diff mbox series

[committed] amdgcn: Implement -msram-ecc=any

Message ID a00f136e-2f3b-4cda-52f6-00f17c1477c1@codesourcery.com
State New
Headers show
Series [committed] amdgcn: Implement -msram-ecc=any | expand

Commit Message

Andrew Stubbs Oct. 7, 2021, 10:45 a.m. UTC
I've committed this patch to implement the -msram-ecc=any feature that 
has been stubbed out awaiting LLVM support for a while now.

When the LLVM assembler supports the "any" feature (v13+) GCC will now 
make use of it. Otherwise, GCC will continue to treat "any" the same as 
"on".

Using the "any" setting for the "sram-ecc" attribute means that we will 
not need a multilib to support differently configure hardware. The 
compiler has always produced code that runs in any configuration, but 
the ELF flags could not reflect this, until now.

Andrew
amdgcn: Implement -msram-ecc=any

The option was already there, but just an alias for -msram-ecc=on.  Now that
LLVM13 supports HSACOv4 and the new ELF flags I can implement the option
properly.

The "any" option is the default in order to ensure that library files work
whichever way the user wants, which means we won't need multilibs to support
the different SRAM ECC hardware configurations.

gcc/ChangeLog:

	* config/gcn/gcn-hsa.h (SRAMOPT): Include the whole option string.
	Adjust for new -msram-ecc=any behaviour.
	(ASM_SPEC): Adjust -mxnack and -msram-ecc usage.
	* config/gcn/gcn.c (output_file_start): Implement -msram-ecc=any.
	* config/gcn/mkoffload.c (EF_AMDGPU_XNACK): Rename to ...
	(EF_AMDGPU_XNACK_V3): ... this.
	(EF_AMDGPU_SRAM_ECC): Rename to ...
	(EF_AMDGPU_SRAM_ECC_V3): ... this.
	(EF_AMDGPU_FEATURE_XNACK_V4): New.
	(EF_AMDGPU_FEATURE_XNACK_UNSUPPORTED_V4): New.
	(EF_AMDGPU_FEATURE_XNACK_ANY_V4): New.
	(EF_AMDGPU_FEATURE_XNACK_OFF_V4): New.
	(EF_AMDGPU_FEATURE_XNACK_ON_V4): New.
	(EF_AMDGPU_FEATURE_SRAMECC_V4): New.
	(EF_AMDGPU_FEATURE_SRAMECC_UNSUPPORTED_V4): New.
	(EF_AMDGPU_FEATURE_SRAMECC_ANY_V4): New.
	(EF_AMDGPU_FEATURE_SRAMECC_OFF_V4): New.
	(EF_AMDGPU_FEATURE_SRAMECC_ON_V4): New.
	(SET_XNACK_ON): New.
	(SET_XNACK_OFF): New.
	(TEST_XNACK): New.
	(SET_SRAM_ECC_ON): New.
	(SET_SRAM_ECC_ANY): New.
	(SET_SRAM_ECC_OFF): New.
	(TEST_SRAM_ECC_ANY): New.
	(TEST_SRAM_ECC_ON): New.
	(main): Implement HSACOv4 and -msram-ecc=any.
diff mbox series

Patch

diff --git a/gcc/config/gcn/gcn-hsa.h b/gcc/config/gcn/gcn-hsa.h
index 60fd40a10b71..6a432d17d99f 100644
--- a/gcc/config/gcn/gcn-hsa.h
+++ b/gcc/config/gcn/gcn-hsa.h
@@ -118,10 +118,12 @@  extern unsigned int gcn_local_sym_hash (const char *name);
 #endif
 
