[glibc] x86: Optimize xstate size calculation

[H.J. Lu]

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=70b648855185e967e54668b101d24704c3fb869d

commit 70b648855185e967e54668b101d24704c3fb869d
Author: Sunil K Pandey <skpgkp2@gmail.com>
Date:   Thu Apr 3 13:00:45 2025 -0700

    x86: Optimize xstate size calculation
    
    Scan xstate IDs up to the maximum supported xstate ID.  Remove the
    separate AMX xstate calculation.  Instead, exclude the AMX space from
    the start of TILECFG to the end of TILEDATA in xsave_state_size.
    
    Completed validation on SKL/SKX/SPR/SDE and compared xsave state size
    with "ld.so --list-diagnostics" option, no regression.
    
    Co-Authored-By: H.J. Lu <hjl.tools@gmail.com>
    Reviewed-by: Sunil K Pandey <skpgkp2@gmail.com>

Diff:
---
 sysdeps/x86/cpu-features.c | 74 +++++++++++-----------------------------------
 sysdeps/x86/sysdep.h       |  6 ++++
 2 files changed, 24 insertions(+), 56 deletions(-)

diff --git a/sysdeps/x86/cpu-features.c b/sysdeps/x86/cpu-features.c
index 6cf7e4caf1..9d136e42ad 100644
--- a/sysdeps/x86/cpu-features.c
+++ b/sysdeps/x86/cpu-features.c
@@ -325,13 +325,8 @@ update_active (struct cpu_features *cpu_features)
 	      /* Check if XSAVEC is available.  */
 	      if (CPU_FEATURES_CPU_P (cpu_features, XSAVEC))
 		{
-		  unsigned int xstate_comp_offsets[32];
-		  unsigned int xstate_comp_sizes[32];
-#ifdef __x86_64__
-		  unsigned int xstate_amx_comp_offsets[32];
-		  unsigned int xstate_amx_comp_sizes[32];
-		  unsigned int amx_ecx;
-#endif
+		  unsigned int xstate_comp_offsets[X86_XSTATE_MAX_ID + 1];
+		  unsigned int xstate_comp_sizes[X86_XSTATE_MAX_ID + 1];
 		  unsigned int i;
 
 		  xstate_comp_offsets[0] = 0;
@@ -339,39 +334,16 @@ update_active (struct cpu_features *cpu_features)
 		  xstate_comp_offsets[2] = 576;
 		  xstate_comp_sizes[0] = 160;
 		  xstate_comp_sizes[1] = 256;
-#ifdef __x86_64__
-		  xstate_amx_comp_offsets[0] = 0;
-		  xstate_amx_comp_offsets[1] = 160;
-		  xstate_amx_comp_offsets[2] = 576;
-		  xstate_amx_comp_sizes[0] = 160;
-		  xstate_amx_comp_sizes[1] = 256;
-#endif
 
-		  for (i = 2; i < 32; i++)
+		  for (i = 2; i <= X86_XSTATE_MAX_ID; i++)
 		    {
 		      if ((FULL_STATE_SAVE_MASK & (1 << i)) != 0)
 			{
 			  __cpuid_count (0xd, i, eax, ebx, ecx, edx);
-#ifdef __x86_64__
-			  /* Include this in xsave_state_full_size.  */
-			  amx_ecx = ecx;
-			  xstate_amx_comp_sizes[i] = eax;
-			  if ((AMX_STATE_SAVE_MASK & (1 << i)) != 0)
-			    {
-			      /* Exclude this from xsave_state_size.  */
-			      ecx = 0;
-			      xstate_comp_sizes[i] = 0;
-			    }
-			  else
-#endif
-			    xstate_comp_sizes[i] = eax;
+			  xstate_comp_sizes[i] = eax;
 			}
 		      else
 			{
-#ifdef __x86_64__
-			  amx_ecx = 0;
-			  xstate_amx_comp_sizes[i] = 0;
-#endif
 			  ecx = 0;
 			  xstate_comp_sizes[i] = 0;
 			}
@@ -380,42 +352,32 @@ update_active (struct cpu_features *cpu_features)
 			{
 			  xstate_comp_offsets[i]
 			    = (xstate_comp_offsets[i - 1]
-			       + xstate_comp_sizes[i -1]);
+			       + xstate_comp_sizes[i - 1]);
 			  if ((ecx & (1 << 1)) != 0)
 			    xstate_comp_offsets[i]
 			      = ALIGN_UP (xstate_comp_offsets[i], 64);
-#ifdef __x86_64__
-			  xstate_amx_comp_offsets[i]
-			    = (xstate_amx_comp_offsets[i - 1]
-			       + xstate_amx_comp_sizes[i - 1]);
-			  if ((amx_ecx & (1 << 1)) != 0)
-			    xstate_amx_comp_offsets[i]
-			      = ALIGN_UP (xstate_amx_comp_offsets[i],
-					  64);
-#endif
 			}
 		    }
 
