767783027, -0.16156275348350616, 0.545279319604191, -0.2787937046284972, -0.14622283971200487, 0.4428911074843226, 1.3064789510512147, -0.3665770754751895, 0.7538101496800835, -1.0994780434653573, -0.6653173099696014, 0.9311209649118, 0.701942432373225, -0.44658977111706105, 0.7510301668553768, -0.7326394147754076, -0.7063414354751946, 1.1229209699055602, -0.8072304779271209, -0.6505212369385155, -0.3125727298378464, 0.4255054706313187, 0.19541382934304083], [0.6651080120448106, 0.5034166659467222, 1.1180498251819484, -0.68303728113005, -0.21848139512982861, -0.4362856938036379, 0.5209084801405685, -0.9813577295719307, 0.18774118368828754, 0.22241416956921406, -0.31734977936208103, -3.2626006044352125, -0.8299346800329955, 1.6903754870539969, 0.3523165962048857, -1.259681927826936, -0.8835992237996873, -1.088400052686317, -1.8053750474611139, 0.9817007125414718, -0.029858451923118867, 0.08927106978214805, -0.05958513278465121, -0.37453947860340325, -0.6761206382004367, 0.2807316073833739, -1.0812676313359173, 0.5159749868460862, -2.0842284526502053, -0.7994430845937481, 0.8989409492602084, -0.704714560706215], [0.4221016244096251, -0.24137347426151173, 0.16192557678136812, 0.05905856261872254, 0.2715339168990874, 0.698816945496721, 0.4761063287800504, 0.18356014513588498, 1.1168805057105133, -0.037990541793860855, -0.7330805465870234, 1.1212281151200858, -1.0391234332791206, 1.194393660228538, 0.9906633124885316, 0.2008768535262879, -1.2557503915994408, -0.47911304654565706, -0.46559060580627704, 1.0097670348922356, -1.1819307735551194, -1.736502402089026, 1.4272155909640265, 0.08807459458471478, -2.4598703184629467, 0.8749321837332923, -0.7283130354977287, 1.0671424420630176, -0.816813014443165, -0.13375167926066417, 0.15135763396958035, -0.1915071411586796]]}, \"w\": {\"shape\": [32], \"truncated\": false, \"values\": [-1.4369850476607804, 0.9505636562188045, -0.2914108003493596, 0.6375240743026724, -0.000811902230876507, 0.49992314539077104, 1.6070034094602563, -2.195877402607221, -0.000893032308986645, -0.6639959220467672, -0.3017111410799019, -0.9742216383431987, 0.21473741641546262, 0.6387917891186016, -0.7795586906916712, -1.1938902047482707, 0.07173229161509213, 0.3277765114169412, -1.4980287526316596, 0.9049194050593735, -1.4839598981990128, 0.7924126698080183, -1.55577180292839, -2.134198113668178, 0.19611532354077255, -0.10662651393379763, 0.23540069809580239, 1.7927786461531043, -1.2602559242286895, 0.5413488234875216, 0.9928153256849107, -1.3156365519275874]}, \"b\": {\"shape\": [32], \"truncated\": false, \"values\": [0.0978678631700693, -1.5005114069098986, 0.3660979843404614, 1.4282581482609957, -0.38559780018130957, -1.2514908975652186, -0.2746253233315594, -0.3064147427576885, -0.12454477748489119, 0.5165554303315025, -0.30547756863900505, -2.4390355924437133, -0.667813573672545, 0.8870430252850763, 1.021226220546675, 1.0246478832438877, -0.859233406949938, -1.0555726300889476, 1.4505716312950572, 0.6667900698355521, 1.515222394691568, -0.6088536343986253, 0.34601494177355213, 0.8533521567032677, 0.6426205339579087, 2.3933119283317623, -1.6816710703305102, -0.2644240676321402, 0.7677252956834795, -1.946470068867866, -0.041458201274717256, -1.9470079518823284]}}", "witness_dims": "{\"M\": 4, \"N\": 32, \"S\": 32}", "numerically_observable": true}
{"family": "reduction", "bug_class": "race", "correct_name": "row_max", "buggy_name": "row_sum_barrier", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_sum_barrier(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.zeros((BLOCK,), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        acc += v\n    if tl.arange(0, BLOCK)[0] < N:\n        tl.debug_barrier()\n    tl.store(out_ptr + row, tl.sum(acc, axis=0))", "why_standard_misses": "divergent barriers produce no numerical error at all until the schedule changes -- they are invisible to any output comparison", "witness_inputs": "", "witness_dims": "", "numerically_observable": false}
