mirror of
https://github.com/tinygrad/tinygrad.git
synced 2026-01-09 15:08:02 -05:00
RDNA3 fp16 assembly gemm 85 TFLOPS (#13990)
This commit is contained in:
@@ -140,11 +140,11 @@ def hand_spec_kernel3():
|
||||
|
||||
return sink.sink(arg=KernelInfo(opts_to_apply=())).simplify()
|
||||
|
||||
def test_matmul(sink:UOp, N=N):
|
||||
def test_matmul(sink:UOp, dtype=dtypes.float32, N=N):
|
||||
rng = np.random.default_rng()
|
||||
a = Tensor(rng.random((N, N), dtype=np.float32)-0.5)
|
||||
b = Tensor(rng.random((N, N), dtype=np.float32)-0.5)
|
||||
hc = Tensor.empty(N, N)
|
||||
a = Tensor(rng.random((N, N), dtype=np.float32)-0.5, dtype=dtype)
|
||||
b = Tensor(rng.random((N, N), dtype=np.float32)-0.5, dtype=dtype)
|
||||
hc = Tensor.empty(N, N, dtype=dtype)
|
||||
Tensor.realize(a, b, hc)
|
||||
|
||||
ei = ExecItem(sink, [t.uop.buffer for t in [hc, a, b]], prg=get_runner(Device.DEFAULT, sink))
|
||||
|
||||
Reference in New Issue
Block a user