x86-64: Use ZMM16-ZMM31 in AVX512 memset family functions

Update ifunc-memset.h/ifunc-wmemset.h to select the function optimized
with AVX512 instructions using ZMM16-ZMM31 registers to avoid RTM abort
with usable AVX512VL and AVX512BW since VZEROUPPER isn't needed at
function exit.

(cherry picked from commit 4e2d8f352774b56078c34648b14a2412c38384f4)

4 files changed, 31 insertions, 24 deletions

diff --git a/sysdeps/x86_64/multiarch/ifunc-impl-list.c b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
index f78318e01a..a51f9b5911 100644
--- a/sysdeps/x86_64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/x86_64/multiarch/ifunc-impl-list.c
@@ -211,10 +211,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			       && CPU_FEATURE_USABLE (AVX512BW)),
 			      __memset_chk_evex_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, __memset_chk,
-			      CPU_FEATURE_USABLE (AVX512F),
+			      (CPU_FEATURE_USABLE (AVX512VL)
+			       && CPU_FEATURE_USABLE (AVX512BW)),
 			      __memset_chk_avx512_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, __memset_chk,
-			      CPU_FEATURE_USABLE (AVX512F),
+			      (CPU_FEATURE_USABLE (AVX512VL)
+			       && CPU_FEATURE_USABLE (AVX512BW)),
 			      __memset_chk_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, __memset_chk,
 			      CPU_FEATURE_USABLE (AVX512F),
@@ -252,10 +254,12 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			       && CPU_FEATURE_USABLE (AVX512BW)),
 			      __memset_evex_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, memset,
-			      CPU_FEATURE_USABLE (AVX512F),
+			      (CPU_FEATURE_USABLE (AVX512VL)
+			       && CPU_FEATURE_USABLE (AVX512BW)),
 			      __memset_avx512_unaligned_erms)
 	      IFUNC_IMPL_ADD (array, i, memset,
-			      CPU_FEATURE_USABLE (AVX512F),
+			      (CPU_FEATURE_USABLE (AVX512VL)
+			       && CPU_FEATURE_USABLE (AVX512BW)),
 			      __memset_avx512_unaligned)
 	      IFUNC_IMPL_ADD (array, i, memset,
 			      CPU_FEATURE_USABLE (AVX512F),
@@ -719,7 +723,7 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
 			      CPU_FEATURE_USABLE (AVX512VL),
 			      __wmemset_evex_unaligned)
 	      IFUNC_IMPL_ADD (array, i, wmemset,
-			      CPU_FEATURE_USABLE (AVX512F),
+			      CPU_FEATURE_USABLE (AVX512VL),
 			      __wmemset_avx512_unaligned))
 
 #ifdef SHARED
diff --git a/sysdeps/x86_64/multiarch/ifunc-memset.h b/sysdeps/x86_64/multiarch/ifunc-memset.h
index c5eb92121b..57029fc17b 100644
--- a/sysdeps/x86_64/multiarch/ifunc-memset.h
+++ b/sysdeps/x86_64/multiarch/ifunc-memset.h
@@ -53,13 +53,16 @@ IFUNC_SELECTOR (void)
   if (CPU_FEATURE_USABLE_P (cpu_features, AVX512F)
       && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512))
     {
-      if (CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER))
-	return OPTIMIZE (avx512_no_vzeroupper);
+      if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)
+	  && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW))
+	{
+	  if (CPU_FEATURE_USABLE_P (cpu_features, ERMS))
+	    return OPTIMIZE (avx512_unaligned_erms);
 
-      if (CPU_FEATURE_USABLE_P (cpu_features, ERMS))
-	return OPTIMIZE (avx512_unaligned_erms);
+	  return OPTIMIZE (avx512_unaligned);
+	}
 
-      return OPTIMIZE (avx512_unaligned);
+      return OPTIMIZE (avx512_no_vzeroupper);
     }
 
   if (CPU_FEATURE_USABLE_P (cpu_features, AVX2))
diff --git a/sysdeps/x86_64/multiarch/ifunc-wmemset.h b/sysdeps/x86_64/multiarch/ifunc-wmemset.h
index e81b96c69f..e06e8b4d80 100644
--- a/sysdeps/x86_64/multiarch/ifunc-wmemset.h
+++ b/sysdeps/x86_64/multiarch/ifunc-wmemset.h
@@ -33,13 +33,13 @@ IFUNC_SELECTOR (void)
   if (CPU_FEATURE_USABLE_P (cpu_features, AVX2)
       && CPU_FEATURES_ARCH_P (cpu_features, AVX_Fast_Unaligned_Load))
     {
-      if (CPU_FEATURE_USABLE_P (cpu_features, AVX512F)
-	  && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512)
-	  && !CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_VZEROUPPER))
-	return OPTIMIZE (avx512_unaligned);
-
       if (CPU_FEATURE_USABLE_P (cpu_features, AVX512VL))
-	return OPTIMIZE (evex_unaligned);
+	{
+	  if (!CPU_FEATURES_ARCH_P (cpu_features, Prefer_No_AVX512))
+	    return OPTIMIZE (avx512_unaligned);
+
+	  return OPTIMIZE (evex_unaligned);
+	}
 
       if (CPU_FEATURE_USABLE_P (cpu_features, RTM))
 	return OPTIMIZE (avx2_unaligned_rtm);
diff --git a/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S b/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
index 0783979ca5..22e7b187c8 100644
--- a/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
+++ b/sysdeps/x86_64/multiarch/memset-avx512-unaligned-erms.S
@@ -1,22 +1,22 @@
 #if IS_IN (libc)
 # define VEC_SIZE	64
-# define VEC(i)		zmm##i
+# define XMM0		xmm16
+# define YMM0		ymm16
+# define VEC0		zmm16
+# define VEC(i)		VEC##i
 # define VMOVU		vmovdqu64
 # define VMOVA		vmovdqa64
+# define VZEROUPPER
 
 # define MEMSET_VDUP_TO_VEC0_AND_SET_RETURN(d, r) \
-  vmovd d, %xmm0; \
   movq r, %rax; \
-  vpbroadcastb %xmm0, %xmm0; \
-  vpbroadcastq %xmm0, %zmm0
+  vpbroadcastb d, %VEC0
 
 # define WMEMSET_VDUP_TO_VEC0_AND_SET_RETURN(d, r) \
-  vmovd d, %xmm0; \
   movq r, %rax; \
-  vpbroadcastd %xmm0, %xmm0; \
-  vpbroadcastq %xmm0, %zmm0
+  vpbroadcastd d, %VEC0
 
-# define SECTION(p)		p##.avx512
+# define SECTION(p)		p##.evex512
 # define MEMSET_SYMBOL(p,s)	p##_avx512_##s
 # define WMEMSET_SYMBOL(p,s)	p##_avx512_##s