Merge "Update Android.bp after XNNPACK rebase"
diff --git a/Android.bp b/Android.bp
index 1343ca9..6f514e2 100644
--- a/Android.bp
+++ b/Android.bp
@@ -45,12 +45,11 @@
     "src/f32-argmaxpool/4x-scalar-c1.c",
     "src/f32-argmaxpool/9p8x-scalar-c1.c",
     "src/f32-argmaxpool/9x-scalar-c1.c",
-    "src/f32-avgpool/mp9p8q-scalar.c",
-    "src/f32-avgpool/up9-scalar.c",
-    "src/f32-bilinear/gen/scalar-c1.c",
-    "src/f32-bilinear/gen/scalar-c2.c",
-    "src/f32-bilinear/gen/scalar-c4.c",
-    "src/f32-clamp/scalar.c",
+    "src/f32-avgpool/9p8x-scalar-c1.c",
+    "src/f32-avgpool/9x-scalar-c1.c",
+    "src/f32-clamp/gen/scalar-x1.c",
+    "src/f32-clamp/gen/scalar-x2.c",
+    "src/f32-clamp/gen/scalar-x4.c",
     "src/f32-conv-hwc/3x3s2p0p1c3x4-scalar-1x1.c",
     "src/f32-conv-hwc/3x3s2p1c3x4-scalar-1x1.c",
     "src/f32-conv-hwc2spchw/3x3s2p1c3x4-scalar-1x1.c",
@@ -71,8 +70,8 @@
     "src/f32-dwconv/gen/up2x9-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x9-scalar.c",
     "src/f32-gavgpool-spchw/scalar-x1.c",
-    "src/f32-gavgpool/mp7p7q-scalar.c",
-    "src/f32-gavgpool/up7-scalar.c",
+    "src/f32-gavgpool/7p7x-scalar-c1.c",
+    "src/f32-gavgpool/7x-scalar-c1.c",
     "src/f32-gemm/gen-inc/1x4-scalar.c",
     "src/f32-gemm/gen-inc/2x4-scalar.c",
     "src/f32-gemm/gen-inc/4x4-scalar.c",
@@ -83,13 +82,16 @@
     "src/f32-hswish/gen/scalar-x1.c",
     "src/f32-hswish/gen/scalar-x2.c",
     "src/f32-hswish/gen/scalar-x4.c",
+    "src/f32-ibilinear/gen/scalar-c1.c",
+    "src/f32-ibilinear/gen/scalar-c2.c",
+    "src/f32-ibilinear/gen/scalar-c4.c",
     "src/f32-igemm/gen/1x4-scalar.c",
     "src/f32-igemm/gen/2x4-scalar.c",
     "src/f32-igemm/gen/4x2-scalar.c",
     "src/f32-igemm/gen/4x4-scalar.c",
     "src/f32-maxpool/9p8x-scalar-c1.c",
-    "src/f32-pavgpool/mp9p8q-scalar.c",
-    "src/f32-pavgpool/up9-scalar.c",
+    "src/f32-pavgpool/9p8x-scalar-c1.c",
+    "src/f32-pavgpool/9x-scalar-c1.c",
     "src/f32-ppmm/gen/2x4-scalar.c",
     "src/f32-ppmm/gen/3x3-scalar.c",
     "src/f32-ppmm/gen/4x2-scalar.c",
@@ -179,15 +181,15 @@
     "src/math/sigmoid-scalar-lut2048-p1-div.c",
     "src/math/sigmoid-scalar-lut64-p2-div.c",
     "src/math/sigmoid-scalar-p5-div.c",
-    "src/q8-avgpool/mp9p8q-scalar.c",
-    "src/q8-avgpool/up9-scalar.c",
+    "src/q8-avgpool/9p8x-scalar-c1.c",
+    "src/q8-avgpool/9x-scalar-c1.c",
     "src/q8-dwconv/up1x9-scalar.c",
-    "src/q8-gavgpool/mp7p7q-scalar.c",
-    "src/q8-gavgpool/up7-scalar.c",
+    "src/q8-gavgpool/7p7x-scalar-c1.c",
+    "src/q8-gavgpool/7x-scalar-c1.c",
     "src/q8-gemm/2x2-scalar.c",
     "src/q8-igemm/2x2-scalar.c",
     "src/q8-vadd/scalar.c",
-    "src/u8-clamp/scalar.c",
+    "src/u8-clamp/scalar-x4.c",
     "src/u8-lut32norm/scalar.c",
