re PR target/65614 (PowerPC VSX systems should use XSCPSGNDP to copy scalar fp data to/from Altivec registers)

[gcc]
2015-04-06  Michael Meissner  <meissner@linux.vnet.ibm.com>

	PR target/65614
	* config/rs6000/rs6000.c (struct processor_costs): Add cost field
	for SF->DF conversions to make FLOAT_EXTEND more expensive, so
	that LFD is used to load double constants instead of LFS.  Add
	defaults for all costs structures.  Add comments for missing
	initialization fields.
	(size32_cost): Likewise.
	(size64_cost): Likewise.
	(rs64a_cost): Likewise.
	(mpccore_cost): Likewise.
	(ppc403_cost): Likewise.
	(ppc405_cost): Likewise.
	(ppc440_cost): Likewise.
	(ppc476_cost): Likewise.
	(ppc601_cost): Likewise.
	(ppc603_cost): Likewise.
	(ppc604_cost): Likewise.
	(ppc604e_cost): Likewise.
	(ppc620_cost): Likewise.
	(ppc630_cost): Likewise.
	(ppccell_cost): Likewise.
	(ppc750_cost): Likewise.
	(ppc7450_cost): Likewise.
	(ppc8540_cost): Likewise.
	(ppce300c2c3_cost): Likewise.
	(ppce500mc_cost): Likewise.
	(ppce500mc64_cost): Likewise.
	(ppce5500_cost): Likewise.
	(ppce6500_cost): Likewise.
	(titan_cost): Likewise.
	(power4_cost): Likewise.
	(power6_cost): Likewise.
	(power7_cost): Likewise.
	(power8_cost): Likewise.
	(ppca2_cost): Likewise.
	(rs6000_rtx_costs): Make FLOAT_EXTEND use SFDF_convert field.

	* config/rs6000/rs6000.md (extendsfdf2_fpr): Generate XSCPSGNDP
	instead of XXLOR to copy SFmode to clear out dirty bits created
	when SFmode denormals are generated.
	(mov<mode>_hardfloat, FMOVE32 case): Likewise.
	(truncdfsf2_fpr): Add support for ISA 2.07 XSRSP instruction.

[gcc/testsuite]
2015-04-06  Michael Meissner  <meissner@linux.vnet.ibm.com>

	PR target/65614
	* gcc.target/powerpc/compress-float-ppc-pic.c: Run test on power5
	to get floating point compression.
	* gcc.target/powerpc/compress-foat-ppc.c: Likewise.

From-SVN: r221888

6 files changed, 106 insertions, 18 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 9d22614..c060283 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,48 @@
+2015-04-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
+
+	PR target/65614
+	* config/rs6000/rs6000.c (struct processor_costs): Add cost field
+	for SF->DF conversions to make FLOAT_EXTEND more expensive, so
+	that LFD is used to load double constants instead of LFS.  Add
+	defaults for all costs structures.  Add comments for missing
+	initialization fields.
+	(size32_cost): Likewise.
+	(size64_cost): Likewise.
+	(rs64a_cost): Likewise.
+	(mpccore_cost): Likewise.
+	(ppc403_cost): Likewise.
+	(ppc405_cost): Likewise.
+	(ppc440_cost): Likewise.
+	(ppc476_cost): Likewise.
+	(ppc601_cost): Likewise.
+	(ppc603_cost): Likewise.
+	(ppc604_cost): Likewise.
+	(ppc604e_cost): Likewise.
+	(ppc620_cost): Likewise.
+	(ppc630_cost): Likewise.
+	(ppccell_cost): Likewise.
+	(ppc750_cost): Likewise.
+	(ppc7450_cost): Likewise.
+	(ppc8540_cost): Likewise.
+	(ppce300c2c3_cost): Likewise.
+	(ppce500mc_cost): Likewise.
+	(ppce500mc64_cost): Likewise.
+	(ppce5500_cost): Likewise.
+	(ppce6500_cost): Likewise.
+	(titan_cost): Likewise.
+	(power4_cost): Likewise.
+	(power6_cost): Likewise.
+	(power7_cost): Likewise.
+	(power8_cost): Likewise.
+	(ppca2_cost): Likewise.
+	(rs6000_rtx_costs): Make FLOAT_EXTEND use SFDF_convert field.
+
+	* config/rs6000/rs6000.md (extendsfdf2_fpr): Generate XSCPSGNDP
+	instead of XXLOR to copy SFmode to clear out dirty bits created
+	when SFmode denormals are generated.
+	(mov<mode>_hardfloat, FMOVE32 case): Likewise.
+	(truncdfsf2_fpr): Add support for ISA 2.07 XSRSP instruction.
+
 2015-04-06  Evandro Menezes  <e.menezes@samsung.com>
 