 #ifdef HAVE_GCN_ASM_V3_SYNTAX
-#define SRAMOPT "sram-ecc"
+#define SRAMOPT "!msram-ecc=off:-mattr=+sram-ecc;:-mattr=-sram-ecc"
 #endif
 #ifdef HAVE_GCN_ASM_V4_SYNTAX
-#define SRAMOPT "sramecc"
+/* In HSACOv4 no attribute setting means the binary supports "any" hardware
+   configuration.  The name of the attribute also changed.  */
+#define SRAMOPT "msram-ecc=on:-mattr=+sramecc;msram-ecc=off:-mattr=-sramecc"
 #endif
 #if !defined(SRAMOPT) && !defined(IN_LIBGCC2)
 #error "No assembler syntax configured"
@@ -143,11 +145,9 @@  extern unsigned int gcn_local_sym_hash (const char *name);
 #define ASM_SPEC  "-triple=amdgcn--amdhsa "  \
 		  "%:last_arg(%{march=*:-mcpu=%*}) " \
 		  HSACO3_SELECT_OPT \
-		  "-mattr=%{" X_FIJI X_900 X_906 X_908 \
-			    "mxnack:+xnack;:-xnack} " \
-		  /* FIXME: support "any" when we move to HSACOv4.  */ \
-		  "-mattr=%{" S_FIJI S_900 S_906 S_908 \
-			    "!msram-ecc=off:+" SRAMOPT ";:-" SRAMOPT "} " \
+		  "%{" X_FIJI X_900 X_906 X_908 \
+		     "mxnack:-mattr=+xnack;:-mattr=-xnack} " \
+		  "%{" S_FIJI S_900 S_906 S_908 SRAMOPT "} " \
 		  "-filetype=obj"
 #define LINK_SPEC "--pie --export-dynamic"
 #define LIB_SPEC  "-lc"
diff --git a/gcc/config/gcn/gcn.c b/gcc/config/gcn/gcn.c
index 8517168ff0ad..2e90f327c451 100644
--- a/gcc/config/gcn/gcn.c
+++ b/gcc/config/gcn/gcn.c
@@ -5265,9 +5265,13 @@  output_file_start (void)
   const char *sram_ecc = (flag_sram_ecc ? "+sram-ecc" : "");
 #endif
 #if HAVE_GCN_ASM_V4_SYNTAX
+  /* In HSACOv4 no attribute setting means the binary supports "any" hardware
+     configuration.  In GCC binaries, this is true for SRAM ECC, but not
+     XNACK.  */
   const char *xnack = (flag_xnack ? ":xnack+" : ":xnack-");
-  /* FIXME: support "any" when we move to HSACOv4.  */
-  const char *sram_ecc = (flag_sram_ecc ? ":sramecc+" : ":sramecc-");
+  const char *sram_ecc = (flag_sram_ecc == SRAM_ECC_ON ? ":sramecc+"
+			  : flag_sram_ecc == SRAM_ECC_OFF ? ":sramecc-"
+			  : "");
 #endif
   if (!use_xnack_attr)
     xnack = "";
diff --git a/gcc/config/gcn/mkoffload.c b/gcc/config/gcn/mkoffload.c
index 732bdfd98e57..a3b22d059b96 100644
--- a/gcc/config/gcn/mkoffload.c
+++ b/gcc/config/gcn/mkoffload.c
@@ -54,8 +54,51 @@ 
 #undef  EF_AMDGPU_MACH_AMDGCN_GFX908
 #define EF_AMDGPU_MACH_AMDGCN_GFX908 0x30
 
