arm neon: disable some FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics

while we figure out the situation in #1099 vcvth_s16_f16 vcvtah_s16_f16 vcvth_n_s32_f16 vcvth_n_s64_f16 vcvth_n_u32_f16 vcvth_n_u64_f16 vcvth_n_f16_s32 vcvth_n_f16_s64 vcvth_n_f16_u32 vcvth_n_f16_u64 vcvtmh_s16_f16 vcvtms_s32_f32 vcvtmd_s64_f64 vcvtnh_s16_f16 vcvtph_s16_f16 vcvtps_s32_f32 vcvtpd_s64_f64 Above removed due to test failures on real hardware vcvt_s16_f16 vcvtq_s16_f16 vcvth_n_s16_f16 vcvt_n_s16_f16 vcvtq_n_s16_f16 (fallback implementations call vcvth_s16_f16) vcvta_s16_f16 vcvtaq_s16_f16 (fallback implementations call vcvtah_s16_f16) vcvtmq_s32_f32 vcvtm_s32_f32 (fallback implementations call vcvtms_s32_f32) vcvtmq_s64_f64 vcvtm_s64_f64 (fallback implementations call vcvtmd_s64_f64) vcvtnq_s16_f16 vcvtn_s16_f16 (fallback implementations call vcvtnh_s16_f16) vcvtpq_s16_f16 vcvtp_s16_f16 (fallback implementations call vcvtph_s16_f16) vcvtpq_s32_f32 vcvtp_s32_f32 (fallback implementations call vcvtps_s32_f32) vcvtpq_s64_f64 vcvtp_s64_f64 (fallback implementations call vcvtpd_s64_f64) vcvtmq_s16_f16 vcvtm_s16_f16 (fallback implementations call vcvtmh_s16_f16)
simd-everywhere · Mar 20, 2024 · 339ffe4 · 339ffe4
1 parent 7ab3240
commit 339ffe4
Show file tree

Hide file tree

Showing 10 changed files with 188 additions and 41 deletions.
diff --git a/simde/arm/neon/cvt.h b/simde/arm/neon/cvt.h
@@ -139,6 +139,8 @@ simde_vcvt_f64_f32(simde_float32x2_t a) {
   #define vcvt_f64_f32(a) simde_vcvt_f64_f32(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int16_t
 simde_vcvth_s16_f16(simde_float16_t a) {
@@ -164,6 +166,7 @@ simde_vcvth_s16_f16(simde_float16_t a) {
   #undef vcvth_s16_f16
   #define vcvth_s16_f16(a) simde_vcvth_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint16_t
@@ -544,6 +547,8 @@ simde_vcvth_f16_u16(uint16_t a) {
   #define vcvth_f16_u16(a) simde_vcvth_f16_u16(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x4_t
 simde_vcvt_s16_f16(simde_float16x4_t a) {
@@ -569,6 +574,7 @@ simde_vcvt_s16_f16(simde_float16x4_t a) {
   #undef vcvt_s16_f16
   #define vcvt_s16_f16(a) simde_vcvt_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int32x2_t
@@ -701,6 +707,8 @@ simde_vcvt_u64_f64(simde_float64x1_t a) {
   #define vcvt_u64_f64(a) simde_vcvt_u64_f64(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x8_t
 simde_vcvtq_s16_f16(simde_float16x8_t a) {
@@ -726,6 +734,7 @@ simde_vcvtq_s16_f16(simde_float16x8_t a) {
   #undef vcvtq_s16_f16
   #define vcvtq_s16_f16(a) simde_vcvtq_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int32x4_t
@@ -1383,6 +1392,8 @@ simde_vcvtq_f64_u64(simde_uint64x2_t a) {
   #define vcvtq_f64_u64(a) simde_vcvtq_f64_u64(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int16_t
 simde_vcvtah_s16_f16(simde_float16_t a) {
@@ -1408,6 +1419,7 @@ simde_vcvtah_s16_f16(simde_float16_t a) {
   #undef vcvtah_s16_f16
   #define vcvtah_s16_f16(a) simde_vcvtah_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint16_t
@@ -1636,6 +1648,8 @@ simde_vcvtas_u32_f32(simde_float32 a) {
   #define vcvtas_u32_f32(a) simde_vcvtas_u32_f32(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x4_t
 simde_vcvta_s16_f16(simde_float16x4_t a) {
@@ -1657,6 +1671,7 @@ simde_vcvta_s16_f16(simde_float16x4_t a) {
   #undef vcvta_s16_f16
   #define vcvta_s16_f16(a) simde_vcvta_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_uint16x4_t
@@ -1746,6 +1761,8 @@ simde_vcvta_s32_f32(simde_float32x2_t a) {
   #define vcvta_s32_f32(a) simde_vcvta_s32_f32(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x8_t
 simde_vcvtaq_s16_f16(simde_float16x8_t a) {
@@ -1767,7 +1784,7 @@ simde_vcvtaq_s16_f16(simde_float16x8_t a) {
   #undef vcvtaq_s16_f16
   #define vcvtaq_s16_f16(a) simde_vcvtaq_s16_f16(a)
 #endif
-
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_uint16x8_t

