simde_float16: prefer __fp16 if available

Fixes: #1104
simd-everywhere · Nov 9, 2023 · aba26f6 · aba26f6
1 parent c877fe5
commit aba26f6
Show file tree

Hide file tree

Showing 5 changed files with 21 additions and 3 deletions.
diff --git a/simde/arm/neon/cadd_rot270.h b/simde/arm/neon/cadd_rot270.h
@@ -33,6 +33,11 @@ HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
 SIMDE_BEGIN_DECLS_
 
+#if defined(__clang__) && SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16
+SIMDE_DIAGNOSTIC_DISABLE_DOUBLE_PROMOTION_
+_Pragma("clang diagnostic ignored \"-Wimplicit-float-conversion\"")
+#endif
+
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16x4_t simde_vcadd_rot270_f16(simde_float16x4_t a, simde_float16x4_t b)
 {

diff --git a/simde/arm/neon/cadd_rot90.h b/simde/arm/neon/cadd_rot90.h
@@ -33,6 +33,11 @@ HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
 SIMDE_BEGIN_DECLS_
 
+#if defined(__clang__) && SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16
+SIMDE_DIAGNOSTIC_DISABLE_DOUBLE_PROMOTION_
+_Pragma("clang diagnostic ignored \"-Wimplicit-float-conversion\"")
+#endif
+
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16x4_t simde_vcadd_rot90_f16(simde_float16x4_t a, simde_float16x4_t b)
 {

diff --git a/simde/simde-f16.h b/simde/simde-f16.h
@@ -65,7 +65,9 @@ SIMDE_BEGIN_DECLS_
  * any ideas on how to improve it.  If you do, patches are definitely
  * welcome. */
 #if !defined(SIMDE_FLOAT16_API)
-  #if !defined(__EMSCRIPTEN__) && !(defined(__clang__) && defined(SIMDE_ARCH_POWER)) && \
+  #if defined(__ARM_FP16_FORMAT_IEEE) && (defined(SIMDE_ARM_NEON_FP16) || defined(__ARM_FP16_ARGS))
+    #define SIMDE_FLOAT16_API SIMDE_FLOAT16_API_FP16
+  #elif !defined(__EMSCRIPTEN__) && !(defined(__clang__) && defined(SIMDE_ARCH_POWER)) && \
     !(defined(HEDLEY_MSVC_VERSION) && defined(__clang__)) && \
     !(defined(__clang__) && defined(SIMDE_ARCH_RISCV64)) && ( \
       defined(SIMDE_X86_AVX512FP16_NATIVE) || \
@@ -79,8 +81,6 @@ SIMDE_BEGIN_DECLS_
     * clang will define the constants even if _Float16 is not
     * supported.  Ideas welcome. */
     #define SIMDE_FLOAT16_API SIMDE_FLOAT16_API_FLOAT16
-  #elif defined(__ARM_FP16_FORMAT_IEEE) && (defined(SIMDE_ARM_NEON_FP16) || defined(__ARM_FP16_ARGS))
-    #define SIMDE_FLOAT16_API SIMDE_FLOAT16_API_FP16
   #elif defined(__FLT16_MIN__) && \
       (defined(__clang__) && \
       (!defined(SIMDE_ARCH_AARCH64) || SIMDE_DETECT_CLANG_VERSION_CHECK(7,0,0)) \

diff --git a/simde/x86/avx512/cmp.h b/simde/x86/avx512/cmp.h
@@ -38,6 +38,10 @@ HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
 SIMDE_BEGIN_DECLS_
 
+#if defined(__clang__) && SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16
+SIMDE_DIAGNOSTIC_DISABLE_DOUBLE_PROMOTION_
+#endif
+
 SIMDE_HUGE_FUNCTION_ATTRIBUTES
 simde__mmask64
 simde_mm512_cmp_epi8_mask (simde__m512i a, simde__m512i b, const int imm8)

diff --git a/simde/x86/avx512/reduce.h b/simde/x86/avx512/reduce.h
@@ -33,6 +33,10 @@ HEDLEY_DIAGNOSTIC_PUSH
 SIMDE_DISABLE_UNWANTED_DIAGNOSTICS
 SIMDE_BEGIN_DECLS_
 
+#if defined(__clang__) && SIMDE_FLOAT16_API == SIMDE_FLOAT16_API_FP16
+SIMDE_DIAGNOSTIC_DISABLE_DOUBLE_PROMOTION_
+#endif
+
 SIMDE_FUNCTION_ATTRIBUTES
 simde_float16
 simde_mm512_reduce_max_ph(simde__m512h a) {