[PATCH], PowerPC ISA 3.0, allow QImode/HImode to go into vector registers
Michael Meissner
meissner@linux.vnet.ibm.com
Wed Nov 9 21:13:00 GMT 2016
Of course it would help, if I actually attached the patches:
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
-------------- next part --------------
Index: gcc/config/rs6000/rs6000.md
===================================================================
--- gcc/config/rs6000/rs6000.md (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/config/rs6000) (revision 241924)
+++ gcc/config/rs6000/rs6000.md (.../gcc/config/rs6000) (working copy)
@@ -325,6 +325,9 @@ (define_mode_iterator INT [QI HI SI DI T
; Any supported integer mode that fits in one register.
(define_mode_iterator INT1 [QI HI SI (DI "TARGET_POWERPC64")])
+; Integer modes supported in VSX registers with ISA 3.0 instructions
+(define_mode_iterator INT_ISA3 [QI HI SI DI])
+
; Everything we can extend QImode to.
(define_mode_iterator EXTQI [SI (DI "TARGET_POWERPC64")])
@@ -334,7 +337,7 @@ (define_mode_iterator EXTHI [SI (DI "TAR
; Everything we can extend SImode to.
(define_mode_iterator EXTSI [(DI "TARGET_POWERPC64")])
-; QImode or HImode for small atomic ops
+; QImode or HImode for small integer moves and small atomic ops
(define_mode_iterator QHI [QI HI])
; QImode, HImode, SImode for fused ops only for GPR loads
@@ -735,13 +738,15 @@ (define_code_attr SMINMAX [(smin "SM
;; complex forms. Basic data transfer is done later.
(define_insn "zero_extendqi<mode>2"
- [(set (match_operand:EXTQI 0 "gpc_reg_operand" "=r,r")
- (zero_extend:EXTQI (match_operand:QI 1 "reg_or_mem_operand" "m,r")))]
+ [(set (match_operand:EXTQI 0 "gpc_reg_operand" "=r,r,?*wJwK,?*wK")
+ (zero_extend:EXTQI (match_operand:QI 1 "reg_or_mem_operand" "m,r,Z,*wK")))]
""
"@
lbz%U1%X1 %0,%1
- rlwinm %0,%1,0,0xff"
- [(set_attr "type" "load,shift")])
+ rlwinm %0,%1,0,0xff
+ lxsibzx %x0,%y1
+ vextractub %0,%1,7"
+ [(set_attr "type" "load,shift,fpload,vecperm")])
(define_insn_and_split "*zero_extendqi<mode>2_dot"
[(set (match_operand:CC 2 "cc_reg_operand" "=x,?y")
@@ -786,13 +791,15 @@ (define_insn_and_split "*zero_extendqi<m
(define_insn "zero_extendhi<mode>2"
- [(set (match_operand:EXTHI 0 "gpc_reg_operand" "=r,r")
- (zero_extend:EXTHI (match_operand:HI 1 "reg_or_mem_operand" "m,r")))]
+ [(set (match_operand:EXTHI 0 "gpc_reg_operand" "=r,r,?*wJwK,?*wK")
+ (zero_extend:EXTHI (match_operand:HI 1 "reg_or_mem_operand" "m,r,Z,wK")))]
""
"@
lhz%U1%X1 %0,%1
- rlwinm %0,%1,0,0xffff"
- [(set_attr "type" "load,shift")])
+ rlwinm %0,%1,0,0xffff
+ lxsihzx %x0,%y1
+ vextractuh %0,%1,6"
+ [(set_attr "type" "load,shift,fpload,vecperm")])
(define_insn_and_split "*zero_extendhi<mode>2_dot"
[(set (match_operand:CC 2 "cc_reg_operand" "=x,?y")
@@ -893,11 +900,13 @@ (define_insn_and_split "*zero_extendsi<m
(define_insn "extendqi<mode>2"
- [(set (match_operand:EXTQI 0 "gpc_reg_operand" "=r")
- (sign_extend:EXTQI (match_operand:QI 1 "gpc_reg_operand" "r")))]
+ [(set (match_operand:EXTQI 0 "gpc_reg_operand" "=r,?*wK")
+ (sign_extend:EXTQI (match_operand:QI 1 "gpc_reg_operand" "r,?*wK")))]
""
- "extsb %0,%1"
