Split scalar production microkernels into portable, AArch32, and Wasm
PiperOrigin-RevId: 417893796
diff --git a/BUILD.bazel b/BUILD.bazel
index d571a59..511825a 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -122,8 +122,15 @@
"src/tables/exp2minus-k-over-2048.c",
]
-PROD_SCALAR_MICROKERNEL_SRCS = [
- "src/f16-f32-vcvt/gen/vcvt-scalar-float-x1.c",
+PROD_SCALAR_PORTABLE_MICROKERNEL_SRCS = [
+ "src/params-init.c",
+ "src/u8-lut32norm/scalar.c",
+ "src/x8-lut/gen/lut-scalar-x4.c",
+ "src/x32-depthtospace2d-chw2hwc/scalar.c",
+ "src/xx-copy/memcpy.c",
+]
+
+PROD_SCALAR_AARCH32_MICROKERNEL_SRCS = [
"src/f16-f32-vcvt/gen/vcvt-scalar-float-x4.c",
"src/f32-argmaxpool/4x-scalar-c1.c",
"src/f32-argmaxpool/9p8x-scalar-c1.c",
@@ -141,15 +148,10 @@
"src/f32-dwconv/gen/up1x9-scalar-acc2.c",
"src/f32-dwconv/gen/up1x25-minmax-scalar-acc2.c",
"src/f32-dwconv/gen/up1x25-scalar-acc2.c",
- "src/f32-dwconv2d-chw/gen/3x3p1-minmax-scalar-2x1-acc2.c",
"src/f32-dwconv2d-chw/gen/3x3p1-minmax-scalar-4x1.c",
- "src/f32-dwconv2d-chw/gen/3x3s2p1-minmax-scalar-1x1-acc2.c",
"src/f32-dwconv2d-chw/gen/3x3s2p1-minmax-scalar-2x1-acc2.c",
- "src/f32-dwconv2d-chw/gen/5x5p2-minmax-scalar-1x1-acc5.c",
"src/f32-dwconv2d-chw/gen/5x5p2-minmax-scalar-2x1-acc2.c",
- "src/f32-dwconv2d-chw/gen/5x5s2p2-minmax-scalar-1x1-acc5.c",
"src/f32-dwconv2d-chw/gen/5x5s2p2-minmax-scalar-2x1-acc2.c",
- "src/f32-f16-vcvt/gen/vcvt-scalar-bitcast-x4.c",
"src/f32-f16-vcvt/gen/vcvt-scalar-fabsf-x2.c",
"src/f32-gavgpool-cw/scalar-x1.c",
"src/f32-gavgpool/7p7x-minmax-scalar-c1.c",
@@ -157,11 +159,7 @@
"src/f32-gemm/gen/1x4-minmax-scalar.c",
"src/f32-gemm/gen/1x4-relu-scalar.c",
"src/f32-gemm/gen/1x4-scalar.c",
- "src/f32-gemm/gen/2x4-minmax-scalar.c",
- "src/f32-gemm/gen/2x4-relu-scalar.c",
- "src/f32-gemm/gen/2x4-scalar.c",
"src/f32-gemm/gen/4x2-minmax-scalar.c",
- "src/f32-gemm/gen/4x2-relu-scalar.c",
"src/f32-gemm/gen/4x2-scalar.c",
"src/f32-gemm/gen/4x4-minmax-scalar.c",
"src/f32-gemm/gen/4x4-relu-scalar.c",
@@ -171,11 +169,7 @@
"src/f32-igemm/gen/1x4-minmax-scalar.c",
"src/f32-igemm/gen/1x4-relu-scalar.c",
"src/f32-igemm/gen/1x4-scalar.c",
- "src/f32-igemm/gen/2x4-minmax-scalar.c",
- "src/f32-igemm/gen/2x4-relu-scalar.c",
- "src/f32-igemm/gen/2x4-scalar.c",
"src/f32-igemm/gen/4x2-minmax-scalar.c",
- "src/f32-igemm/gen/4x2-relu-scalar.c",
"src/f32-igemm/gen/4x2-scalar.c",
"src/f32-igemm/gen/4x4-minmax-scalar.c",