{"family": "matmul", "bug_class": "dtype_promotion", "correct_name": "matmul_fp16", "buggy_name": "matmul_fp16_acc16", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_M = 32\n# litmus: constexpr BLOCK_N = 32\n# litmus: constexpr BLOCK_K = 32\n# litmus: buffer a_ptr = M*K\n# litmus: buffer b_ptr = K*N\n# litmus: buffer c_ptr = M*N\n# litmus: grid = cdiv(M, BLOCK_M), cdiv(N, BLOCK_N)\n\n@triton.jit\ndef matmul(a_ptr, b_ptr, c_ptr, M, N, K,\n           BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr):\n    pm = tl.program_id(0)\n    pn = tl.program_id(1)\n    offs_m = pm * BLOCK_M + tl.arange(0, BLOCK_M)\n    offs_n = pn * BLOCK_N + tl.arange(0, BLOCK_N)\n    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float32)\n    for k in range(0, tl.cdiv(K, BLOCK_K)):\n        offs_k = k * BLOCK_K + tl.arange(0, BLOCK_K)\n        a = tl.load(a_ptr + offs_m[:, None] * K + offs_k[None, :],\n                    mask=(offs_m[:, None] < M) & (offs_k[None, :] < K), other=0.0)\n        b = tl.load(b_ptr + offs_k[:, None] * N + offs_n[None, :],\n                    mask=(offs_k[:, None] < K) & (offs_n[None, :] < N), other=0.0)\n        acc += tl.dot(a, b)\n    tl.store(c_ptr + offs_m[:, None] * N + offs_n[None, :], acc,\n             mask=(offs_m[:, None] < M) & (offs_n[None, :] < N))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK_M = 32\n# litmus: constexpr BLOCK_N = 32\n# litmus: constexpr BLOCK_K = 32\n# litmus: buffer a_ptr = M*K\n# litmus: buffer b_ptr = K*N\n# litmus: buffer c_ptr = M*N\n# litmus: grid = cdiv(M, BLOCK_M), cdiv(N, BLOCK_N)\n\n@triton.jit\ndef matmul(a_ptr, b_ptr, c_ptr, M, N, K,\n           BLOCK_M: tl.constexpr, BLOCK_N: tl.constexpr, BLOCK_K: tl.constexpr):\n    pm = tl.program_id(0)\n    pn = tl.program_id(1)\n    offs_m = pm * BLOCK_M + tl.arange(0, BLOCK_M)\n    offs_n = pn * BLOCK_N + tl.arange(0, BLOCK_N)\n    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=tl.float16)\n    for k in range(0, tl.cdiv(K, BLOCK_K)):\n        offs_k = k * BLOCK_K + tl.arange(0, BLOCK_K)\n        a = tl.load(a_ptr + offs_m[:, None] * K + offs_k[None, :],\n                    mask=(offs_m[:, None] < M) & (offs_k[None, :] < K), other=0.0)\n        b = tl.load(b_ptr + offs_k[:, None] * N + offs_n[None, :],\n                    mask=(offs_k[:, None] < K) & (offs_n[None, :] < N), other=0.0)\n        acc += tl.dot(a, b)\n    tl.store(c_ptr + offs_m[:, None] * N + offs_n[None, :], acc,\n             mask=(offs_m[:, None] < M) & (offs_n[None, :] < N))", "why_standard_misses": "at K=512 with values in [0,1) the fp16 accumulator has not yet stagnated; the error stays under rtol=1e-2", "witness_inputs": "{\"a\": {\"shape\": [32, 2048], \"truncated\": true, \"values\": [3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4], \"note\": \"65536 elements; first 4096 kept\"}, \"b\": {\"shape\": [2048, 32], \"truncated\": true, \"values\": [3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4, 3.4], \"note\": \"65536 elements; first 4096 kept\"}}", "witness_dims": "{\"M\": 32, \"N\": 32, \"K\": 2048}", "numerically_observable": true}
{"family": "reduction", "bug_class": "dtype_promotion", "correct_name": "row_max", "buggy_name": "row_sum_fp16_acc", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_reduce(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.zeros((BLOCK,), dtype=tl.float16)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        acc += v\n    tl.store(out_ptr + row, tl.sum(acc, axis=0))", "why_standard_misses": "an fp16 accumulator stagnates above ~2048 and overflows above 65504; with N=512 and values in [0,1) the running total only reaches about 256", "witness_inputs": "{\"x\": {\"shape\": [2, 1024], \"truncated\": false, \"values\": [[100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0], [100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0, 100.0]]}}", "witness_dims": "{\"M\": 2, \"N\": 1024, \"S\": 1024}", "numerically_observable": true}
