vp9_f(dct|ht): disable avx2 variants

tests failing under Win32/Win64

+ dct16x16_test: add missing avx2 functions (partially disabled)

exercises the forward transforms
no idct/iht implementations, so the c-code is used

Change-Id: I04f64a457fa0828a00f32b5c9fe4f55294f21f61

test/dct16x16_test.cc

diff --git a/test/dct16x16_test.cc b/test/dct16x16_test.cc
index 99c8d0c..e6a20fb 100644
--- a/test/dct16x16_test.cc
+++ b/test/dct16x16_test.cc
@@ -606,4 +606,29 @@
     ::testing::Values(
         make_tuple(&vp9_fdct16x16_c, &vp9_idct16x16_256_add_ssse3, 0)));
 #endif
+
+#if HAVE_AVX2
+// TODO(jzern): these prototypes can be removed after the avx2 versions are
+// reenabled in vp9_rtcd_defs.pl.
+extern "C" {
+void vp9_fdct16x16_avx2(const int16_t *input, int16_t *output, int stride);
+void vp9_fht16x16_avx2(const int16_t *input, int16_t *output, int stride,
+                       int tx_type);
+}
+INSTANTIATE_TEST_CASE_P(
+    DISABLED_AVX2, Trans16x16DCT,
+    ::testing::Values(
+        make_tuple(&vp9_fdct16x16_avx2,
+                   &vp9_idct16x16_256_add_c, 0)));
+INSTANTIATE_TEST_CASE_P(
+    AVX2, Trans16x16HT,
+    ::testing::Values(
+        make_tuple(&vp9_fht16x16_avx2, &vp9_iht16x16_256_add_c, 3)));
+INSTANTIATE_TEST_CASE_P(
+    DISABLED_AVX2, Trans16x16HT,
+    ::testing::Values(
+        make_tuple(&vp9_fht16x16_avx2, &vp9_iht16x16_256_add_c, 0),
+        make_tuple(&vp9_fht16x16_avx2, &vp9_iht16x16_256_add_c, 1),
+        make_tuple(&vp9_fht16x16_avx2, &vp9_iht16x16_256_add_c, 2)));
+#endif
 }  // namespace

vp9/common/vp9_rtcd_defs.pl

diff --git a/vp9/common/vp9_rtcd_defs.pl b/vp9/common/vp9_rtcd_defs.pl
index d5ba6a2..e5358a7 100644
--- a/vp9/common/vp9_rtcd_defs.pl
+++ b/vp9/common/vp9_rtcd_defs.pl
@@ -739,7 +739,7 @@
 specialize qw/vp9_fht8x8 sse2 avx2/;
 
 add_proto qw/void vp9_fht16x16/, "const int16_t *input, int16_t *output, int stride, int tx_type";
-specialize qw/vp9_fht16x16 sse2 avx2/;
+specialize qw/vp9_fht16x16 sse2/;
 
 add_proto qw/void vp9_fwht4x4/, "const int16_t *input, int16_t *output, int stride";
 specialize qw/vp9_fwht4x4/, "$mmx_x86inc";
@@ -751,7 +751,7 @@
 specialize qw/vp9_fdct8x8 sse2 avx2/, "$ssse3_x86_64";
 
 add_proto qw/void vp9_fdct16x16/, "const int16_t *input, int16_t *output, int stride";
-specialize qw/vp9_fdct16x16 sse2 avx2/;
+specialize qw/vp9_fdct16x16 sse2/;
 
 add_proto qw/void vp9_fdct32x32/, "const int16_t *input, int16_t *output, int stride";
 specialize qw/vp9_fdct32x32 sse2 avx2/;