//
	//            c=constant k=key b=blockcount n=nonce
	var (
		c0, c1, c2, c3   = j0, j1, j2, j3
		c4, c5, c6, c7   = s.key[0], s.key[1], s.key[2], s.key[3]
		c8, c9, c10, c11 = s.key[4], s.key[5], s.key[6], s.key[7]
		_, c13, c14, c15 = s.counter, s.nonce[0], s.nonce[1], s.nonce[2]
	)

	// Three quarters of the first round don't depend on the counter, so we can

	}
	// arm64:`CMN`,-`CMP`
	if a*c+b != 0 {
		c9 = 1
	}
	// arm64:`CMP`,-`CMN`
	if b*c-a == 0 {
		c10 = 1
	}
	// arm64:`CMP`,-`CMN`
	if a*d-b != 0 {
		c11 = 1
	}
	return c1 + c2 + c3 + c4 + c5 + c6 + c7 + c8 + c9 + c10 + c11
}

func cmpToCmnLessThan(a, b, c, d int) int {
	var c1, c2, c3, c4 int
	// arm64:`CMN`,`CSET\tMI`,-`CMP`
	if a+1 < 0 {
		c1 = 1
	}

	}
	if runtime.GOOS == "windows" {
		t.Skipf("skipping: test requires pthread support")
		// TODO: Can this test be rewritten to use the C11 thread API instead?
	}

	testenv.MustHaveGoRun(t)

	// This test requires building various packages with -race, so
	// it's somewhat slow.
	if testing.Short() {
		t.Skip("skipping test in -short mode")

        "//tensorflow/core:session_options",
        "//tensorflow/core:test",
    ],
)

tf_cc_test(
    name = "c_test",
    srcs = ["c_test.c"],
    extra_copts = ["-std=c11"],
    deps = [
        ":c_api",
        ":c_api_experimental",
        ":env",
        ":kernels",
    ],
)

tf_cuda_cc_test(
    name = "c_api_test",
    size = "medium",

// sections of code).
// RaceAcquire establishes a happens-before relation with the preceding
// RaceReleaseMerge on addr up to and including the last RaceRelease on addr.
// In terms of the C memory model (C11 §5.1.2.4, §7.17.3),
// RaceAcquire is equivalent to atomic_load(memory_order_acquire).
//
//go:nosplit
func RaceAcquire(addr unsafe.Pointer) {
	raceacquire(addr)
}