 	* doc/invoke.texi (AARCH64/mtune): Add exynos-m1 as an option.
diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c
index 31b46ea..4bb4221 100644
--- a/gcc/config/rs6000/rs6000.c
+++ b/gcc/config/rs6000/rs6000.c
@@ -457,6 +457,7 @@ struct processor_costs {
   const int l2_cache_size;	/* size of l2 cache, in kilobytes.  */
   const int simultaneous_prefetches; /* number of parallel prefetch
 					operations.  */
+  const int sfdf_convert;	/* cost of SF->DF conversion.  */
 };
 
 const struct processor_costs *rs6000_cost;
@@ -476,10 +477,11 @@ struct processor_costs size32_cost = {
   COSTS_N_INSNS (1),    /* dmul */
   COSTS_N_INSNS (1),    /* sdiv */
   COSTS_N_INSNS (1),    /* ddiv */
-  32,
-  0,
-  0,
-  0,
+  32,			/* cache line size */
+  0,			/* l1 cache */
+  0,			/* l2 cache */
+  0,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction size costs on 64bit processors.  */
@@ -495,10 +497,11 @@ struct processor_costs size64_cost = {
   COSTS_N_INSNS (1),    /* dmul */
   COSTS_N_INSNS (1),    /* sdiv */
   COSTS_N_INSNS (1),    /* ddiv */
-  128,
-  0,
-  0,
-  0,
+  128,			/* cache line size */
+  0,			/* l1 cache */
+  0,			/* l2 cache */
+  0,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on RS64A processors.  */
@@ -518,6 +521,7 @@ struct processor_costs rs64a_cost = {
   128,			/* l1 cache */
   2048,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on MPCCORE processors.  */
@@ -537,6 +541,7 @@ struct processor_costs mpccore_cost = {
   4,			/* l1 cache */
   16,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC403 processors.  */
@@ -556,6 +561,7 @@ struct processor_costs ppc403_cost = {
   4,			/* l1 cache */
   16,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC405 processors.  */
@@ -575,6 +581,7 @@ struct processor_costs ppc405_cost = {
   16,			/* l1 cache */
   128,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC440 processors.  */
@@ -594,6 +601,7 @@ struct processor_costs ppc440_cost = {
   32,			/* l1 cache */
   256,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC476 processors.  */
@@ -613,6 +621,7 @@ struct processor_costs ppc476_cost = {
   32,			/* l1 cache */
   512,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC601 processors.  */
@@ -632,6 +641,7 @@ struct processor_costs ppc601_cost = {
   32,			/* l1 cache */
   256,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC603 processors.  */
@@ -651,6 +661,7 @@ struct processor_costs ppc603_cost = {
   8,			/* l1 cache */
   64,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC604 processors.  */
@@ -670,6 +681,7 @@ struct processor_costs ppc604_cost = {
   16,			/* l1 cache */
   512,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC604e processors.  */
@@ -689,6 +701,7 @@ struct processor_costs ppc604e_cost = {
   32,			/* l1 cache */
   1024,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC620 processors.  */
@@ -708,6 +721,7 @@ struct processor_costs ppc620_cost = {
   32,			/* l1 cache */
   1024,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC630 processors.  */
@@ -727,6 +741,7 @@ struct processor_costs ppc630_cost = {
   64,			/* l1 cache */
   1024,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on Cell processor.  */
@@ -747,6 +762,7 @@ struct processor_costs ppccell_cost = {
   32,			/* l1 cache */
   512,			/* l2 cache */
   6,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC750 and PPC7400 processors.  */
@@ -766,6 +782,7 @@ struct processor_costs ppc750_cost = {
   32,			/* l1 cache */
   512,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC7450 processors.  */
@@ -785,6 +802,7 @@ struct processor_costs ppc7450_cost = {
   32,			/* l1 cache */
   1024,			/* l2 cache */
   1,			/* streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPC8540 processors.  */
@@ -804,6 +822,7 @@ struct processor_costs ppc8540_cost = {
   32,			/* l1 cache */
   256,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on E300C2 and E300C3 cores.  */
@@ -823,6 +842,7 @@ struct processor_costs ppce300c2c3_cost = {
   16,			/* l1 cache */
   16,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPCE500MC processors.  */
@@ -842,6 +862,7 @@ struct processor_costs ppce500mc_cost = {
   32,			/* l1 cache */
   128,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPCE500MC64 processors.  */
@@ -861,6 +882,7 @@ struct processor_costs ppce500mc64_cost = {
   32,			/* l1 cache */
   128,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPCE5500 processors.  */
@@ -880,6 +902,7 @@ struct processor_costs ppce5500_cost = {
   32,			/* l1 cache */
   128,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on PPCE6500 processors.  */
@@ -899,6 +922,7 @@ struct processor_costs ppce6500_cost = {
   32,			/* l1 cache */
   128,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on AppliedMicro Titan processors.  */
@@ -918,6 +942,7 @@ struct processor_costs titan_cost = {
   32,			/* l1 cache */
   512,			/* l2 cache */
   1,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on POWER4 and POWER5 processors.  */
@@ -937,6 +962,7 @@ struct processor_costs power4_cost = {
   32,			/* l1 cache */
   1024,			/* l2 cache */
   8,			/* prefetch streams /*/
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on POWER6 processors.  */
@@ -956,6 +982,7 @@ struct processor_costs power6_cost = {
   64,			/* l1 cache */
   2048,			/* l2 cache */
   16,			/* prefetch streams */
+  0,			/* SF->DF convert */
 };
 