{"family": "reduction", "bug_class": "stride", "correct_name": "row_max", "buggy_name": "row_sum_wrong_stride", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_max(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.full((BLOCK,), float('-inf'), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=float('-inf'))\n        acc = tl.maximum(acc, v)\n    tl.store(out_ptr + row, tl.max(acc, axis=0))", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer out_ptr = M\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef row_reduce(x_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    acc = tl.zeros((BLOCK,), dtype=tl.float32)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=0.0)\n        acc += v\n    tl.store(out_ptr + row, tl.sum(acc, axis=0))", "why_standard_misses": "benchmarks always pass contiguous tensors, where the row stride equals the row width and the bug is a no-op", "witness_inputs": "{\"x\": {\"shape\": [3, 40], \"truncated\": false, \"values\": [[-0.7111066620607831, -0.18152747526685256, -0.5776135416552802, -0.1570444641441432, 1.023144014263097, -0.6279538720442022, -0.5224700162677491, 1.964888119715187, -2.007477339053849, -0.6355140596028792, 0.8284850190307785, -0.5158542499703735, -0.491047728186838, -1.7519032811239368, -1.914127349116291, 1.1443453419182168, 1.6512646699514324, 0.29142250799284164, -0.7206785692962974, -1.5410245016803001, -0.1884183002862263, 0.2567322687087859, -1.899208717879165, 0.4581405355649837, -0.46190139474424924, 1.405363499223668, 0.5235088198805894, 0.7185388800840798, -0.21640862091288507, -0.5772274719226567, -0.8631238077255025, 1.335149077344922, 0.5828965688072903, 0.02813924091965331, -0.1406443203806588, -0.16346815064781, 1.0143379657456404, 1.322055361650601, 1.5209658797118788, -1.7404112286296933], [0.22381979115666684, -1.7604352767783027, -0.16156275348350616, 0.545279319604191, -0.2787937046284972, -0.14622283971200487, 0.4428911074843226, 1.3064789510512147, -0.3665770754751895, 0.7538101496800835, -1.0994780434653573, -0.6653173099696014, 0.9311209649118, 0.701942432373225, -0.44658977111706105, 0.7510301668553768, -0.7326394147754076, -0.7063414354751946, 1.1229209699055602, -0.8072304779271209, -0.6505212369385155, -0.3125727298378464, 0.4255054706313187, 0.19541382934304083, 0.6651080120448106, 0.5034166659467222, 1.1180498251819484, -0.68303728113005, -0.21848139512982861, -0.4362856938036379, 0.5209084801405685, -0.9813577295719307, 0.18774118368828754, 0.22241416956921406, -0.31734977936208103, -3.2626006044352125, -0.8299346800329955, 1.6903754870539969, 0.3523165962048857, -1.259681927826936], [-0.8835992237996873, -1.088400052686317, -1.8053750474611139, 0.9817007125414718, -0.029858451923118867, 0.08927106978214805, -0.05958513278465121, -0.37453947860340325, -0.6761206382004367, 0.2807316073833739, -1.0812676313359173, 0.5159749868460862, -2.0842284526502053, -0.7994430845937481, 0.8989409492602084, -0.704714560706215, 0.4221016244096251, -0.24137347426151173, 0.16192557678136812, 0.05905856261872254, 0.2715339168990874, 0.698816945496721, 0.4761063287800504, 0.18356014513588498, 1.1168805057105133, -0.037990541793860855, -0.7330805465870234, 1.1212281151200858, -1.0391234332791206, 1.194393660228538, 0.9906633124885316, 0.2008768535262879, -1.2557503915994408, -0.47911304654565706, -0.46559060580627704, 1.0097670348922356, -1.1819307735551194, -1.736502402089026, 1.4272155909640265, 0.08807459458471478]]}, \"w\": {\"shape\": [32], \"truncated\": false, \"values\": [-2.4598703184629467, 0.8749321837332923, -0.7283130354977287, 