"src/f32-igemm/gen/4x4-relu-scalar.c",
@@ -184,9 +178,7 @@
"src/f32-pavgpool/9p8x-minmax-scalar-c1.c",
"src/f32-pavgpool/9x-minmax-scalar-c1.c",
"src/f32-prelu/gen/scalar-2x4.c",
- "src/f32-qs8-vcvt/gen/vcvt-scalar-magic-iminmax-x1.c",
"src/f32-qs8-vcvt/gen/vcvt-scalar-magic-iminmax-x4.c",
- "src/f32-qu8-vcvt/gen/vcvt-scalar-magic-iminmax-x1.c",
"src/f32-qu8-vcvt/gen/vcvt-scalar-magic-iminmax-x4.c",
"src/f32-raddstoreexpminusmax/gen/scalar-p5-x4-acc2.c",
"src/f32-rmax/scalar.c",
@@ -196,9 +188,7 @@
"src/f32-vbinary/gen/vadd-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vaddc-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vdiv-minmax-scalar-x2.c",
- "src/f32-vbinary/gen/vdiv-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vdivc-minmax-scalar-x2.c",
- "src/f32-vbinary/gen/vdivc-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vmax-scalar-x8.c",
"src/f32-vbinary/gen/vmaxc-scalar-x8.c",
"src/f32-vbinary/gen/vmin-scalar-x8.c",
@@ -206,6 +196,138 @@
"src/f32-vbinary/gen/vmul-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vmulc-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vrdivc-minmax-scalar-x2.c",
+ "src/f32-vbinary/gen/vrsubc-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vsqrdiff-scalar-x8.c",
+ "src/f32-vbinary/gen/vsqrdiffc-scalar-x8.c",
+ "src/f32-vbinary/gen/vsub-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vsubc-minmax-scalar-x8.c",
+ "src/f32-vclamp/gen/vclamp-scalar-x4.c",
+ "src/f32-velu/gen/velu-scalar-rr2-lut16-p3-x4.c",
+ "src/f32-vhswish/gen/vhswish-scalar-x4.c",
+ "src/f32-vlrelu/gen/vlrelu-scalar-x4.c",
+ "src/f32-vmulcaddc/gen/c1-minmax-scalar-2x.c",
+ "src/f32-vrelu/gen/vrelu-scalar-x8.c",
+ "src/f32-vrnd/gen/vrndd-scalar-libm-x1.c",
+ "src/f32-vrnd/gen/vrndne-scalar-libm-x1.c",
+ "src/f32-vrnd/gen/vrndu-scalar-libm-x1.c",
+ "src/f32-vrnd/gen/vrndz-scalar-libm-x1.c",
+ "src/f32-vsigmoid/gen/vsigmoid-scalar-lut64-p2-div-x2.c",
+ "src/f32-vsqrt/gen/scalar-sqrt-x1.c",
+ "src/f32-vunary/gen/vabs-scalar-x4.c",
+ "src/f32-vunary/gen/vneg-scalar-x4.c",
+ "src/f32-vunary/gen/vsqr-scalar-x4.c",
+ "src/qc8-dwconv/gen/up2x9-minmax-fp32-scalar-magic.c",
+ "src/qc8-dwconv/gen/up2x25-minmax-fp32-scalar-magic.c",
+ "src/qc8-gemm/gen/1x2-minmax-fp32-scalar-magic.c",
+ "src/qc8-gemm/gen/2x2-minmax-fp32-scalar-magic.c",
+ "src/qc8-igemm/gen/1x2-minmax-fp32-scalar-magic.c",
+ "src/qc8-igemm/gen/2x2-minmax-fp32-scalar-magic.c",