     "src/u8-maxpool/9p8x-scalar-c1.c",
     "src/u8-rmax/scalar.c",
@@ -205,17 +207,20 @@
     "src/x8-zip/x3-scalar.c",
     "src/x8-zip/x4-scalar.c",
     "src/x8-zip/xm-scalar.c",
+    "src/requantization/precise-scalar.c",
+    "src/requantization/fp32-scalar.c",
+    "src/requantization/q31-scalar.c",
+    "src/requantization/gemmlowp-scalar.c",
 ]
 
 PSIMD_FASTMATH_UKERNELS = [
     "src/f32-argmaxpool/4x-psimd-c4.c",
     "src/f32-argmaxpool/9p8x-psimd-c4.c",
     "src/f32-argmaxpool/9x-psimd-c4.c",
-    "src/f32-avgpool/mp9p8q-psimd.c",
-    "src/f32-avgpool/up9-psimd.c",
-    "src/f32-bilinear/gen/psimd-c4.c",
-    "src/f32-bilinear/gen/psimd-c8.c",
-    "src/f32-clamp/psimd.c",
+    "src/f32-avgpool/9p8x-psimd-c4.c",
+    "src/f32-avgpool/9x-psimd-c4.c",
+    "src/f32-clamp/gen/psimd-x4.c",
+    "src/f32-clamp/gen/psimd-x8.c",
     "src/f32-dwconv/gen/up4x25-psimd-acc2.c",
     "src/f32-dwconv/gen/up4x25-psimd.c",
     "src/f32-dwconv/gen/up4x4-psimd-acc2.c",
@@ -228,11 +233,12 @@
     "src/f32-dwconv/gen/up8x4-psimd.c",
     "src/f32-dwconv/gen/up8x9-psimd-acc2.c",
     "src/f32-dwconv/gen/up8x9-psimd.c",
-    "src/f32-gavgpool/mp7p7q-psimd.c",
-    "src/f32-gavgpool/up7-psimd.c",
+    "src/f32-gavgpool/7p7x-psimd-c4.c",
+    "src/f32-gavgpool/7x-psimd-c4.c",
     "src/f32-gemm/gen/1x8-psimd-loadsplat.c",
     "src/f32-gemm/gen/1x8-psimd-splat.c",
     "src/f32-gemm/gen/1x8s4-psimd.c",
+    "src/f32-gemm/gen/4x2c4-psimd.c",
     "src/f32-gemm/gen/4x8-psimd-loadsplat.c",
     "src/f32-gemm/gen/4x8-psimd-splat.c",
     "src/f32-gemm/gen/4x8s4-psimd.c",
@@ -250,6 +256,8 @@
     "src/f32-gemm/gen-inc/6x8s4-psimd.c",
     "src/f32-hswish/gen/psimd-x4.c",
     "src/f32-hswish/gen/psimd-x8.c",
+    "src/f32-ibilinear/gen/psimd-c4.c",
+    "src/f32-ibilinear/gen/psimd-c8.c",
     "src/f32-igemm/gen/1x8-psimd-loadsplat.c",
     "src/f32-igemm/gen/1x8-psimd-splat.c",
     "src/f32-igemm/gen/1x8s4-psimd.c",
@@ -261,8 +269,8 @@
     "src/f32-igemm/gen/6x8-psimd-splat.c",
     "src/f32-igemm/gen/6x8s4-psimd.c",
     "src/f32-maxpool/9p8x-psimd-c4.c",
-    "src/f32-pavgpool/mp9p8q-psimd.c",
-    "src/f32-pavgpool/up9-psimd.c",
+    "src/f32-pavgpool/9p8x-psimd-c4.c",
+    "src/f32-pavgpool/9x-psimd-c4.c",
     "src/f32-ppmm/gen/4x8-psimd.c",
     "src/f32-prelu/gen/psimd-2x4.c",
     "src/f32-prelu/gen/psimd-2x8.c",
@@ -304,6 +312,8 @@
     "src/x32-zip/x3-psimd.c",
     "src/x32-zip/x4-psimd.c",
     "src/x32-zip/xm-psimd.c",
+    "src/requantization/precise-psimd.c",
+    "src/requantization/fp32-psimd.c",
 ]
 
 PSIMD_ACCMATH_UKERNELS = [
@@ -330,18 +340,17 @@
 
 // ISA-specific micro-kernels
 NEON_UKERNELS = [
-    "src/f32-avgpool/mp9p8q-neon.c",
-    "src/f32-avgpool/up9-neon.c",
-    "src/f32-bilinear/gen/neon-c4.c",