1.0671424420630176, -0.816813014443165, -0.13375167926066417, 0.15135763396958035, -0.1915071411586796, -1.4369850476607804, 0.9505636562188045, -0.2914108003493596, 0.6375240743026724, -0.000811902230876507, 0.49992314539077104, 1.6070034094602563, -2.195877402607221, -0.000893032308986645, -0.6639959220467672, -0.3017111410799019, -0.9742216383431987, 0.21473741641546262, 0.6387917891186016, -0.7795586906916712, -1.1938902047482707, 0.07173229161509213, 0.3277765114169412, -1.4980287526316596, 0.9049194050593735, -1.4839598981990128, 0.7924126698080183, -1.55577180292839, -2.134198113668178]}, \"b\": {\"shape\": [32], \"truncated\": false, \"values\": [0.19611532354077255, -0.10662651393379763, 0.23540069809580239, 1.7927786461531043, -1.2602559242286895, 0.5413488234875216, 0.9928153256849107, -1.3156365519275874, 0.0978678631700693, -1.5005114069098986, 0.3660979843404614, 1.4282581482609957, -0.38559780018130957, -1.2514908975652186, -0.2746253233315594, -0.3064147427576885, -0.12454477748489119, 0.5165554303315025, -0.30547756863900505, -2.4390355924437133, -0.667813573672545, 0.8870430252850763, 1.021226220546675, 1.0246478832438877, -0.859233406949938, -1.0555726300889476, 1.4505716312950572, 0.6667900698355521, 1.515222394691568, -0.6088536343986253, 0.34601494177355213, 0.8533521567032677]}}", "witness_dims": "{\"M\": 3, \"N\": 32, \"S\": 40}", "numerically_observable": true}
{"family": "layernorm", "bug_class": "stride", "correct_name": "layernorm", "buggy_name": "layernorm_wrong_stride", "correct_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer w_ptr = N\n# litmus: buffer b_ptr = N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef layernorm(x_ptr, w_ptr, b_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    s1 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        s1 += tl.sum(v, axis=0)\n    mean = s1 / N\n    s2 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        d = tl.where(mask, v - mean, 0.0)\n        s2 += tl.sum(d * d, axis=0)\n    var = s2 / N\n    rstd = 1.0 / tl.sqrt(var + 1e-5)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * S + cols, mask=mask, other=0.0)\n        w = tl.load(w_ptr + cols, mask=mask, other=0.0)\n        b = tl.load(b_ptr + cols, mask=mask, other=0.0)\n        tl.store(out_ptr + row * N + cols, (v - mean) * rstd * w + b, mask=mask)", "buggy_source": "import triton\nimport triton.language as tl\n\n# litmus: constexpr BLOCK = 32\n# litmus: buffer x_ptr = M*S\n# litmus: buffer w_ptr = N\n# litmus: buffer b_ptr = N\n# litmus: buffer out_ptr = M*N\n# litmus: grid = M\n# litmus: subst S = N + P\n# litmus: sym P in [0, 8]\n\n@triton.jit\ndef layernorm(x_ptr, w_ptr, b_ptr, out_ptr, M, N, S, BLOCK: tl.constexpr):\n    row = tl.program_id(0)\n    s1 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=0.0)\n        s1 += tl.sum(v, axis=0)\n    mean = s1 / N\n    s2 = 0.0\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=0.0)\n        d = tl.where(mask, v - mean, 0.0)\n        s2 += tl.sum(d * d, axis=0)\n    var = s2 / N\n    rstd = 1.0 / tl.sqrt(var + 1e-5)\n    for start in range(0, N, BLOCK):\n        cols = start + tl.arange(0, BLOCK)\n        mask = cols < N\n        v = tl.load(x_ptr + row * N + cols, mask=mask, other=0.0)\n        w = tl.load(w_ptr + cols, mask=mask, other=0.0)\n        b = tl.load(b_ptr + cols, mask=mask, other=0.0)\n        tl.store(out_ptr + row * N + cols, (v - mean) * rstd * w + b, mask=mask)", "why_standard_misses": "benchmarks always pass contiguous tensors, where the row stride equals the row width and the bug is a no-op", "witness_inputs": "{\"x\": {\"shape\": [3, 40], \"truncated\": false, \"values\": [[-0.7111066620607831, -0.18152747526685256, -0.5776135416552802, -0.1570444641441432, 1.023144014263097, -0.6279538720442022, -0.5224700162677491, 