- [(set_attr "type" "exts")])
+ "@
+ extsb %0,%1
+ vextsb2d %0,%1"
+ [(set_attr "type" "exts,vecperm")])
(define_insn_and_split "*extendqi<mode>2_dot"
[(set (match_operand:CC 2 "cc_reg_operand" "=x,?y")
@@ -948,14 +957,30 @@ (define_expand "extendhi<mode>2"
"")
(define_insn "*extendhi<mode>2"
- [(set (match_operand:EXTHI 0 "gpc_reg_operand" "=r,r")
- (sign_extend:EXTHI (match_operand:HI 1 "reg_or_mem_operand" "m,r")))]
+ [(set (match_operand:EXTHI 0 "gpc_reg_operand" "=r,r,?*wK,?*wK")
+ (sign_extend:EXTHI (match_operand:HI 1 "reg_or_mem_operand" "m,r,Z,wK")))]
"rs6000_gen_cell_microcode"
"@
lha%U1%X1 %0,%1
- extsh %0,%1"
- [(set_attr "type" "load,exts")
- (set_attr "sign_extend" "yes")])
+ extsh %0,%1
+ #
+ vextsh2d %0,%1"
+ [(set_attr "type" "load,exts,fpload,vecperm")
+ (set_attr "sign_extend" "yes")
+ (set_attr "length" "4,4,8,4")])
+
+(define_split
+ [(set (match_operand:EXTHI 0 "altivec_register_operand" "")
+ (sign_extend:EXTHI
+ (match_operand:HI 1 "indexed_or_indirect_operand" "")))]
+ "TARGET_P9_VECTOR && reload_completed"
+ [(set (match_dup 2)
+ (match_dup 1))
+ (set (match_dup 0)
+ (sign_extend:EXTHI (match_dup 2)))]
+{
+ operands[2] = gen_rtx_REG (HImode, REGNO (operands[1]));
+})
(define_insn "*extendhi<mode>2_noload"
[(set (match_operand:EXTHI 0 "gpc_reg_operand" "=r")
@@ -5299,30 +5324,33 @@ (define_insn_and_split "*floatunssidf2_i
(set_attr "type" "fp")])
;; ISA 3.0 adds instructions lxsi[bh]zx to directly load QImode and HImode to
-;; vector registers. At the moment, QI/HImode are not allowed in floating
-;; point or vector registers, so we use UNSPEC's to use the load byte and
-;; half-word instructions.
+;; vector registers. These insns favor doing the sign/zero extension in
+;; the vector registers, rather then loading up a GPR, doing a sign/zero
+;; extension and then a direct move.
(define_expand "float<QHI:mode><FP_ISA3:mode>2"
[(parallel [(set (match_operand:FP_ISA3 0 "vsx_register_operand")
(float:FP_ISA3
(match_operand:QHI 1 "input_operand")))
(clobber (match_scratch:DI 2))
- (clobber (match_scratch:DI 3))])]
- "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64"
+ (clobber (match_scratch:DI 3))
+ (clobber (match_scratch:<QHI:MODE> 4))])]
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64
+ && TARGET_VSX_SMALL_INTEGER"
{
if (MEM_P (operands[1]))
operands[1] = rs6000_address_for_fpconvert (operands[1]);
})
(define_insn_and_split "*float<QHI:mode><FP_ISA3:mode>2_internal"
- [(set (match_operand:FP_ISA3 0 "vsx_register_operand" "=<Fv>,<Fv>")
+ [(set (match_operand:FP_ISA3 0 "vsx_register_operand" "=<Fv>,<Fv>,<Fv>")
(float:FP_ISA3
- (match_operand:QHI 1 "reg_or_indexed_operand" "r,Z")))
- (clobber (match_scratch:DI 2 "=wi,v"))
- (clobber (match_scratch:DI 3 "=r,X"))]
+ (match_operand:QHI 1 "reg_or_indexed_operand" "wK,r,Z")))
+ (clobber (match_scratch:DI 2 "=wK,wi,wK"))
+ (clobber (match_scratch:DI 3 "=X,r,X"))
+ (clobber (match_scratch:<QHI:MODE> 4 "=X,X,wK"))]
"TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64
- && TARGET_UPPER_REGS_DI"
+ && TARGET_UPPER_REGS_DI && TARGET_VSX_SMALL_INTEGER"
"#"
"&& reload_completed"
[(const_int 0)]