+ "src/qs8-dwconv/gen/up1x9-minmax-fp32-scalar-magic.c",
+ "src/qs8-dwconv/gen/up1x25-minmax-fp32-scalar-magic.c",
+ "src/qs8-f32-vcvt/gen/vcvt-scalar-x4.c",
+ "src/qs8-gavgpool/gen/7p7x-minmax-scalar-c1.c",
+ "src/qs8-gavgpool/gen/7x-minmax-scalar-c1.c",
+ "src/qs8-gemm/gen/1x2-minmax-fp32-scalar-magic.c",
+ "src/qs8-gemm/gen/2x2-minmax-fp32-scalar-magic.c",
+ "src/qs8-igemm/gen/1x2-minmax-fp32-scalar-magic.c",
+ "src/qs8-igemm/gen/2x2-minmax-fp32-scalar-magic.c",
+ "src/qs8-vadd/gen/minmax-scalar-x1.c",
+ "src/qs8-vaddc/gen/minmax-scalar-x1.c",
+ "src/qs8-vmul/gen/minmax-fp32-scalar-x4.c",
+ "src/qs8-vmulc/gen/minmax-fp32-scalar-x4.c",
+ "src/qu8-avgpool/9p8x-minmax-scalar-c1.c",
+ "src/qu8-avgpool/9x-minmax-scalar-c1.c",
+ "src/qu8-dwconv/gen/up1x9-minmax-fp32-scalar-magic.c",
+ "src/qu8-dwconv/gen/up1x25-minmax-fp32-scalar-magic.c",
+ "src/qu8-f32-vcvt/gen/vcvt-scalar-x4.c",
+ "src/qu8-gavgpool/7p7x-minmax-scalar-c1.c",
+ "src/qu8-gavgpool/7x-minmax-scalar-c1.c",
+ "src/qu8-gemm/gen/1x2-minmax-fp32-scalar-magic.c",
+ "src/qu8-gemm/gen/2x2-minmax-fp32-scalar-magic.c",
+ "src/qu8-igemm/gen/1x2-minmax-fp32-scalar-magic.c",
+ "src/qu8-igemm/gen/2x2-minmax-fp32-scalar-magic.c",
+ "src/qu8-vadd/gen/minmax-scalar-x1.c",
+ "src/qu8-vaddc/gen/minmax-scalar-x1.c",
+ "src/qu8-vmul/gen/minmax-fp32-scalar-x4.c",
+ "src/qu8-vmulc/gen/minmax-fp32-scalar-x4.c",
+ "src/s8-ibilinear/gen/scalar-c1.c",
+ "src/s8-maxpool/9p8x-minmax-scalar-c1.c",
+ "src/s8-vclamp/scalar-x4.c",
+ "src/u8-ibilinear/gen/scalar-c1.c",
+ "src/u8-maxpool/9p8x-minmax-scalar-c1.c",
+ "src/u8-rmax/scalar.c",
+ "src/u8-vclamp/scalar-x4.c",
+ "src/x8-zip/x2-scalar.c",
+ "src/x8-zip/x3-scalar.c",
+ "src/x8-zip/x4-scalar.c",
+ "src/x8-zip/xm-scalar.c",
+ "src/x32-packx/x2-scalar.c",
+ "src/x32-packx/x3-scalar.c",
+ "src/x32-packx/x4-scalar.c",
+ "src/x32-unpool/scalar.c",
+ "src/x32-zip/x2-scalar.c",
+ "src/x32-zip/x3-scalar.c",
+ "src/x32-zip/x4-scalar.c",
+ "src/x32-zip/xm-scalar.c",
+ "src/xx-fill/scalar-x16.c",
+ "src/xx-pad/scalar.c",
+]
+
+PROD_SCALAR_WASM_MICROKERNEL_SRCS = [
+ "src/f16-f32-vcvt/gen/vcvt-scalar-float-x1.c",
+ "src/f32-argmaxpool/4x-scalar-c1.c",
+ "src/f32-argmaxpool/9p8x-scalar-c1.c",
+ "src/f32-argmaxpool/9x-scalar-c1.c",
+ "src/f32-avgpool/9p8x-minmax-scalar-c1.c",
+ "src/f32-avgpool/9x-minmax-scalar-c1.c",
+ "src/f32-conv-hwc/3x3s2p0p1c3x4-scalar-1x1.c",
+ "src/f32-conv-hwc/3x3s2p1c3x4-scalar-1x1.c",
