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>

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_offsets[X86_XSTATE_MAX_ID + 1];
-		  unsigned int xstate_comp_sizes[32];
+		  unsigned int xstate_comp_sizes[X86_XSTATE_MAX_ID + 1];
-#ifdef __x86_64__
-		  unsigned int xstate_amx_comp_offsets[32];
-		  unsigned int xstate_amx_comp_sizes[32];
-		  unsigned int amx_ecx;
-#endif
 		  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;
 			}
 		      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
+		      _dl_x86_features_tlsdesc_state_size = size;
-			= (xstate_amx_comp_offsets[31]
+		      /* Exclude the AMX space from the start of TILECFG
-			   + xstate_amx_comp_sizes[31]);
+			 space to the end of TILEDATA space.  If CPU
-		      amx_size
+			 doesn't support AMX, TILECFG offset is the same
-			= ALIGN_UP ((amx_size
+			 as TILEDATA + 1 offset.  Otherwise, they are
-				     + TLSDESC_CALL_REGISTER_SAVE_AREA),
+			 multiples of 64.  */
-				    64);
+		      size -= (xstate_comp_offsets[X86_XSTATE_TILEDATA_ID + 1]
-		      /* Set TLSDESC state size to the compact AMX
+			       - xstate_comp_offsets[X86_XSTATE_TILECFG_ID]);
-			 state size for XSAVEC.  */
-		      _dl_x86_features_tlsdesc_state_size = amx_size;
 #endif
-		      cpu_features->xsave_state_size
+		      cpu_features->xsave_state_size = size;
-			= ALIGN_UP (size + TLSDESC_CALL_REGISTER_SAVE_AREA,
-				    64);
 		      CPU_FEATURE_SET (cpu_features, XSAVEC);
 		    }
 		}

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