mmx: work around some clang <= 11 bugs on POWER9

These were causing ICEs prior to clang-12 on POWER9 when compiled
without optimizitaion.  Here is a trivial reproducer:

  typedef short i16x4 __attribute__((__vector_size__(8)));

  i16x4 shr(i16x4 v, long n) {
    return v >> n;
  }

AFAICT the problem has been fixed in clang-12, though I haven't
bothered to track down a bug report, git commit, phabricator review,
etc. so no SIMDE_BUG_* macro.

simde-common.h

@@ -820,6 +820,9 @@ HEDLEY_DIAGNOSTIC_POP
 #    if defined(SIMDE_ARCH_POWER)
 #      define SIMDE_BUG_CLANG_46770
 #    endif
+#    if defined(_ARCH_PWR9) && !SIMDE_DETECT_CLANG_VERSION_CHECK(12,0,0) && !defined(__OPTIMIZE__)
+#      define SIMDE_BUG_CLANG_POWER9_16x4_BAD_SHIFT
+#    endif
 #    if defined(SIMDE_ARCH_X86) || defined(SIMDE_ARCH_AMD64)
 #      if HEDLEY_HAS_WARNING("-Wsign-conversion") && SIMDE_DETECT_CLANG_VERSION_NOT(11,0,0)
 #        define SIMDE_BUG_CLANG_45931

x86/mmx.h

@@ -1390,6 +1390,11 @@ simde_mm_sll_pi16 (simde__m64 a, simde__m64 count) {
       #endif
       r_.neon_i16 = vshl_s16(a_.neon_i16, vmov_n_s16(HEDLEY_STATIC_CAST(int16_t, vget_lane_u64(count_.neon_u64, 0))));
       HEDLEY_DIAGNOSTIC_POP
+    #elif defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR) && defined(SIMDE_BUG_CLANG_POWER9_16x4_BAD_SHIFT)
+      if (HEDLEY_UNLIKELY(count_.u64[0] > 15))
+        return simde_mm_setzero_si64();
+
+      r_.i16 = a_.i16 << HEDLEY_STATIC_CAST(int16_t, count_.u64[0]);
     #elif defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR)
       r_.i16 = a_.i16 << count_.u64[0];
     #else
@@ -1462,8 +1467,14 @@ simde_mm_slli_pi16 (simde__m64 a, int count) {
     simde__m64_private r_;
     simde__m64_private a_ = simde__m64_to_private(a);
 
-    #if defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR)
+    #if defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR) && defined(SIMDE_BUG_CLANG_POWER9_16x4_BAD_SHIFT)
+      if (HEDLEY_UNLIKELY(count > 15))
+        return simde_mm_setzero_si64();
+
+      r_.i16 = a_.i16 << HEDLEY_STATIC_CAST(int16_t, count);
+    #elif defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR)
       r_.i16 = a_.i16 << count;
+    #elif defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR)
     #elif defined(SIMDE_ARM_NEON_A32V7_NATIVE)
       r_.neon_i16 = vshl_s16(a_.neon_i16, vmov_n_s16((int16_t) count));
     #elif defined(SIMDE_MIPS_LOONGSON_MMI_NATIVE)
@@ -1583,7 +1594,12 @@ simde_mm_srl_pi16 (simde__m64 a, simde__m64 count) {
     simde__m64_private a_ = simde__m64_to_private(a);
     simde__m64_private count_ = simde__m64_to_private(count);
 
-    #if defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR)
+    #if defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR) && defined(SIMDE_BUG_CLANG_POWER9_16x4_BAD_SHIFT)
+      if (HEDLEY_UNLIKELY(count_.u64[0] > 15))
+        return simde_mm_setzero_si64();
+
+      r_.i16 = a_.i16 >> HEDLEY_STATIC_CAST(int16_t, count_.u64[0]);
+    #elif defined(SIMDE_VECTOR_SUBSCRIPT_SCALAR)
       r_.u16 = a_.u16 >> count_.u64[0];
     #elif defined(SIMDE_ARM_NEON_A32V7_NATIVE)
       r_.neon_u16 = vshl_u16(a_.neon_u16, vmov_n_s16(-((int16_t) vget_lane_u64(count_.neon_u64, 0))));