+ "src/f32-conv-hwc2chw/3x3s2p1c3x4-scalar-1x1.c",
+ "src/f32-dwconv/gen/up1x3-minmax-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x3-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x4-minmax-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x4-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x9-minmax-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x9-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x25-minmax-scalar-acc2.c",
+ "src/f32-dwconv/gen/up1x25-scalar-acc2.c",
+ "src/f32-dwconv2d-chw/gen/3x3p1-minmax-scalar-2x1-acc2.c",
+ "src/f32-dwconv2d-chw/gen/3x3s2p1-minmax-scalar-1x1-acc2.c",
+ "src/f32-dwconv2d-chw/gen/5x5p2-minmax-scalar-1x1-acc5.c",
+ "src/f32-dwconv2d-chw/gen/5x5s2p2-minmax-scalar-1x1-acc5.c",
+ "src/f32-f16-vcvt/gen/vcvt-scalar-bitcast-x4.c",
+ "src/f32-gavgpool-cw/scalar-x1.c",
+ "src/f32-gavgpool/7p7x-minmax-scalar-c1.c",
+ "src/f32-gavgpool/7x-minmax-scalar-c1.c",
+ "src/f32-gemm/gen/2x4-minmax-scalar.c",
+ "src/f32-gemm/gen/2x4-relu-scalar.c",
+ "src/f32-gemm/gen/2x4-scalar.c",
+ "src/f32-ibilinear-chw/gen/scalar-p4.c",
+ "src/f32-ibilinear/gen/scalar-c2.c",
+ "src/f32-igemm/gen/2x4-minmax-scalar.c",
+ "src/f32-igemm/gen/2x4-relu-scalar.c",
+ "src/f32-igemm/gen/2x4-scalar.c",
+ "src/f32-maxpool/9p8x-minmax-scalar-c1.c",
+ "src/f32-pavgpool/9p8x-minmax-scalar-c1.c",
+ "src/f32-pavgpool/9x-minmax-scalar-c1.c",
+ "src/f32-prelu/gen/scalar-2x4.c",
+ "src/f32-qs8-vcvt/gen/vcvt-scalar-magic-iminmax-x1.c",
+ "src/f32-qu8-vcvt/gen/vcvt-scalar-magic-iminmax-x1.c",
+ "src/f32-raddstoreexpminusmax/gen/scalar-p5-x4-acc2.c",
+ "src/f32-rmax/scalar.c",
+ "src/f32-spmm/gen/8x1-minmax-scalar.c",
+ "src/f32-spmm/gen/8x2-minmax-scalar.c",
+ "src/f32-spmm/gen/8x4-minmax-scalar.c",
+ "src/f32-vbinary/gen/vadd-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vaddc-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vdiv-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vdivc-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vmaxc-scalar-x8.c",
+ "src/f32-vbinary/gen/vmin-scalar-x8.c",
+ "src/f32-vbinary/gen/vminc-scalar-x8.c",
+ "src/f32-vbinary/gen/vmul-minmax-scalar-x8.c",
+ "src/f32-vbinary/gen/vmulc-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vrdivc-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vrsubc-minmax-scalar-x8.c",
"src/f32-vbinary/gen/vsqrdiff-scalar-x8.c",
@@ -214,25 +336,19 @@
"src/f32-vbinary/gen/vsubc-minmax-scalar-x8.c",