-#define EF_AMDGPU_XNACK    0x100
-#define EF_AMDGPU_SRAM_ECC 0x200
+#define EF_AMDGPU_XNACK_V3    0x100
+#define EF_AMDGPU_SRAM_ECC_V3 0x200
+
+#define EF_AMDGPU_FEATURE_XNACK_V4	0x300  /* Mask.  */
+#define EF_AMDGPU_FEATURE_XNACK_UNSUPPORTED_V4	0x000
+#define EF_AMDGPU_FEATURE_XNACK_ANY_V4	0x100
+#define EF_AMDGPU_FEATURE_XNACK_OFF_V4	0x200
+#define EF_AMDGPU_FEATURE_XNACK_ON_V4	0x300
+
+#define EF_AMDGPU_FEATURE_SRAMECC_V4	0xc00  /* Mask.  */
+#define EF_AMDGPU_FEATURE_SRAMECC_UNSUPPORTED_V4	0x000
+#define EF_AMDGPU_FEATURE_SRAMECC_ANY_V4	0x400
+#define EF_AMDGPU_FEATURE_SRAMECC_OFF_V4	0x800
+#define EF_AMDGPU_FEATURE_SRAMECC_ON_V4		0xc00
+
+#ifdef HAVE_GCN_ASM_V3_SYNTAX
+#define SET_XNACK_ON(VAR) VAR |= EF_AMDGPU_XNACK_V3
+#define SET_XNACK_OFF(VAR) VAR &= ~EF_AMDGPU_XNACK_V3
+#define TEST_XNACK(VAR) (VAR & EF_AMDGPU_XNACK_V3)
+
+#define SET_SRAM_ECC_ON(VAR) VAR |= EF_AMDGPU_SRAM_ECC_V3
+#define SET_SRAM_ECC_ANY(VAR) SET_SRAM_ECC_ON (VAR)
+#define SET_SRAM_ECC_OFF(VAR) VAR &= ~EF_AMDGPU_SRAM_ECC_V3
+#define TEST_SRAM_ECC_ANY(VAR) 0 /* Not supported.  */
+#define TEST_SRAM_ECC_ON(VAR) (VAR & EF_AMDGPU_SRAM_ECC_V3)
+#endif
+#ifdef HAVE_GCN_ASM_V4_SYNTAX
+#define SET_XNACK_ON(VAR) VAR = ((VAR & ~EF_AMDGPU_FEATURE_XNACK_V4) \
+				 | EF_AMDGPU_FEATURE_XNACK_ON_V4)
+#define SET_XNACK_OFF(VAR) VAR = ((VAR & ~EF_AMDGPU_FEATURE_XNACK_V4) \
+				  | EF_AMDGPU_FEATURE_XNACK_OFF_V4)
+#define TEST_XNACK(VAR) ((VAR & EF_AMDGPU_FEATURE_XNACK_V4) \
+			 == EF_AMDGPU_FEATURE_XNACK_ON_V4)
+
+#define SET_SRAM_ECC_ON(VAR) VAR = ((VAR & ~EF_AMDGPU_FEATURE_SRAMECC_V4) \
+				    | EF_AMDGPU_FEATURE_SRAMECC_ON_V4)
+#define SET_SRAM_ECC_ANY(VAR) VAR = ((VAR & ~EF_AMDGPU_FEATURE_SRAMECC_V4) \
+				     | EF_AMDGPU_FEATURE_SRAMECC_ANY_V4)
+#define SET_SRAM_ECC_OFF(VAR) VAR = ((VAR & ~EF_AMDGPU_FEATURE_SRAMECC_V4) \
+				     | EF_AMDGPU_FEATURE_SRAMECC_OFF_V4)
+#define TEST_SRAM_ECC_ANY(VAR) ((VAR & EF_AMDGPU_FEATURE_SRAMECC_V4) \
+				== EF_AMDGPU_FEATURE_SRAMECC_ANY_V4)
+#define TEST_SRAM_ECC_ON(VAR) ((VAR & EF_AMDGPU_FEATURE_SRAMECC_V4) \
+			       == EF_AMDGPU_FEATURE_SRAMECC_ON_V4)
+#endif
 
 #ifndef R_AMDGPU_NONE
 #define R_AMDGPU_NONE		0
@@ -80,7 +123,13 @@  static struct obstack files_to_cleanup;
 
 enum offload_abi offload_abi = OFFLOAD_ABI_UNSET;
 uint32_t elf_arch = EF_AMDGPU_MACH_AMDGCN_GFX803;  // Default GPU architecture.
-uint32_t elf_flags = 0;
+uint32_t elf_flags =
+#ifdef HAVE_GCN_ASM_V3_SYNTAX
+    0;
+#endif
+#ifdef HAVE_GCN_ASM_V4_SYNTAX
+    (EF_AMDGPU_FEATURE_XNACK_ANY_V4 | EF_AMDGPU_FEATURE_SRAMECC_ANY_V4);
+#endif
 