 /* Instruction costs on POWER7 processors.  */
@@ -975,6 +1002,7 @@ struct processor_costs power7_cost = {
   32,			/* l1 cache */
   256,			/* l2 cache */
   12,			/* prefetch streams */
+  COSTS_N_INSNS (3),	/* SF->DF convert */
 };
 
 /* Instruction costs on POWER8 processors.  */
@@ -994,6 +1022,7 @@ struct processor_costs power8_cost = {
   32,			/* l1 cache */
   256,			/* l2 cache */
   12,			/* prefetch streams */
+  COSTS_N_INSNS (3),	/* SF->DF convert */
 };
 
 /* Instruction costs on POWER A2 processors.  */
@@ -1013,6 +1042,7 @@ struct processor_costs ppca2_cost = {
   16,			/* l1 cache */
   2048,			/* l2 cache */
   16,			/* prefetch streams */
+  0,			/* SF->DF convert */
 };
 
 
@@ -30480,7 +30510,7 @@ rs6000_rtx_costs (rtx x, int code, int outer_code, int opno ATTRIBUTE_UNUSED,
 
     case FLOAT_EXTEND:
       if (mode == DFmode)
-	*total = 0;
+	*total = rs6000_cost->sfdf_convert;
       else
 	*total = rs6000_cost->fp;
       return false;
diff --git a/gcc/config/rs6000/rs6000.md b/gcc/config/rs6000/rs6000.md
index 5025e60..0178bf4 100644
--- a/gcc/config/rs6000/rs6000.md
+++ b/gcc/config/rs6000/rs6000.md
@@ -5222,7 +5222,7 @@
    fmr %0,%1
    lfs%U1%X1 %0,%1
    #
-   xxlor %x0,%x1,%x1
+   xscpsgndp %x0,%x1,%x1
    lxsspx %x0,%y1"
   "&& reload_completed && REG_P (operands[1]) && REGNO (operands[0]) == REGNO (operands[1])"
   [(const_int 0)]
@@ -5230,7 +5230,7 @@
   emit_note (NOTE_INSN_DELETED);
   DONE;
 }
-  [(set_attr "type" "fp,fp,fpload,fp,vecsimple,fpload")])
+  [(set_attr "type" "fp,fp,fpload,fp,fp,fpload")])
 