"src/f32-vclamp/gen/vclamp-scalar-x4.c",
"src/f32-velu/gen/velu-scalar-rr2-lut16-p3-x2.c",
- "src/f32-velu/gen/velu-scalar-rr2-lut16-p3-x4.c",
"src/f32-vhswish/gen/vhswish-scalar-x4.c",
"src/f32-vlrelu/gen/vlrelu-scalar-x4.c",
"src/f32-vmulcaddc/gen/c1-minmax-scalar-2x.c",
"src/f32-vrelu/gen/vrelu-scalar-x8.c",
- "src/f32-vrnd/gen/vrndd-scalar-libm-x1.c",
"src/f32-vrnd/gen/vrndd-scalar-libm-x4.c",
- "src/f32-vrnd/gen/vrndne-scalar-libm-x1.c",
"src/f32-vrnd/gen/vrndne-scalar-libm-x4.c",
- "src/f32-vrnd/gen/vrndu-scalar-libm-x1.c",
"src/f32-vrnd/gen/vrndu-scalar-libm-x4.c",
- "src/f32-vrnd/gen/vrndz-scalar-libm-x1.c",
"src/f32-vrnd/gen/vrndz-scalar-libm-x4.c",
"src/f32-vsigmoid/gen/vsigmoid-scalar-lut64-p2-div-x2.c",
"src/f32-vsqrt/gen/scalar-sqrt-x1.c",
"src/f32-vunary/gen/vabs-scalar-x4.c",
"src/f32-vunary/gen/vneg-scalar-x4.c",
"src/f32-vunary/gen/vsqr-scalar-x4.c",
- "src/params-init.c",
"src/qc8-dwconv/gen/up2x9-minmax-fp32-scalar-magic.c",
"src/qc8-dwconv/gen/up2x25-minmax-fp32-scalar-magic.c",
"src/qc8-gemm/gen/1x2-minmax-fp32-scalar-magic.c",
@@ -243,42 +359,28 @@
"src/qc8-igemm/gen/1x4-minmax-fp32-scalar-magic.c",
"src/qc8-igemm/gen/2x2-minmax-fp32-scalar-magic.c",
"src/qc8-igemm/gen/4x4-minmax-fp32-scalar-magic.c",
- "src/qs8-dwconv/gen/up1x9-minmax-fp32-scalar-magic.c",
- "src/qs8-dwconv/gen/up1x25-minmax-fp32-scalar-magic.c",
"src/qs8-dwconv/gen/up2x9-minmax-fp32-scalar-magic.c",
"src/qs8-dwconv/gen/up2x25-minmax-fp32-scalar-magic.c",
"src/qs8-f32-vcvt/gen/vcvt-scalar-x1.c",
- "src/qs8-f32-vcvt/gen/vcvt-scalar-x4.c",
- "src/qs8-gavgpool/gen/7p7x-minmax-scalar-c1.c",
"src/qs8-gavgpool/gen/7p7x-minmax-scalar-c4.c",
- "src/qs8-gavgpool/gen/7x-minmax-scalar-c1.c",
"src/qs8-gavgpool/gen/7x-minmax-scalar-c4.c",
"src/qs8-gemm/gen/1x2-minmax-fp32-scalar-magic.c",
"src/qs8-gemm/gen/1x4-minmax-fp32-scalar-magic.c",
- "src/qs8-gemm/gen/1x4-minmax-rndnu-scalar.c",
"src/qs8-gemm/gen/2x2-minmax-fp32-scalar-magic.c",
- "src/qs8-gemm/gen/3x4-minmax-rndnu-scalar.c",
"src/qs8-gemm/gen/4x4-minmax-fp32-scalar-magic.c",
"src/qs8-igemm/gen/1x2-minmax-fp32-scalar-magic.c",
"src/qs8-igemm/gen/1x4-minmax-fp32-scalar-magic.c",
- "src/qs8-igemm/gen/1x4-minmax-rndnu-scalar.c",
"src/qs8-igemm/gen/2x2-minmax-fp32-scalar-magic.c",
- "src/qs8-igemm/gen/3x4-minmax-rndnu-scalar.c",
"src/qs8-igemm/gen/4x4-minmax-fp32-scalar-magic.c",
- "src/qs8-vadd/gen/minmax-scalar-x1.c",
