This is the mail archive of the binutils@sourceware.org mailing list for the binutils project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[PATCH] MIPS/GAS/testsuite: Fine-grained MIPS FP ISA matching


Hi,

 While merging test suite changes I've noticed some recent 24K erratum 
workaround cases have suboptimal coverage.  Upon further investigation I 
discovered we have no proper mechanism to depend on for FP ISA revisions 
in multi-architecture tests.  Therefore I decided to make one up.

 The MIPS FP ISA evolved a bit differently to the integer ISA and the 
dependencies are (roughly) as follows -- as far as the instruction set is 
concerned:

1. MIPS I   -- base FP instruction set (MIPS I FP ISA).

2. MIPS II  -- 64-bit FP memory transfers, 32-bit fixed-point FP 
               instructions added (MIPS II FP ISA).

3. MIPS III -- 64-bit fixed-point FP instructions added (MIPS III FP ISA).

4. MIPS IV  -- indexed FP memory transfers, FP multiply-accumulate, 
               reciprocal and conditional-move instructions added (MIPS IV 
               FP ISA).

5. MIPS V   -- unaligned indexed FP memory transfers, paired-single FP 
               instructions added (MIPS V FP ISA).

 Then MIPS architecture processors defined their FP ISAs as follows:

1. MIPS32   -- MIPS II FP ISA plus FP conditional-move instructions.

2. MIPS64   -- MIPS V FP ISA.

3. MIPS32r2 -- MIPS V FP ISA plus upper-half FP register transfers.

4. MIPS64r2  -- MIPS V FP ISA plus upper-half FP register transfers.

 Therefore I propose to add a set of FP ISA properties as follows to let 
test cases that include MIPS III, MIPS IV or MIPS V FP ISA instructions to 
have better coverage.  The change is imperfect as it doesn't cover MIPS32 
ISA FP conditional-move instructions, but I think it is good enough 
anyway, especially as I'm told original (r1) MIPS32 FP parts have never 
been made.  If someone needs to test such a combination, then they may 
propose a further property to cover it (like with "ror", etc.).

 With these bits in place, the change now extends the 24K erratum 
workaround coverage to some older ISAs too.  Ultimately, I think these 
cases should be passed through microMIPS testing too, to make sure the 
workaround is correctly disabled when assembling microMIPS code (there are 
no 24K parts supporting the microMIPS ASE).

 Comments?

2011-08-09  Maciej W. Rozycki  <macro@codesourcery.com>

	gas/testsuite/
	* gas/mips/mips.exp: Define new "fpisa3", "fpisa4" and "fpisa5" 
	architecture properties adding them to "mips3", "mips4", "mips5" 
	and "mips32r2" architectures.  Run the "24k-triple-stores-1"
	test for "fpisa5" instead of "mips32r2".  Run the 
	"24k-triple-stores-3" test for "mips2" instead of "mips3".

  Maciej

binutils-24k-test-mips3.diff
Index: binutils-fsf-trunk-quilt/gas/testsuite/gas/mips/mips.exp
===================================================================
--- binutils-fsf-trunk-quilt.orig/gas/testsuite/gas/mips/mips.exp	2011-08-09 00:57:25.000000000 +0100
+++ binutils-fsf-trunk-quilt/gas/testsuite/gas/mips/mips.exp	2011-08-09 23:08:44.000000000 +0100
@@ -55,6 +55,10 @@
 #		The architecture includes the instructions defined
 #		by that MIPS ISA.
 #
+#	fpisa3, fpisa4, fpisa5
+#		The architecture includes the floating-point
+#		instructions defined by that MIPS ISA.
+#
 #	gpr_ilocks
 #		The architecture interlocks GPRs accesses.  (That is,
 #		there are no load delay slots.)
@@ -369,16 +373,16 @@ mips_arch_create mips1 	32	{}	{} \
 			{ -march=mips1 -mtune=mips1 } { -mmips:3000 }
 mips_arch_create mips2 	32	mips1	{ gpr_ilocks } \
 		        { -march=mips2 -mtune=mips2 } { -mmips:6000 }
-mips_arch_create mips3 	64	mips2	{} \
+mips_arch_create mips3 	64	mips2	{ fpisa3 } \
 			{ -march=mips3 -mtune=mips3 } { -mmips:4000 }
-mips_arch_create mips4 	64	mips3	{} \
+mips_arch_create mips4 	64	mips3	{ fpisa4 } \
 			{ -march=mips4 -mtune=mips4 } { -mmips:8000 }
-mips_arch_create mips5 	64	mips4	{} \
+mips_arch_create mips5 	64	mips4	{ fpisa5 } \
 			{ -march=mips5 -mtune=mips5 } { -mmips:mips5 }
 mips_arch_create mips32	32	mips2	{} \
 			{ -march=mips32 -mtune=mips32 } { -mmips:isa32 } \
 			{ mipsisa32-*-* mipsisa32el-*-* }
-mips_arch_create mips32r2 32	mips32	{ ror } \
+mips_arch_create mips32r2 32	mips32	{ fpisa3 fpisa4 fpisa5 ror } \
 			{ -march=mips32r2 -mtune=mips32r2 } \
 			{ -mmips:isa32r2 } \
 			{ mipsisa32r2-*-* mipsisa32r2el-*-* }
@@ -492,11 +496,11 @@ if { [istarget mips*-*-vxworks*] } {
     run_dump_test_arches "24k-branch-delay-1" \
 	[mips_arch_list_matching mips1 !micromips]
     run_dump_test_arches "24k-triple-stores-1" \
-	[mips_arch_list_matching mips32r2 !octeon !micromips]
+	[mips_arch_list_matching fpisa5 !octeon !micromips]
     run_dump_test_arches "24k-triple-stores-2" \
 	[mips_arch_list_matching mips2 !micromips]
     run_dump_test_arches "24k-triple-stores-3" \
-	[mips_arch_list_matching mips3 !micromips]
+	[mips_arch_list_matching mips2 !micromips]
     run_dump_test_arches "24k-triple-stores-4" \
 	[mips_arch_list_matching mips2 !micromips]
     run_dump_test_arches "24k-triple-stores-5" \


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]