diff --git a/simde/arm/neon/cvt_n.h b/simde/arm/neon/cvt_n.h
@@ -34,6 +34,8 @@ HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
 SIMDE_BEGIN_DECLS_
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int16_t
 simde_vcvth_n_s16_f16(simde_float16_t a, const int n)
@@ -81,6 +83,7 @@ simde_vcvth_n_s64_f16(simde_float16_t a, const int n)
   #undef vcvth_n_s64_f16
   #define vcvth_n_s64_f16(a, n) simde_vcvth_n_s64_f16(a, n)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint16_t
@@ -98,6 +101,8 @@ simde_vcvth_n_u16_f16(simde_float16_t a, const int n)
   #define vcvth_n_u16_f16(a, n) simde_vcvth_n_u16_f16(a, n)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 uint32_t
 simde_vcvth_n_u32_f16(simde_float16_t a, const int n)
@@ -129,6 +134,7 @@ simde_vcvth_n_u64_f16(simde_float16_t a, const int n)
   #undef vcvth_n_u64_f16
   #define vcvth_n_u64_f16(a, n) simde_vcvth_n_u64_f16(a, n)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16_t
@@ -146,6 +152,8 @@ simde_vcvth_n_f16_s16(int16_t a, const int n)
   #define vcvth_n_f16_s16(a, n) simde_vcvth_n_f16_s16(a, n)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16_t
 simde_vcvth_n_f16_s32(int32_t a, const int n)
@@ -177,6 +185,7 @@ simde_vcvth_n_f16_s64(int64_t a, const int n)
   #undef vcvth_n_f16_s64
   #define vcvth_n_f16_s64(a, n) simde_vcvth_n_f16_s64(a, n)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16_t
@@ -194,6 +203,8 @@ simde_vcvth_n_f16_u16(uint16_t a, const int n)
   #define vcvth_n_f16_u16(a, n) simde_vcvth_n_f16_u16(a, n)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16_t
 simde_vcvth_n_f16_u32(uint32_t a, const int n)
@@ -225,6 +236,7 @@ simde_vcvth_n_f16_u64(uint64_t a, const int n)
   #undef vcvth_n_f16_u64
   #define vcvth_n_f16_u64(a, n) simde_vcvth_n_f16_u64(a, n)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 int32_t
@@ -340,6 +352,8 @@ simde_vcvtd_n_f64_u64(uint64_t a, const int n)
   #define vcvtd_n_f64_u64(a, n) simde_vcvtd_n_f64_u64(a, n)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x4_t
 simde_vcvt_n_s16_f16(simde_float16x4_t a, const int n)
@@ -363,6 +377,7 @@ simde_vcvt_n_s16_f16(simde_float16x4_t a, const int n)
   #undef vcvt_n_s16_f16
   #define vcvt_n_s16_f16(a, n) simde_vcvt_n_s16_f16((a), (n))
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int32x2_t
@@ -476,6 +491,8 @@ simde_vcvt_n_u64_f64(simde_float64x1_t a, const int n)
   #define vcvt_n_u64_f64(a, n) simde_vcvt_n_u64_f64((a), (n))
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x8_t
 simde_vcvtq_n_s16_f16(simde_float16x8_t a, const int n)
@@ -499,6 +516,7 @@ simde_vcvtq_n_s16_f16(simde_float16x8_t a, const int n)
   #undef vcvtq_n_s16_f16
   #define vcvtq_n_s16_f16(a, n) simde_vcvtq_n_s16_f16((a), (n))
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int32x4_t