 		  /* Use XSAVEC.  */
 		  unsigned int size
-		    = xstate_comp_offsets[31] + xstate_comp_sizes[31];
+		    = (xstate_comp_offsets[X86_XSTATE_MAX_ID]
+		       + xstate_comp_sizes[X86_XSTATE_MAX_ID]);
 		  if (size)
 		    {
+		      size = ALIGN_UP (size + TLSDESC_CALL_REGISTER_SAVE_AREA,
+				       64);
 #ifdef __x86_64__
-		      unsigned int amx_size
-			= (xstate_amx_comp_offsets[31]
-			   + xstate_amx_comp_sizes[31]);
-		      amx_size
-			= ALIGN_UP ((amx_size
-				     + TLSDESC_CALL_REGISTER_SAVE_AREA),
-				    64);
-		      /* Set TLSDESC state size to the compact AMX
-			 state size for XSAVEC.  */
-		      _dl_x86_features_tlsdesc_state_size = amx_size;
+		      _dl_x86_features_tlsdesc_state_size = size;
+		      /* Exclude the AMX space from the start of TILECFG
+			 space to the end of TILEDATA space.  If CPU
+			 doesn't support AMX, TILECFG offset is the same
+			 as TILEDATA + 1 offset.  Otherwise, they are
+			 multiples of 64.  */
+		      size -= (xstate_comp_offsets[X86_XSTATE_TILEDATA_ID + 1]
+			       - xstate_comp_offsets[X86_XSTATE_TILECFG_ID]);
 #endif
-		      cpu_features->xsave_state_size
-			= ALIGN_UP (size + TLSDESC_CALL_REGISTER_SAVE_AREA,
-				    64);
+		      cpu_features->xsave_state_size = size;
 		      CPU_FEATURE_SET (cpu_features, XSAVEC);
 		    }
 		}
diff --git a/sysdeps/x86/sysdep.h b/sysdeps/x86/sysdep.h
index 541393f1dc..c3c73e75dd 100644
--- a/sysdeps/x86/sysdep.h
+++ b/sysdeps/x86/sysdep.h
@@ -102,6 +102,9 @@
    | (1 << X86_XSTATE_ZMM_ID)		\
    | (1 << X86_XSTATE_APX_F_ID))
 
+/* The maximum supported xstate ID.  */
+# define X86_XSTATE_MAX_ID	X86_XSTATE_APX_F_ID
+
 /* AMX state mask.  */
 # define AMX_STATE_SAVE_MASK		\
   ((1 << X86_XSTATE_TILECFG_ID) | (1 << X86_XSTATE_TILEDATA_ID))
@@ -123,6 +126,9 @@
    | (1 << X86_XSTATE_K_ID)		\
    | (1 << X86_XSTATE_ZMM_H_ID))
 
+/* The maximum supported xstate ID.  */
+# define X86_XSTATE_MAX_ID	X86_XSTATE_ZMM_H_ID
+
 /* States to be included in xsave_state_size.  */
 # define FULL_STATE_SAVE_MASK		STATE_SAVE_MASK
 #endif