@@ -5333,26 +5361,20 @@ (define_insn_and_split "*float<QHI:mode>
if (!MEM_P (input))
{
- rtx tmp = operands[3];
- emit_insn (gen_extend<QHI:mode>di2 (tmp, input));
- emit_move_insn (di, tmp);
+ if (altivec_register_operand (input, <QHI:MODE>mode))
+ emit_insn (gen_extend<QHI:mode>di2 (di, input));
+ else
+ {
+ rtx tmp = operands[3];
+ emit_insn (gen_extend<QHI:mode>di2 (tmp, input));
+ emit_move_insn (di, tmp);
+ }
}
else
{
- machine_mode vmode;
- rtx di_vector;
-
- emit_insn (gen_p9_lxsi<QHI:wd>zx (di, input));
-
- if (<MODE>mode == QImode)
- vmode = V16QImode;
- else if (<MODE>mode == HImode)
- vmode = V8HImode;
- else
- gcc_unreachable ();
-
- di_vector = gen_rtx_REG (vmode, REGNO (di));
- emit_insn (gen_vsx_sign_extend_<QHI:mode>_di (di, di_vector));
+ rtx tmp = operands[4];
+ emit_move_insn (tmp, input);
+ emit_insn (gen_extend<QHI:mode>di2 (di, tmp));
}
emit_insn (gen_floatdi<FP_ISA3:mode>2 (result, di));
@@ -5365,19 +5387,21 @@ (define_expand "floatuns<QHI:mode><FP_IS
(match_operand:QHI 1 "input_operand" "")))
(clobber (match_scratch:DI 2 ""))
(clobber (match_scratch:DI 3 ""))])]
- "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64"
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64
+ && TARGET_VSX_SMALL_INTEGER"
{
if (MEM_P (operands[1]))
operands[1] = rs6000_address_for_fpconvert (operands[1]);
})
(define_insn_and_split "*floatuns<QHI:mode><FP_ISA3:mode>2_internal"
- [(set (match_operand:FP_ISA3 0 "vsx_register_operand" "=<Fv>,<Fv>")
+ [(set (match_operand:FP_ISA3 0 "vsx_register_operand" "=<Fv>,<Fv>,<Fv>")
(unsigned_float:FP_ISA3
- (match_operand:QHI 1 "reg_or_indexed_operand" "r,Z")))
- (clobber (match_scratch:DI 2 "=wi,wi"))
- (clobber (match_scratch:DI 3 "=r,X"))]
- "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64"
+ (match_operand:QHI 1 "reg_or_indexed_operand" "wJwK,r,Z")))
+ (clobber (match_scratch:DI 2 "=wK,wi,wJwK"))
+ (clobber (match_scratch:DI 3 "=X,r,X"))]
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64
+ && TARGET_VSX_SMALL_INTEGER"
"#"
"&& reload_completed"
[(const_int 0)]
@@ -5385,15 +5409,15 @@ (define_insn_and_split "*floatuns<QHI:mo
rtx result = operands[0];
rtx input = operands[1];
rtx di = operands[2];
- rtx tmp = operands[3];
- if (!MEM_P (input))
+ if (MEM_P (input) || altivec_register_operand (input, <QHI:MODE>mode))
+ emit_insn (gen_zero_extend<QHI:mode>di2 (di, input));
+ else
{
+ rtx tmp = operands[3];
emit_insn (gen_zero_extend<QHI:mode>di2 (tmp, input));
emit_move_insn (di, tmp);
}
- else
- emit_insn (gen_p9_lxsi<QHI:wd>zx (di, input));
emit_insn (gen_floatdi<FP_ISA3:mode>2 (result, di));
DONE;
@@ -5508,19 +5532,43 @@ (define_insn "*fix_trunc<mode>di2_fctidz
[(set_attr "type" "fp")])
(define_expand "fix_trunc<SFDF:mode><QHI:mode>2"
- [(use (match_operand:QHI 0 "rs6000_nonimmediate_operand" ""))
- (use (match_operand:SFDF 1 "vsx_register_operand" ""))]
- "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64"
+ [(parallel [(set (match_operand:<QHI:MODE> 0 "nonimmediate_operand")
+ (fix:QHI (match_operand:SFDF 1 "gpc_reg_operand")))
+ (clobber (match_scratch:DI 2))])]
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE_64BIT