 /* Delete tempfiles.  */
 
@@ -851,23 +900,22 @@  main (int argc, char **argv)
       else if (strcmp (argv[i], "-fpic") == 0)
 	fpic = true;
       else if (strcmp (argv[i], "-mxnack") == 0)
-	elf_flags |= EF_AMDGPU_XNACK;
+	SET_XNACK_ON (elf_flags);
       else if (strcmp (argv[i], "-mno-xnack") == 0)
-	elf_flags &= ~EF_AMDGPU_XNACK;
+	SET_XNACK_OFF (elf_flags);
       else if (strcmp (argv[i], "-msram-ecc=on") == 0)
 	{
-	  elf_flags |= EF_AMDGPU_SRAM_ECC;
+	  SET_SRAM_ECC_ON (elf_flags);
 	  sram_seen = true;
 	}
       else if (strcmp (argv[i], "-msram-ecc=any") == 0)
 	{
-	  /* FIXME: change this when we move to HSACOv4.  */
-	  elf_flags |= EF_AMDGPU_SRAM_ECC;
+	  SET_SRAM_ECC_ANY (elf_flags);
 	  sram_seen = true;
 	}
       else if (strcmp (argv[i], "-msram-ecc=off") == 0)
 	{
-	  elf_flags &= ~EF_AMDGPU_SRAM_ECC;
+	  SET_SRAM_ECC_OFF (elf_flags);
 	  sram_seen = true;
 	}
       else if (strcmp (argv[i], "-save-temps") == 0)
@@ -890,23 +938,27 @@  main (int argc, char **argv)
   if (!(fopenacc ^ fopenmp))
     fatal_error (input_location, "either -fopenacc or -fopenmp must be set");
 
-  /* The SRAM-ECC feature defaults to "any" on GPUs where the feature is
-     available.  */
   if (!sram_seen)
-    switch (elf_arch)
-      {
-      case EF_AMDGPU_MACH_AMDGCN_GFX803:
-      case EF_AMDGPU_MACH_AMDGCN_GFX900:
-      case EF_AMDGPU_MACH_AMDGCN_GFX906:
+    {
+#ifdef HAVE_GCN_ASM_V3_SYNTAX
+      /* For HSACOv3, the SRAM-ECC feature defaults to "on" on GPUs where the
+	 feature is available.
+	 (HSACOv4 has elf_flags initialsed to "any" in all cases.)  */
+      switch (elf_arch)
+	{
+	case EF_AMDGPU_MACH_AMDGCN_GFX803:
+	case EF_AMDGPU_MACH_AMDGCN_GFX900:
+	case EF_AMDGPU_MACH_AMDGCN_GFX906:
 #ifndef HAVE_GCN_SRAM_ECC_GFX908
-      case EF_AMDGPU_MACH_AMDGCN_GFX908:
+	case EF_AMDGPU_MACH_AMDGCN_GFX908:
 #endif
-	break;
-      default:
-	/* FIXME: change this when we move to HSACOv4.  */
-	elf_flags |= EF_AMDGPU_SRAM_ECC;
-	break;
-      }
+	  break;
+	default:
+	  SET_SRAM_ECC_ON (elf_flags);
+	  break;
+	}
+#endif
+    }
 
   const char *abi;
   switch (offload_abi)
@@ -936,11 +988,12 @@  main (int argc, char **argv)
   if (fopenmp)
     obstack_ptr_grow (&cc_argv_obstack, "-mgomp");
   obstack_ptr_grow (&cc_argv_obstack,
-		    (elf_flags & EF_AMDGPU_XNACK
+		    (TEST_XNACK (elf_flags)
 		     ? "-mxnack" : "-mno-xnack"));
   obstack_ptr_grow (&cc_argv_obstack,
-		    (elf_flags & EF_AMDGPU_SRAM_ECC
-		     ? "-msram-ecc=on" : "-msram-ecc=off"));
+		    (TEST_SRAM_ECC_ON (elf_flags) ? "-msram-ecc=on"
+		     : TEST_SRAM_ECC_ANY (elf_flags) ? "-msram-ecc=any"
+		     : "-msram-ecc=off"));
 
   for (int ix = 1; ix != argc; ix++)
     {
@@ -1043,11 +1096,12 @@  main (int argc, char **argv)
       obstack_ptr_grow (&ld_argv_obstack, gcn_s2_name);
       obstack_ptr_grow (&ld_argv_obstack, "-lgomp");
       obstack_ptr_grow (&ld_argv_obstack,
-			(elf_flags & EF_AMDGPU_XNACK
+			(TEST_XNACK (elf_flags)
 			 ? "-mxnack" : "-mno-xnack"));
       obstack_ptr_grow (&ld_argv_obstack,
-			(elf_flags & EF_AMDGPU_SRAM_ECC
-			 ? "-msram-ecc=on" : "-msram-ecc=off"));
+			(TEST_SRAM_ECC_ON (elf_flags) ? "-msram-ecc=on"
+			 : TEST_SRAM_ECC_ANY (elf_flags) ? "-msram-ecc=any"
+			 : "-msram-ecc=off"));
       if (verbose)
 	obstack_ptr_grow (&ld_argv_obstack, "-v");