Add complex64 support for tf.abs().

[nsthorat]

We hide this behind tf.abs.

y = sqrt(real ^ 2 + imag ^ 2)

---

This change is 

[dsmilkov]

Reviewed 2 of 4 files at r1.
Reviewable status: 0 of 1 approvals obtained (waiting on @nsthorat and @dsmilkov)

---

src/kernels/webgl/complex_abs_gpu.ts, line 32 at r1 (raw file):

        float imag = getImagAtOutCoords();

        setOutput(sqrt(real * real + imag * imag));

vec2 v = vec2(real, imag)
setOutput(dot(v, v))

---

src/ops/unary_ops_test.ts, line 142 at r1 (raw file):

      Math.sqrt(0 * 0 + 3 * 3), Math.sqrt(1 * 1 + 0 * 0),
      Math.sqrt(2 * 2 + -1 * -1)
    ]);

add couple more unit tests for different ranks

[dsmilkov]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @nsthorat and @dsmilkov)

---

src/kernels/webgl/complex_abs_gpu.ts, line 32 at r1 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

vec2 v = vec2(real, imag)
setOutput(dot(v, v))

forgot sqrt()

[nsthorat]

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @nsthorat)

---

src/kernels/webgl/complex_abs_gpu.ts, line 32 at r1 (raw file):

vec2 v = vec2(real, imag)

Nice. Done.

---

src/ops/unary_ops_test.ts, line 142 at r1 (raw file):

Previously, dsmilkov (Daniel Smilkov) wrote…

add couple more unit tests for different ranks

Done

[DirkToewe]

There's also a length() function in GLSL ES 1.00. It might be slightly faster and also avoid some fringe cases of float32 underflow ( float32( (1e-23)² ) = 0 ).

[dsmilkov]

That's a great point @DirkToewe. Want to send a PR?

[DirkToewe]

Here You go