+ && TARGET_VSX_SMALL_INTEGER"
+{
+ if (MEM_P (operands[0]))
+ operands[0] = rs6000_address_for_fpconvert (operands[0]);
+})
+
+(define_insn_and_split "*fix_trunc<SFDF:mode><QHI:mode>2_internal"
+ [(set (match_operand:<QHI:MODE> 0 "reg_or_indexed_operand" "=wIwJ,rZ")
+ (fix:QHI
+ (match_operand:SFDF 1 "gpc_reg_operand" "<SFDF:Fv>,<SFDF:Fv>")))
+ (clobber (match_scratch:DI 2 "=X,wi"))]
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE_64BIT
+ && TARGET_VSX_SMALL_INTEGER"
+ "#"
+ "&& reload_completed"
+ [(const_int 0)]
{
- rtx op0 = operands[0];
- rtx op1 = operands[1];
- rtx di_tmp = gen_reg_rtx (DImode);
+ rtx dest = operands[0];
+ rtx src = operands[1];
- if (MEM_P (op0))
- op0 = rs6000_address_for_fpconvert (op0);
+ if (vsx_register_operand (dest, <QHI:MODE>mode))
+ {
+ rtx di_dest = gen_rtx_REG (DImode, REGNO (dest));
+ emit_insn (gen_fix_trunc<SFDF:mode>di2 (di_dest, src));
+ }
+ else
+ {
+ rtx tmp = operands[2];
+ rtx tmp2 = gen_rtx_REG (<QHI:MODE>mode, REGNO (tmp));
- emit_insn (gen_fctiwz_<SFDF:mode> (di_tmp, op1));
- emit_insn (gen_p9_stxsi<QHI:wd>x (op0, di_tmp));
+ emit_insn (gen_fix_trunc<SFDF:mode>di2 (tmp, src));
+ emit_move_insn (dest, tmp2);
+ }
DONE;
})
@@ -5597,22 +5645,45 @@ (define_insn "*fixuns_trunc<mode>di2_fct
[(set_attr "type" "fp")])
(define_expand "fixuns_trunc<SFDF:mode><QHI:mode>2"
- [(use (match_operand:QHI 0 "rs6000_nonimmediate_operand" ""))
- (use (match_operand:SFDF 1 "vsx_register_operand" ""))]
- "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE && TARGET_POWERPC64"
+ [(parallel [(set (match_operand:<QHI:MODE> 0 "nonimmediate_operand")
+ (unsigned_fix:QHI (match_operand:SFDF 1 "gpc_reg_operand")))
+ (clobber (match_scratch:DI 2))])]
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE_64BIT
+ && TARGET_VSX_SMALL_INTEGER"
+{
+ if (MEM_P (operands[0]))
+ operands[0] = rs6000_address_for_fpconvert (operands[0]);
+})
+
+(define_insn_and_split "*fixuns_trunc<SFDF:mode><QHI:mode>2_internal"
+ [(set (match_operand:<QHI:MODE> 0 "reg_or_indexed_operand" "=wIwJ,rZ")
+ (unsigned_fix:QHI
+ (match_operand:SFDF 1 "gpc_reg_operand" "<SFDF:Fv>,<SFDF:Fv>")))
+ (clobber (match_scratch:DI 2 "=X,wi"))]
+ "TARGET_P9_VECTOR && TARGET_DIRECT_MOVE_64BIT
+ && TARGET_VSX_SMALL_INTEGER"
+ "#"
+ "&& reload_completed"
+ [(const_int 0)]
{
- rtx op0 = operands[0];
- rtx op1 = operands[1];
- rtx di_tmp = gen_reg_rtx (DImode);
+ rtx dest = operands[0];
+ rtx src = operands[1];
- if (MEM_P (op0))
- op0 = rs6000_address_for_fpconvert (op0);
+ if (vsx_register_operand (dest, <QHI:MODE>mode))
+ {
+ rtx di_dest = gen_rtx_REG (DImode, REGNO (dest));
+ emit_insn (gen_fixuns_trunc<SFDF:mode>di2 (di_dest, src));
+ }
+ else
+ {
+ rtx tmp = operands[2];
+ rtx tmp2 = gen_rtx_REG (<QHI:MODE>mode, REGNO (tmp));
- emit_insn (gen_fctiwuz_<SFDF:mode> (di_tmp, op1));
- emit_insn (gen_p9_stxsi<QHI:wd>x (op0, di_tmp));
+ emit_insn (gen_fixuns_trunc<SFDF:mode>di2 (tmp, src));
+ emit_move_insn (dest, tmp2);
+ }
DONE;
})
-
; Here, we use (set (reg) (unspec:DI [(fix:SI ...)] UNSPEC_FCTIWZ))
; rather than (set (subreg:SI (reg)) (fix:SI ...))
