CMake build fix

- Move scalar kernels from SSE source section to scalar section.
- Copy scalar source file list from CMake to BUILD so they match.

PiperOrigin-RevId: 325564233
diff --git a/BUILD.bazel b/BUILD.bazel
index cf5b7fb..4a6a0c0 100644
--- a/BUILD.bazel
+++ b/BUILD.bazel
@@ -121,43 +121,40 @@
     "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-hwc/3x3s2p0p1c3x4-scalar-1x1.c",
     "src/f32-conv-hwc2chw/3x3s2p1c3x4-scalar-1x1.c",
     "src/f32-dwconv-chw/3x3p1-scalar.c",
     "src/f32-dwconv-chw/3x3s2p1-scalar.c",
     "src/f32-dwconv-chw/5x5p2-scalar.c",
     "src/f32-dwconv-chw/5x5s2p2-scalar.c",
-    "src/f32-dwconv/gen/up1x4-scalar-acc2.c",
     "src/f32-dwconv/gen/up1x4-scalar.c",
-    "src/f32-dwconv/gen/up1x9-scalar-acc2.c",
+    "src/f32-dwconv/gen/up1x4-scalar-acc2.c",
     "src/f32-dwconv/gen/up1x9-scalar.c",
-    "src/f32-dwconv/gen/up1x25-scalar-acc2.c",
+    "src/f32-dwconv/gen/up1x9-scalar-acc2.c",
     "src/f32-dwconv/gen/up1x25-scalar.c",
-    "src/f32-dwconv/gen/up2x4-scalar-acc2.c",
+    "src/f32-dwconv/gen/up1x25-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x4-scalar.c",
-    "src/f32-dwconv/gen/up2x9-scalar-acc2.c",
+    "src/f32-dwconv/gen/up2x4-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x9-scalar.c",
-    "src/f32-dwconv/gen/up2x25-scalar-acc2.c",
+    "src/f32-dwconv/gen/up2x9-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x25-scalar.c",
-    "src/f32-dwconv/gen/up1x4-minmax-scalar-acc2.c",
+    "src/f32-dwconv/gen/up2x25-scalar-acc2.c",
     "src/f32-dwconv/gen/up1x4-minmax-scalar.c",
-    "src/f32-dwconv/gen/up1x9-minmax-scalar-acc2.c",
+    "src/f32-dwconv/gen/up1x4-minmax-scalar-acc2.c",
     "src/f32-dwconv/gen/up1x9-minmax-scalar.c",
-    "src/f32-dwconv/gen/up1x25-minmax-scalar-acc2.c",
+    "src/f32-dwconv/gen/up1x9-minmax-scalar-acc2.c",
     "src/f32-dwconv/gen/up1x25-minmax-scalar.c",
-    "src/f32-dwconv/gen/up2x4-minmax-scalar-acc2.c",
+    "src/f32-dwconv/gen/up1x25-minmax-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x4-minmax-scalar.c",
-    "src/f32-dwconv/gen/up2x9-minmax-scalar-acc2.c",
+    "src/f32-dwconv/gen/up2x4-minmax-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x9-minmax-scalar.c",
-    "src/f32-dwconv/gen/up2x25-minmax-scalar-acc2.c",
+    "src/f32-dwconv/gen/up2x9-minmax-scalar-acc2.c",
     "src/f32-dwconv/gen/up2x25-minmax-scalar.c",
+    "src/f32-dwconv/gen/up2x25-minmax-scalar-acc2.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-inc/1x4inc-minmax-scalar.c",
-    "src/f32-gemm/gen-inc/2x4inc-minmax-scalar.c",
-    "src/f32-gemm/gen-inc/4x4inc-minmax-scalar.c",
     "src/f32-gemm/gen/1x4-scalar.c",
     "src/f32-gemm/gen/2x4-scalar.c",
     "src/f32-gemm/gen/4x2-scalar.c",
@@ -170,6 +167,9 @@
     "src/f32-gemm/gen/2x4-minmax-scalar.c",
     "src/f32-gemm/gen/4x2-minmax-scalar.c",
     "src/f32-gemm/gen/4x4-minmax-scalar.c",
+    "src/f32-gemm/gen-inc/1x4inc-minmax-scalar.c",
+    "src/f32-gemm/gen-inc/2x4inc-minmax-scalar.c",
+    "src/f32-gemm/gen-inc/4x4inc-minmax-scalar.c",
     "src/f32-hswish/gen/hswish-scalar-x1.c",
     "src/f32-hswish/gen/hswish-scalar-x2.c",
     "src/f32-hswish/gen/hswish-scalar-x4.c",
@@ -188,6 +188,9 @@
     "src/f32-igemm/gen/2x4-minmax-scalar.c",
     "src/f32-igemm/gen/4x2-minmax-scalar.c",
     "src/f32-igemm/gen/4x4-minmax-scalar.c",
+    "src/f32-vlrelu/gen/vlrelu-scalar-x1.c",
+    "src/f32-vlrelu/gen/vlrelu-scalar-x2.c",
+    "src/f32-vlrelu/gen/vlrelu-scalar-x4.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",
@@ -209,10 +212,10 @@
     "src/f32-raddstoreexpminusmax/gen/scalar-p5-x4.c",