 (define_expand "truncdfsf2"
   [(set (match_operand:SF 0 "gpc_reg_operand" "")
@@ -5239,10 +5239,12 @@
   "")
 
 (define_insn "*truncdfsf2_fpr"
-  [(set (match_operand:SF 0 "gpc_reg_operand" "=f")
-	(float_truncate:SF (match_operand:DF 1 "gpc_reg_operand" "d")))]
+  [(set (match_operand:SF 0 "gpc_reg_operand" "=f,wy")
+	(float_truncate:SF (match_operand:DF 1 "gpc_reg_operand" "d,ws")))]
   "TARGET_HARD_FLOAT && TARGET_FPRS && TARGET_DOUBLE_FLOAT"
-  "frsp %0,%1"
+  "@
+   frsp %0,%1
+   xsrsp %x0,%x1"
   [(set_attr "type" "fp")])
 
 ;; This expander is here to avoid FLOAT_WORDS_BIGENDIAN tests in
@@ -8058,7 +8060,7 @@
    lwz%U1%X1 %0,%1
    stw%U0%X0 %1,%0
    fmr %0,%1
-   xxlor %x0,%x1,%x1
+   xscpsgndp %x0,%x1,%x1
    xxlxor %x0,%x0,%x0
    li %0,0
    <f32_li>
@@ -8070,7 +8072,7 @@
    mt%0 %1
    mf%1 %0
    nop"
-  [(set_attr "type" "*,load,store,fp,vecsimple,vecsimple,integer,fpload,fpstore,fpload,fpstore,mftgpr,mffgpr,mtjmpr,mfjmpr,*")
+  [(set_attr "type" "*,load,store,fp,fp,vecsimple,integer,fpload,fpstore,fpload,fpstore,mftgpr,mffgpr,mtjmpr,mfjmpr,*")
    (set_attr "length" "4")])
 
 (define_insn "*mov<mode>_softfloat"
diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog
index 6a0d000..2c01f72 100644
--- a/gcc/testsuite/ChangeLog
+++ b/gcc/testsuite/ChangeLog
@@ -1,3 +1,10 @@
+2015-04-06  Michael Meissner  <meissner@linux.vnet.ibm.com>
+
+	PR target/65614
+	* gcc.target/powerpc/compress-float-ppc-pic.c: Run test on power5
+	to get floating point compression.
+	* gcc.target/powerpc/compress-foat-ppc.c: Likewise.
+
 2015-04-06  Jakub Jelinek  <jakub@redhat.com>
 
 	PR preprocessor/61977
diff --git a/gcc/testsuite/gcc.target/powerpc/compress-float-ppc-pic.c b/gcc/testsuite/gcc.target/powerpc/compress-float-ppc-pic.c
index aa1da52..1c8814b 100644
--- a/gcc/testsuite/gcc.target/powerpc/compress-float-ppc-pic.c
+++ b/gcc/testsuite/gcc.target/powerpc/compress-float-ppc-pic.c
@@ -1,5 +1,7 @@
 /* { dg-do compile { target powerpc_fprs } } */
-/* { dg-options "-O2 -fpic" } */
+/* { dg-options "-O2 -fpic -mcpu=power5" } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power5" } } */
+
 double foo (double x) {
   return x + 1.75;
 }
diff --git a/gcc/testsuite/gcc.target/powerpc/compress-float-ppc.c b/gcc/testsuite/gcc.target/powerpc/compress-float-ppc.c
index 312642e..290ab2b 100644
--- a/gcc/testsuite/gcc.target/powerpc/compress-float-ppc.c
+++ b/gcc/testsuite/gcc.target/powerpc/compress-float-ppc.c
@@ -1,5 +1,7 @@
 /* { dg-do compile { target powerpc_fprs } } */
-/* { dg-options "-O2" } */
+/* { dg-options "-O2 -mcpu=power5" } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power5" } } */
+
 double foo (double x) {
   return x + 1.75;
 }