diff --git a/simde/arm/neon/cvtm.h b/simde/arm/neon/cvtm.h
@@ -88,6 +88,8 @@ simde_vcvtmh_s32_f16(simde_float16_t a) {
   #define vcvtmh_s32_f16(a) simde_vcvtmh_s32_f16(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int16_t
 simde_vcvtmh_s16_f16(simde_float16_t a) {
@@ -114,6 +116,7 @@ simde_vcvtmh_s16_f16(simde_float16_t a) {
   #undef vcvtmh_s16_f16
   #define vcvtmh_s16_f16(a) simde_vcvtmh_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint64_t
@@ -196,6 +199,8 @@ simde_vcvtmh_u16_f16(simde_float16_t a) {
   #define vcvtmh_u16_f16(a) simde_vcvtmh_u16_f16(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int32_t
 simde_vcvtms_s32_f32(simde_float32 a) {
@@ -219,6 +224,7 @@ simde_vcvtms_s32_f32(simde_float32 a) {
   #undef vcvtms_s32_f32
   #define vcvtms_s32_f32(a) simde_vcvtms_s32_f32(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint32_t
@@ -244,6 +250,8 @@ simde_vcvtms_u32_f32(simde_float32 a) {
   #define vcvtms_u32_f32(a) simde_vcvtms_u32_f32(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int64_t
 simde_vcvtmd_s64_f64(simde_float64 a) {
@@ -267,6 +275,7 @@ simde_vcvtmd_s64_f64(simde_float64 a) {
   #undef vcvtmd_s64_f64
   #define vcvtmd_s64_f64(a) simde_vcvtmd_s64_f64(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint64_t
@@ -292,6 +301,8 @@ simde_vcvtmd_u64_f64(simde_float64 a) {
   #define vcvtmd_u64_f64(a) simde_vcvtmd_u64_f64(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x8_t
 simde_vcvtmq_s16_f16(simde_float16x8_t a) {
@@ -357,6 +368,7 @@ simde_vcvtmq_s64_f64(simde_float64x2_t a) {
   #undef vcvtmq_s64_f64
   #define vcvtmq_s64_f64(a) simde_vcvtmq_s64_f64(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_uint16x8_t
@@ -450,6 +462,8 @@ simde_vcvtmq_u64_f64(simde_float64x2_t a) {
   #define vcvtmq_u64_f64(a) simde_vcvtmq_u64_f64(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x4_t
 simde_vcvtm_s16_f16(simde_float16x4_t a) {
@@ -471,6 +485,7 @@ simde_vcvtm_s16_f16(simde_float16x4_t a) {
   #undef vcvtm_s16_f16
   #define vcvtm_s16_f16(a) simde_vcvtm_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_uint16x4_t
@@ -516,6 +531,8 @@ simde_vcvtm_u32_f32(simde_float32x2_t a) {
   #define vcvtm_u32_f32(a) simde_vcvtm_u32_f32(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int32x2_t
 simde_vcvtm_s32_f32(simde_float32x2_t a) {
@@ -559,6 +576,7 @@ simde_vcvtm_s64_f64(simde_float64x1_t a) {
   #undef vcvtm_s64_f64
   #define vcvtm_s64_f64(a) simde_vcvtm_s64_f64(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_uint64x1_t

diff --git a/simde/arm/neon/cvtn.h b/simde/arm/neon/cvtn.h
@@ -153,6 +153,8 @@ simde_vcvtnh_s32_f16(simde_float16_t a) {
   #define vcvtnh_s32_f16(a) simde_vcvtnh_s32_f16(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 int16_t
 simde_vcvtnh_s16_f16(simde_float16_t a) {
@@ -177,6 +179,7 @@ simde_vcvtnh_s16_f16(simde_float16_t a) {
   #undef vcvtnh_s16_f16
   #define vcvtnh_s16_f16(a) simde_vcvtnh_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 uint64_t
@@ -417,6 +420,8 @@ simde_vcvtnq_u64_f64(simde_float64x2_t a) {
   #define vcvtnq_u64_f64(a) simde_vcvtnq_u64_f64(a)
 #endif
 
+/* Disabled until we fix the FCVTZS/FCVTMS/FCVTPS/FCVTNS family intrinsics
+ * https://github.com/simd-everywhere/simde/issues/1099
 SIMDE_FUNCTION_ATTRIBUTES
 simde_int16x8_t
 simde_vcvtnq_s16_f16(simde_float16x8_t a) {
@@ -460,6 +465,7 @@ simde_vcvtn_s16_f16(simde_float16x4_t a) {
   #undef vcvtn_s16_f16
   #define vcvtn_s16_f16(a) simde_vcvtn_s16_f16(a)
 #endif
+*/
 
 SIMDE_FUNCTION_ATTRIBUTES
 simde_uint16x8_t