1.964888119715187, -2.007477339053849, -0.6355140596028792, 0.8284850190307785, -0.5158542499703735, -0.491047728186838, -1.7519032811239368, -1.914127349116291, 1.1443453419182168, 1.6512646699514324, 0.29142250799284164, -0.7206785692962974, -1.5410245016803001, -0.1884183002862263, 0.2567322687087859, -1.899208717879165, 0.4581405355649837, -0.46190139474424924, 1.405363499223668, 0.5235088198805894, 0.7185388800840798, -0.21640862091288507, -0.5772274719226567, -0.8631238077255025, 1.335149077344922, 0.5828965688072903, 0.02813924091965331, -0.1406443203806588, -0.16346815064781, 1.0143379657456404, 1.322055361650601, 1.5209658797118788, -1.7404112286296933], [0.22381979115666684, -1.7604352767783027, -0.16156275348350616, 0.545279319604191, -0.2787937046284972, -0.14622283971200487, 0.4428911074843226, 1.3064789510512147, -0.3665770754751895, 0.7538101496800835, -1.0994780434653573, -0.6653173099696014, 0.9311209649118, 0.701942432373225, -0.44658977111706105, 0.7510301668553768, -0.7326394147754076, -0.7063414354751946, 1.1229209699055602, -0.8072304779271209, -0.6505212369385155, -0.3125727298378464, 0.4255054706313187, 0.19541382934304083, 0.6651080120448106, 0.5034166659467222, 1.1180498251819484, -0.68303728113005, -0.21848139512982861, -0.4362856938036379, 0.5209084801405685, -0.9813577295719307, 0.18774118368828754, 0.22241416956921406, -0.31734977936208103, -3.2626006044352125, -0.8299346800329955, 1.6903754870539969, 0.3523165962048857, -1.259681927826936], [-0.8835992237996873, -1.088400052686317, -1.8053750474611139, 0.9817007125414718, -0.029858451923118867, 0.08927106978214805, -0.05958513278465121, -0.37453947860340325, -0.6761206382004367, 0.2807316073833739, -1.0812676313359173, 0.5159749868460862, -2.0842284526502053, -0.7994430845937481, 0.8989409492602084, -0.704714560706215, 0.4221016244096251, -0.24137347426151173, 0.16192557678136812, 0.05905856261872254, 0.2715339168990874, 0.698816945496721, 0.4761063287800504, 0.18356014513588498, 1.1168805057105133, -0.037990541793860855, -0.7330805465870234, 1.1212281151200858, -1.0391234332791206, 1.194393660228538, 0.9906633124885316, 0.2008768535262879, -1.2557503915994408, -0.47911304654565706, -0.46559060580627704, 1.0097670348922356, -1.1819307735551194, -1.736502402089026, 1.4272155909640265, 0.08807459458471478]]}, \"w\": {\"shape\": [32], \"truncated\": false, \"values\": [-2.4598703184629467, 0.8749321837332923, -0.7283130354977287, 1.0671424420630176, -0.816813014443165, -0.13375167926066417, 0.15135763396958035, -0.1915071411586796, -1.4369850476607804, 0.9505636562188045, -0.2914108003493596, 0.6375240743026724, -0.000811902230876507, 0.49992314539077104, 1.6070034094602563, -2.195877402607221, -0.000893032308986645, -0.6639959220467672, -0.3017111410799019, -0.9742216383431987, 0.21473741641546262, 0.6387917891186016, -0.7795586906916712, -1.1938902047482707, 0.07173229161509213, 0.3277765114169412, -1.4980287526316596, 0.9049194050593735, -1.4839598981990128, 0.7924126698080183, -1.55577180292839, -2.134198113668178]}, \"b\": {\"shape\": [32], \"truncated\": false, \"values\": [0.19611532354077255, -0.10662651393379763, 0.23540069809580239, 1.7927786461531043, -1.2602559242286895, 0.5413488234875216, 0.9928153256849107, -1.3156365519275874, 0.0978678631700693, -1.5005114069098986, 0.3660979843404614, 1.4282581482609957, -0.38559780018130957, -1.2514908975652186, -0.2746253233315594, -0.3064147427576885, -0.12454477748489119, 0.5165554303315025, -0.30547756863900505, -2.4390355924437133, -0.667813573672545, 0.8870430252850763, 1.021226220546675, 1.0246478832438877, -0.859233406949938, -1.0555726300889476, 1.4505716312950572, 0.6667900698355521, 1.515222394691568, -0.6088536343986253, 0.34601494177355213, 0.8533521567032677]}}", "witness_dims": "{\"M\": 3, \"N\": 32, \"S\": 40}", "numerically_observable": true}