"src/qs8-vadd/gen/minmax-scalar-x4.c",
- "src/qs8-vaddc/gen/minmax-scalar-x1.c",
"src/qs8-vaddc/gen/minmax-scalar-x4.c",
"src/qs8-vmul/gen/minmax-fp32-scalar-x4.c",
"src/qs8-vmulc/gen/minmax-fp32-scalar-x4.c",
"src/qu8-avgpool/9p8x-minmax-scalar-c1.c",
"src/qu8-avgpool/9x-minmax-scalar-c1.c",
- "src/qu8-dwconv/gen/up1x9-minmax-fp32-scalar-magic.c",
- "src/qu8-dwconv/gen/up1x25-minmax-fp32-scalar-magic.c",
"src/qu8-dwconv/gen/up2x9-minmax-fp32-scalar-magic.c",
"src/qu8-dwconv/gen/up2x25-minmax-fp32-scalar-magic.c",
"src/qu8-f32-vcvt/gen/vcvt-scalar-x1.c",
- "src/qu8-f32-vcvt/gen/vcvt-scalar-x4.c",
"src/qu8-gavgpool/7p7x-minmax-scalar-c1.c",
"src/qu8-gavgpool/7x-minmax-scalar-c1.c",
"src/qu8-gemm/gen/1x2-minmax-fp32-scalar-magic.c",
@@ -289,9 +391,7 @@
"src/qu8-igemm/gen/1x4-minmax-fp32-scalar-magic.c",
"src/qu8-igemm/gen/2x2-minmax-fp32-scalar-magic.c",
"src/qu8-igemm/gen/4x4-minmax-fp32-scalar-magic.c",
- "src/qu8-vadd/gen/minmax-scalar-x1.c",
"src/qu8-vadd/gen/minmax-scalar-x4.c",
- "src/qu8-vaddc/gen/minmax-scalar-x1.c",
"src/qu8-vaddc/gen/minmax-scalar-x4.c",
"src/qu8-vmul/gen/minmax-fp32-scalar-x4.c",
"src/qu8-vmulc/gen/minmax-fp32-scalar-x4.c",
@@ -299,16 +399,13 @@
"src/s8-maxpool/9p8x-minmax-scalar-c1.c",
"src/s8-vclamp/scalar-x4.c",
"src/u8-ibilinear/gen/scalar-c1.c",
- "src/u8-lut32norm/scalar.c",
"src/u8-maxpool/9p8x-minmax-scalar-c1.c",
"src/u8-rmax/scalar.c",
"src/u8-vclamp/scalar-x4.c",
- "src/x8-lut/gen/lut-scalar-x4.c",
"src/x8-zip/x2-scalar.c",
"src/x8-zip/x3-scalar.c",
"src/x8-zip/x4-scalar.c",
"src/x8-zip/xm-scalar.c",
- "src/x32-depthtospace2d-chw2hwc/scalar.c",
"src/x32-packx/x2-scalar.c",
"src/x32-packx/x3-scalar.c",
"src/x32-packx/x4-scalar.c",
@@ -317,7 +414,6 @@
"src/x32-zip/x3-scalar.c",
"src/x32-zip/x4-scalar.c",
"src/x32-zip/xm-scalar.c",
- "src/xx-copy/memcpy.c",
"src/xx-fill/scalar-x16.c",
"src/xx-pad/scalar.c",
]
@@ -6717,11 +6813,15 @@
xnnpack_cc_library(
name = "scalar_prod_microkernels",
- srcs = PROD_SCALAR_MICROKERNEL_SRCS,
+ srcs = PROD_SCALAR_PORTABLE_MICROKERNEL_SRCS,
hdrs = INTERNAL_HDRS,
aarch32_copts = ["-marm"],
+ aarch32_srcs = PROD_SCALAR_AARCH32_MICROKERNEL_SRCS,
gcc_copts = xnnpack_gcc_std_copts(),
msvc_copts = xnnpack_msvc_std_copts(),
+ wasm_srcs = PROD_SCALAR_WASM_MICROKERNEL_SRCS,
+ wasmrelaxedsimd_srcs = PROD_SCALAR_WASM_MICROKERNEL_SRCS,
+ wasmsimd_srcs = PROD_SCALAR_WASM_MICROKERNEL_SRCS,
deps = [
":tables",
"@FP16",