-    "src/f32-bilinear/gen/neon-c8.c",
-    "src/f32-clamp/neon.c",
+    "src/f32-avgpool/9p8x-neon-c4.c",
+    "src/f32-avgpool/9x-neon-c4.c",
+    "src/f32-clamp/gen/neon-x4.c",
+    "src/f32-clamp/gen/neon-x8.c",
     "src/f32-dwconv/gen/up4x9-neon.c",
     "src/f32-dwconv/gen/up4x9-neon-acc2.c",
     "src/f32-dwconv/gen/up8x9-neon.c",
     "src/f32-dwconv/gen/up8x9-neon-acc2.c",
     "src/f32-gavgpool-spchw/neon-x4.c",
-    "src/f32-gavgpool/mp7p7q-neon.c",
-    "src/f32-gavgpool/up7-neon.c",
+    "src/f32-gavgpool/7p7x-neon-c4.c",
+    "src/f32-gavgpool/7x-neon-c4.c",
     "src/f32-gemm/gen/1x8-neon-lane-ld64.c",
     "src/f32-gemm/gen/4x2-neon-lane-ld64.c",
     "src/f32-gemm/gen/4x8-neon-lane-ld128.c",
@@ -375,6 +384,8 @@
     "src/f32-gemm/gen-inc/8x8s4-neon.c",
     "src/f32-hswish/gen/neon-x4.c",
     "src/f32-hswish/gen/neon-x8.c",
+    "src/f32-ibilinear/gen/neon-c4.c",
+    "src/f32-ibilinear/gen/neon-c8.c",
     "src/f32-igemm/gen/1x8-neon-lane-ld64.c",
     "src/f32-igemm/gen/4x2-neon-lane-ld64.c",
     "src/f32-igemm/gen/4x4-neon-lane-ld64.c",
@@ -391,8 +402,9 @@
     "src/f32-igemm/gen/4x8s4-neon.c",
     "src/f32-igemm/gen/6x8s4-neon.c",
     "src/f32-igemm/gen/8x8s4-neon.c",
-    "src/f32-pavgpool/mp9p8q-neon.c",
-    "src/f32-pavgpool/up9-neon.c",
+    "src/f32-maxpool/9p8x-neon-c4.c",
+    "src/f32-pavgpool/9p8x-neon-c4.c",
+    "src/f32-pavgpool/9x-neon-c4.c",
     "src/f32-ppmm/gen/4x8-neon.c",
     "src/f32-ppmm/gen/8x8-neon.c",
     "src/f32-prelu/gen/neon-2x4.c",
@@ -465,17 +477,17 @@
     "src/f32-vbinary/gen/vsubc-neon-x8.c",
     "src/f32-vmulcaddc/gen/c4-neon-2x.c",
     "src/f32-vmulcaddc/gen/c8-neon-2x.c",
-    "src/q8-avgpool/mp9p8q-neon.c",
-    "src/q8-avgpool/up9-neon.c",
+    "src/q8-avgpool/9p8x-neon-c8.c",
+    "src/q8-avgpool/9x-neon-c8.c",
     "src/q8-dwconv/up8x9-neon.c",
-    "src/q8-gavgpool/mp7p7q-neon.c",
-    "src/q8-gavgpool/up7-neon.c",
+    "src/q8-gavgpool/7p7x-neon-c8.c",
+    "src/q8-gavgpool/7x-neon-c8.c",
     "src/q8-gemm/4x8-neon.c",
     "src/q8-gemm/8x8-neon.c",
     "src/q8-igemm/4x8-neon.c",
     "src/q8-igemm/8x8-neon.c",
     "src/q8-vadd/neon.c",
-    "src/u8-clamp/neon.c",
+    "src/u8-clamp/neon-x64.c",
     "src/u8-maxpool/9p8x-neon-c16.c",
     "src/u8-rmax/neon.c",
     "src/x32-packx/x4-neon-st4.c",
@@ -495,11 +507,15 @@
     "src/math/sigmoid-neon-rr2-lut2048-p1-nr2recps.c",
     "src/math/sigmoid-neon-rr2-lut64-p2-nr2recps.c",
     "src/math/sigmoid-neon-rr2-p5-nr2recps.c",
+    "src/requantization/precise-neon.c",
+    "src/requantization/fp32-neon.c",
+    "src/requantization/q31-neon.c",
+    "src/requantization/gemmlowp-neon.c",
 ]
 
 NEONFMA_UKERNELS = [
-    "src/f32-bilinear/gen/neonfma-c4.c",
-    "src/f32-bilinear/gen/neonfma-c8.c",
