Enable support for Emscripten in CMake

[aiusepsi]

Addresses #3455

Takes the approach of detecting the use of Wasm SIMD by detecting if the requisite flags have been added to CMAKE_C_FLAGS or CMAKE_CXX_FLAGS. Adding the SIMD flags to these variables is the right way to enable Wasm SIMD globally for CMake builds, so this should do the right thing.

The use of SET(XNNPACK_ENABLE_RISCV_VECTOR OFF) is to suppress a compile error; without this RISC-specific code is compiled for Wasm; turning this option off here seemed the minimally invasive solution.

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[fbarchard]

RISCV in particular?

[aiusepsi]

Yeah, the XNNPACK_ENABLE_RISCV_VECTOR option defaults to ON, which causes the define XNN_ENABLE_RISCV_VECTOR to be set, which causes this line:

XNNPACK/src/configs/transpose-config.c

Line 203 in 4b857f3

if (hardware_config->vlenb >= 128) {

to be compiled, which (along with a few others) falls over:

XNNPACK/src/configs/transpose-config.c:203:28: error: no member named 'vlenb' in 'struct xnn_hardware_config'
  203 |       if (hardware_config->vlenb >= 128) {
      |           ~~~~~~~~~~~~~~~  ^
XNNPACK/src/configs/transpose-config.c:205:61: error: use of undeclared identifier 'xnn_x32_transposec_ukernel__32x8_rvv'
  205 |           .const_size_ukernel = (xnn_transposec_ukernel_fn) xnn_x32_transposec_ukernel__32x8_rvv,
      |                                                             ^
XNNPACK/src/configs/transpose-config.c:208:35: error: no member named 'vlenb' in 'struct xnn_hardware_config'
  208 |       } else if (hardware_config->vlenb == 64) {
      |                  ~~~~~~~~~~~~~~~  ^
XNNPACK/src/configs/transpose-config.c:210:61: error: use of undeclared identifier 'xnn_x32_transposec_ukernel__16x8_rvv'
  210 |           .const_size_ukernel = (xnn_transposec_ukernel_fn) xnn_x32_transposec_ukernel__16x8_rvv,
      |                                                             ^
XNNPACK/src/configs/transpose-config.c:213:35: error: no member named 'vlenb' in 'struct xnn_hardware_config'
  213 |       } else if (hardware_config->vlenb == 32) {
      |                  ~~~~~~~~~~~~~~~  ^
XNNPACK/src/configs/transpose-config.c:215:61: error: use of undeclared identifier 'xnn_x32_transposec_ukernel__8x8_rvv'
  215 |           .const_size_ukernel = (xnn_transposec_ukernel_fn) xnn_x32_transposec_ukernel__8x8_rvv,
      |                                                             ^
XNNPACK/src/configs/transpose-config.c:218:35: error: no member named 'vlenb' in 'struct xnn_hardware_config'
  218 |       } else if (hardware_config->vlenb == 16) {
      |                  ~~~~~~~~~~~~~~~  ^
XNNPACK/src/configs/transpose-config.c:220:61: error: use of undeclared identifier 'xnn_x32_transposec_ukernel__4x4_rvv'
  220 |           .const_size_ukernel = (xnn_transposec_ukernel_fn) xnn_x32_transposec_ukernel__4x4_rvv,

There's possible alternative fixes (e.g. fixing the ifdefs in transpose-config.c so that code isn't compiled, or defaulting the XNNPACK_ENABLE_RISCV_VECTOR to OFF except for RISCV builds, etc), but switching it off here in the Emscripten section was just the lowest risk of affecting anything else option I could see.