     "src/f32-raddstoreexpminusmax/gen/scalar-p5-x4-acc2.c",
     "src/f32-raddstoreexpminusmax/gen/scalar-p5-x4-acc4.c",
+    "src/f32-rmax/scalar.c",
     "src/f32-relu/gen/scalar-x1.c",
     "src/f32-relu/gen/scalar-x2.c",
     "src/f32-relu/gen/scalar-x4.c",
-    "src/f32-rmax/scalar.c",
     "src/f32-sigmoid/gen/scalar-lut2048-p1-div-x1.c",
     "src/f32-sigmoid/gen/scalar-lut2048-p1-div-x2.c",
     "src/f32-sigmoid/gen/scalar-lut2048-p1-div-x4.c",
@@ -310,24 +313,21 @@
     "src/f32-vbinary/gen/vsubc-scalar-x1.c",
     "src/f32-vbinary/gen/vsubc-scalar-x2.c",
     "src/f32-vbinary/gen/vsubc-scalar-x4.c",
-    "src/f32-vlrelu/gen/vlrelu-scalar-x1.c",
-    "src/f32-vlrelu/gen/vlrelu-scalar-x2.c",
-    "src/f32-vlrelu/gen/vlrelu-scalar-x4.c",
     "src/f32-vmulcaddc/gen/c1-minmax-scalar-2x.c",
     "src/f32-vmulcaddc/gen/c2-minmax-scalar-2x.c",
     "src/f32-vmulcaddc/gen/c4-minmax-scalar-2x.c",
-    "src/f32-vrnd/gen/vrndne-scalar-libm-x1.c",
-    "src/f32-vrnd/gen/vrndne-scalar-libm-x2.c",
-    "src/f32-vrnd/gen/vrndne-scalar-libm-x4.c",
-    "src/f32-vrnd/gen/vrndz-scalar-libm-x1.c",
-    "src/f32-vrnd/gen/vrndz-scalar-libm-x2.c",
-    "src/f32-vrnd/gen/vrndz-scalar-libm-x4.c",
-    "src/f32-vrnd/gen/vrndu-scalar-libm-x1.c",
-    "src/f32-vrnd/gen/vrndu-scalar-libm-x2.c",
-    "src/f32-vrnd/gen/vrndu-scalar-libm-x4.c",
     "src/f32-vrnd/gen/vrndd-scalar-libm-x1.c",
     "src/f32-vrnd/gen/vrndd-scalar-libm-x2.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-x2.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-x2.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-x2.c",
+    "src/f32-vrnd/gen/vrndz-scalar-libm-x4.c",
     "src/f32-vsqrt/gen/scalar-sqrt-x1.c",
     "src/f32-vsqrt/gen/scalar-sqrt-x2.c",
     "src/f32-vsqrt/gen/scalar-sqrt-x4.c",
@@ -358,19 +358,21 @@
     "src/math/sigmoid-scalar-lut2048-p1-div.c",
     "src/math/sigmoid-scalar-lut64-p2-div.c",
     "src/math/sigmoid-scalar-p5-div.c",
+    "src/qs8-gemm/gen/8x8c4-minmax-scalar.c",
+    "src/qs8-gemm/gen/12x4c4-minmax-scalar.c",
     "src/qs8-requantization/fp32-scalar-lrintf.c",
     "src/qs8-requantization/fp32-scalar-magic.c",
     "src/qs8-requantization/precise-scalar-signed64.c",
     "src/qs8-requantization/precise-scalar-unsigned32.c",
     "src/qs8-requantization/precise-scalar-unsigned64.c",
     "src/qs8-requantization/q31-scalar.c",
-    "src/qs8-gemm/gen/8x8c4-minmax-scalar.c",
-    "src/qs8-gemm/gen/12x4c4-minmax-scalar.c",
     "src/qu8-avgpool/9p8x-minmax-scalar-c1.c",
     "src/qu8-avgpool/9x-minmax-scalar-c1.c",
     "src/qu8-dwconv/up1x9-minmax-scalar.c",
     "src/qu8-gavgpool/7p7x-minmax-scalar-c1.c",
     "src/qu8-gavgpool/7x-minmax-scalar-c1.c",
+    "src/qu8-gemm/gen/8x8c4-minmax-scalar.c",
+    "src/qu8-gemm/gen/12x4c4-minmax-scalar.c",
     "src/qu8-gemm/2x2-minmax-scalar.c",
     "src/qu8-igemm/2x2-minmax-scalar.c",
     "src/qu8-requantization/fp32-scalar-lrintf.c",
@@ -380,8 +382,6 @@
     "src/qu8-requantization/precise-scalar-unsigned64.c",
     "src/qu8-requantization/q31-scalar.c",
     "src/qu8-vadd/minmax-scalar.c",
-    "src/qu8-gemm/gen/8x8c4-minmax-scalar.c",
-    "src/qu8-gemm/gen/12x4c4-minmax-scalar.c",
     "src/u8-clamp/scalar-x4.c",
     "src/u8-lut32norm/scalar.c",
     "src/u8-maxpool/9p8x-minmax-scalar-c1.c",
@@ -391,8 +391,8 @@
     "src/x32-packx/x2-scalar.c",
     "src/x32-packx/x3-scalar.c",
     "src/x32-packx/x4-scalar.c",
-    "src/x32-pad/scalar-int.c",
     "src/x32-pad/scalar-float.c",
+    "src/x32-pad/scalar-int.c",
     "src/x32-unpool/scalar.c",
     "src/x32-zip/x2-scalar.c",
     "src/x32-zip/x3-scalar.c",