+    "src/f32-ibilinear/gen/neonfma-c4.c",
+    "src/f32-ibilinear/gen/neonfma-c8.c",
     "src/f32-igemm/gen/1x8-neonfma-dup-ld64.c",
     "src/f32-igemm/gen/4x8-neonfma-dup-ld128.c",
     "src/f32-igemm/gen/4x8-neonfma-dup-ld64.c",
@@ -725,11 +741,10 @@
 ]
 
 SSE_UKERNELS = [
-    "src/f32-avgpool/mp9p8q-sse.c",
-    "src/f32-avgpool/up9-sse.c",
-    "src/f32-bilinear/gen/sse-c4.c",
-    "src/f32-bilinear/gen/sse-c8.c",
-    "src/f32-clamp/sse.c",
+    "src/f32-avgpool/9p8x-sse-c4.c",
+    "src/f32-avgpool/9x-sse-c4.c",
+    "src/f32-clamp/gen/sse-x4.c",
+    "src/f32-clamp/gen/sse-x8.c",
     "src/f32-dwconv-spchw/3x3p1-sse.c",
     "src/f32-dwconv-spchw/3x3s2p1-sse.c",
     "src/f32-dwconv/gen/up4x25-sse-acc2.c",
@@ -745,11 +760,12 @@
     "src/f32-dwconv/gen/up8x9-sse-acc2.c",
     "src/f32-dwconv/gen/up8x9-sse.c",
     "src/f32-gavgpool-spchw/sse-x4.c",
-    "src/f32-gavgpool/mp7p7q-sse.c",
-    "src/f32-gavgpool/up7-sse.c",
+    "src/f32-gavgpool/7p7x-sse-c4.c",
+    "src/f32-gavgpool/7x-sse-c4.c",
     "src/f32-gemm/gen/1x8-sse-dup.c",
     "src/f32-gemm/gen/1x8-sse-load1.c",
     "src/f32-gemm/gen/1x8s4-sse.c",
+    "src/f32-gemm/gen/4x2c4-sse.c",
     "src/f32-gemm/gen/4x8-sse-dup.c",
     "src/f32-gemm/gen/4x8-sse-load1.c",
     "src/f32-gemm/gen/4x8s4-sse.c",
@@ -761,6 +777,8 @@
     "src/f32-gemm/gen-inc/4x8s4-sse.c",
     "src/f32-hswish/gen/sse-x4.c",
     "src/f32-hswish/gen/sse-x8.c",
+    "src/f32-ibilinear/gen/sse-c4.c",
+    "src/f32-ibilinear/gen/sse-c8.c",
     "src/f32-igemm/gen/1x8-sse-dup.c",
     "src/f32-igemm/gen/1x8-sse-load1.c",
     "src/f32-igemm/gen/1x8s4-sse.c",
@@ -769,8 +787,8 @@
     "src/f32-igemm/gen/4x8-sse-load1.c",
     "src/f32-igemm/gen/4x8s4-sse.c",
     "src/f32-maxpool/9p8x-sse-c4.c",
-    "src/f32-pavgpool/mp9p8q-sse.c",
-    "src/f32-pavgpool/up9-sse.c",
+    "src/f32-pavgpool/9p8x-sse-c4.c",
+    "src/f32-pavgpool/9x-sse-c4.c",
     "src/f32-ppmm/gen/4x8-sse.c",
     "src/f32-rmax/sse.c",
     "src/f32-spmm/gen/4x1-sse.c",
@@ -832,16 +850,16 @@
     "src/f32-sigmoid/gen/sse2-p5-div-x16.c",
     "src/f32-sigmoid/gen/sse2-p5-div-x20.c",
     "src/f32-sigmoid/gen/sse2-p5-div-x24.c",
-    "src/q8-avgpool/mp9p8q-sse2.c",
-    "src/q8-avgpool/up9-sse2.c",
+    "src/q8-avgpool/9p8x-sse2-c8.c",
+    "src/q8-avgpool/9x-sse2-c8.c",
     "src/q8-igemm/4x4c2-sse2.c",
     "src/q8-dwconv/up8x9-sse2.c",
-    "src/q8-gavgpool/mp7p7q-sse2.c",
-    "src/q8-gavgpool/up7-sse2.c",
+    "src/q8-gavgpool/7p7x-sse2-c8.c",
+    "src/q8-gavgpool/7x-sse2-c8.c",
     "src/q8-gemm/2x4c8-sse2.c",
     "src/q8-gemm/4x4c2-sse2.c",
     "src/q8-vadd/sse2.c",
-    "src/u8-clamp/sse2.c",
+    "src/u8-clamp/sse2-x64.c",
     "src/u8-maxpool/9p8x-sse2-c16.c",
     "src/u8-rmax/sse2.c",
