Move the rest of the math ops to logical

[dsmilkov]

This resolves #18

This change is 

[nsthorat]

---

Reviewed 6 of 7 files at r1, 17 of 18 files at r2.
Review status: all files reviewed at latest revision, 4 unresolved discussions.

---

src/math/math_gpu.ts, line 101 at r2 (raw file):

    const program = new Copy2DProgram(sourceSizeRowCol[1], destSizeRowCol[1]);
    const customSetup = copy_gpu.getCustomSetupFunc(
        sourceBeginRowCol, destBeginRowCol, destSizeRowCol);

one thought: getCustomSetupFunc could be part of the GPGPUProgram, though you lose a little flexibility. I'm fine either way, just throwing ideas out there :)

---

src/math/math_gpu.ts, line 359 at r2 (raw file):

      alignCorners: boolean): Array3D {
    const program =
        new ResizeBilinearProgram(x.shape, newShape2D, alignCorners);

rename this to ResizeBilinear3DProgram

---

src/math/webgl/batchnorm_gpu_test.ts, line 130 at r2 (raw file):

    const result = uploadBatchNormDownload(
        x, [2, 3, 3], mean, [1, 3], variance, [1, 3], offset, [1, 3], scale,

does 2, 9 not work, or do you not like that test?

---

src/math/webgl/resize_bilinear_gpu.ts, line 42 at r2 (raw file):

      const vec2 effectiveInputOverOutputRatioRC = vec2(
          ${effectiveInputShape[0] /
        effectiveOutputShape[0]},

i think this fits on prev line

---

Comments from Reviewable

[dsmilkov]

Review status: 17 of 22 files reviewed at latest revision, 4 unresolved discussions, some commit checks pending.

---

src/math/math_gpu.ts, line 101 at r2 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

one thought: getCustomSetupFunc could be part of the GPGPUProgram, though you lose a little flexibility. I'm fine either way, just throwing ideas out there :)

Done.

---

src/math/math_gpu.ts, line 359 at r2 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

rename this to ResizeBilinear3DProgram

Done.

---

src/math/webgl/batchnorm_gpu_test.ts, line 130 at r2 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

does 2, 9 not work, or do you not like that test?

[2, 9] corresponds to the texture shape, which is what the old uploadBatchNormDownload was expecting. [2, 3, 3] is the logical shape which is what the rewritten uploadDownload is expecting. (I didn't want any dependency/usage of texture shapes in the tests)

---

src/math/webgl/resize_bilinear_gpu.ts, line 42 at r2 (raw file):

Previously, nsthorat (Nikhil Thorat) wrote…

i think this fits on prev line

Done.

---

Comments from Reviewable