     "src/x32-pad/x2-sse2.c",
@@ -856,6 +874,16 @@
     "src/math/exp-sse2-p5.c",
     "src/math/expminus-sse2-p5.c",
     "src/math/sigmoid-sse2-p5-div.c",
+    "src/requantization/precise-sse2.c",
+    "src/requantization/fp32-sse2.c",
+    "src/requantization/q31-sse2.c",
+    "src/requantization/gemmlowp-sse2.c",
+]
+
+SSSE3_UKERNELS = [
+    "src/requantization/precise-ssse3.c",
+    "src/requantization/q31-ssse3.c",
+    "src/requantization/gemmlowp-ssse3.c",
 ]
 
 SSE41_UKERNELS = [
@@ -867,10 +895,14 @@
     "src/f32-sigmoid/gen/sse41-p5-div-x16.c",
     "src/f32-sigmoid/gen/sse41-p5-div-x20.c",
     "src/f32-sigmoid/gen/sse41-p5-div-x24.c",
+    "src/requantization/precise-sse4.c",
+    "src/requantization/q31-sse4.c",
+    "src/requantization/gemmlowp-sse4.c",
 ]
 
 AVX_UKERNELS = [
-    "src/f32-clamp/avx.c",
+    "src/f32-clamp/gen/avx-x8.c",
+    "src/f32-clamp/gen/avx-x16.c",
     "src/f32-dwconv/gen/up16x4-avx-acc2.c",
     "src/f32-dwconv/gen/up16x4-avx.c",
     "src/f32-dwconv/gen/up8x4-avx-acc2.c",
@@ -912,6 +944,8 @@
     "src/f32-igemm/gen/3x16-avx-broadcast.c",
     "src/f32-igemm/gen/4x16-avx-broadcast.c",
     "src/f32-igemm/gen/5x16-avx-broadcast.c",
+    "src/f32-prelu/gen/avx-2x8.c",
+    "src/f32-prelu/gen/avx-2x16.c",
     "src/f32-rmax/avx.c",
     "src/f32-vbinary/gen/vadd-avx-x8.c",
     "src/f32-vbinary/gen/vadd-avx-x16.c",
@@ -1108,7 +1142,8 @@
 ]
 
 AVX512F_UKERNELS = [
-    "src/f32-clamp/avx512f.c",
+    "src/f32-clamp/gen/avx512f-x16.c",
+    "src/f32-clamp/gen/avx512f-x32.c",
     "src/f32-dwconv/gen/up32x4-avx512f-acc2.c",
     "src/f32-dwconv/gen/up32x4-avx512f.c",
     "src/f32-dwconv/gen/up16x4-avx512f-acc2.c",
@@ -1141,6 +1176,8 @@
     "src/f32-igemm/gen/6x16-avx512f-broadcast.c",
     "src/f32-igemm/gen/7x16-avx512f-broadcast.c",
     "src/f32-igemm/gen/8x16-avx512f-broadcast.c",
+    "src/f32-prelu/gen/avx512f-2x16.c",
+    "src/f32-prelu/gen/avx512f-2x32.c",
     "src/f32-raddexpminusmax/gen/avx512f-p5-scalef-x128.c",
     "src/f32-raddexpminusmax/gen/avx512f-p5-scalef-x128-acc2.c",
     "src/f32-raddexpminusmax/gen/avx512f-p5-scalef-x128-acc4.c",
@@ -1241,12 +1278,15 @@
 AARCH32_ASM_UKERNELS = [
     "src/q8-dwconv/up8x9-aarch32-neon.S",
     "src/f32-gemm/4x8-aarch32-neon-cortex-a53.S",
+    "src/f32-gemm/4x8-aarch32-neon-cortex-a55.S",
     "src/f32-gemm/gen/4x8-aarch32-neon-cortex-a75.S",
     "src/f32-gemm/gen/4x8-aarch32-neon-pld-cortex-a75.S",
     "src/f32-gemm/4x8-aarch32-neon-ld64.S",
     "src/f32-igemm/4x8-aarch32-neon-ld64.S",
     "src/f32-igemm/gen/4x8-aarch32-neon-cortex-a75.S",
     "src/f32-igemm/gen/4x8-aarch32-neon-pld-cortex-a75.S",
+    "src/f32-igemm/4x8-aarch32-neon-cortex-a53.S",
+    "src/f32-igemm/4x8-aarch32-neon-cortex-a55.S",
 ]
 
 AARCH64_ASM_UKERNELS = [
@@ -1265,6 +1305,7 @@
     "src/f32-gemm/gen/5x8-aarch64-neonfma-cortex-a57.S",
     "src/f32-gemm/gen/5x8-aarch64-neonfma-cortex-a75.S",
     "src/f32-gemm/gen/6x8-aarch64-neonfma-cortex-a53.S",
+    "src/f32-gemm/gen/6x8-aarch64-neonfma-cortex-a55.S",
     "src/f32-gemm/gen/6x8-aarch64-neonfma-cortex-a73.S",
     "src/f32-gemm/gen/6x8-aarch64-neonfma-cortex-a57.S",
     "src/f32-gemm/gen/6x8-aarch64-neonfma-cortex-a75.S",
@@ -1283,6 +1324,7 @@
     "src/f32-gemm/gen-inc/5x8-aarch64-neonfma-cortex-a57.S",
     "src/f32-gemm/gen-inc/5x8-aarch64-neonfma-cortex-a75.S",
     "src/f32-gemm/gen-inc/6x8-aarch64-neonfma-cortex-a53.S",
+    "src/f32-gemm/gen-inc/6x8-aarch64-neonfma-cortex-a55.S",
     "src/f32-gemm/gen-inc/6x8-aarch64-neonfma-cortex-a73.S",
     "src/f32-gemm/gen-inc/6x8-aarch64-neonfma-cortex-a57.S",
     "src/f32-gemm/gen-inc/6x8-aarch64-neonfma-cortex-a75.S",
@@ -1299,6 +1341,7 @@
     "src/f32-igemm/gen/5x8-aarch64-neonfma-cortex-a57.S",
     "src/f32-igemm/gen/5x8-aarch64-neonfma-cortex-a75.S",
     "src/f32-igemm/6x8-aarch64-neonfma-cortex-a53.S",
+    "src/f32-igemm/6x8-aarch64-neonfma-cortex-a55.S",
     "src/f32-igemm/6x8-aarch64-neonfma-cortex-a73.S",
     "src/f32-igemm/gen/6x8-aarch64-neonfma-cortex-a57.S",
     "src/f32-igemm/gen/6x8-aarch64-neonfma-cortex-a75.S",
@@ -1574,6 +1617,34 @@
 }
 
 cc_library_static {
+    name: "xnnpack_ssse3_ukernels",
+    defaults: ["xnnpack_internal_default"],
+    arch: {
+        arm: { enabled: false, },
+        arm64:  { enabled: false, },
+        x86: {
+            srcs: SSSE3_UKERNELS,
+            cflags: [
+                "-mssse3",
+            ],
+        },
+        x86_64: {
+            srcs: SSSE3_UKERNELS,
+            cflags: [
+                "-mssse3",
+            ],
+        },
+    },
+    header_libs: [
+        "fp16_headers",
+    ],
+    static_libs: [
+        "libpthreadpool",
+        "xnnpack_tables",
+    ],
+}
+
+cc_library_static {
     name: "xnnpack_sse41_ukernels",
     defaults: ["xnnpack_internal_default"],
     arch: {
@@ -1743,6 +1814,7 @@
                 "xnnpack_psimd_fastmath_ukernels",
                 "xnnpack_psimd_accmath_ukernels",
                 "xnnpack_sse2_ukernels",
+                "xnnpack_ssse3_ukernels",
                 "xnnpack_sse41_ukernels",
                 "xnnpack_avx_ukernels",
                 "xnnpack_fma3_ukernels",
@@ -1755,6 +1827,7 @@
                 "xnnpack_psimd_fastmath_ukernels",
                 "xnnpack_psimd_accmath_ukernels",
                 "xnnpack_sse2_ukernels",
+                "xnnpack_ssse3_ukernels",
                 "xnnpack_sse41_ukernels",
                 "xnnpack_avx_ukernels",
                 "xnnpack_fma3_ukernels",