; because the first makes it clear that operand 0 is not live
@@ -6676,41 +6747,55 @@ (define_split
(const_int 0)))]
"")
-(define_insn "*movhi_internal"
- [(set (match_operand:HI 0 "nonimmediate_operand" "=r,r,m,r,r,*c*l,*h")
- (match_operand:HI 1 "input_operand" "r,m,r,i,*h,r,0"))]
- "gpc_reg_operand (operands[0], HImode)
- || gpc_reg_operand (operands[1], HImode)"
- "@
- mr %0,%1
- lhz%U1%X1 %0,%1
- sth%U0%X0 %1,%0
- li %0,%w1
- mf%1 %0
- mt%0 %1
- nop"
- [(set_attr "type" "*,load,store,*,mfjmpr,mtjmpr,*")])
-
(define_expand "mov<mode>"
[(set (match_operand:INT 0 "general_operand" "")
(match_operand:INT 1 "any_operand" ""))]
""
"{ rs6000_emit_move (operands[0], operands[1], <MODE>mode); DONE; }")
-(define_insn "*movqi_internal"
- [(set (match_operand:QI 0 "nonimmediate_operand" "=r,r,m,r,r,*c*l,*h")
- (match_operand:QI 1 "input_operand" "r,m,r,i,*h,r,0"))]
- "gpc_reg_operand (operands[0], QImode)
- || gpc_reg_operand (operands[1], QImode)"
+;; MR LHZ/LBZ LXSI*ZX STH/STB STXSI*X LI
+;; XXLOR load 0 load -1 VSPLTI* # MFVSRWZ
+;; MTVSRWZ MF%1 MT%1 NOP
+(define_insn "*mov<mode>_internal"
+ [(set (match_operand:QHI 0 "nonimmediate_operand"
+ "=r, r, ?*wJwK, m, Z, r,
+ ?*wJwK, ?*wJwK, ?*wJwK, ?*wK, ?*wK, r,
+ ?*wJwK, r, *c*l, *h")
+
+ (match_operand:QHI 1 "input_operand"
+ "r, m, Z, r, wJwK, i,
+ wJwK, O, wM, wB, wS, ?*wJwK,
+ r, *h, r, 0"))]
+
+ "gpc_reg_operand (operands[0], <MODE>mode)
+ || gpc_reg_operand (operands[1], <MODE>mode)"
"@
mr %0,%1
- lbz%U1%X1 %0,%1
- stb%U0%X0 %1,%0
+ l<wd>z%U1%X1 %0,%1
+ lxsi<wd>zx %x0,%y1
+ st<wd>%U0%X0 %1,%0
+ stxsi<wd>x %1,%y0
li %0,%1
+ xxlor %x0,%x1,%x1
+ xxspltib %x0,0
+ xxspltib %x0,255
+ vspltis<wd> %0,%1
+ #
+ mfvsrwz %0,%x1
+ mtvsrwz %x0,%1
mf%1 %0
mt%0 %1
nop"
- [(set_attr "type" "*,load,store,*,mfjmpr,mtjmpr,*")])
+ [(set_attr "type"
+ "*, load, fpload, store, fpstore, *,
+ vecsimple, vecperm, vecperm, vecperm, vecperm, mftgpr,
+ mffgpr, mfjmpr, mtjmpr, *")
+
+ (set_attr "length"
+ "4, 4, 4, 4, 4, 4,
+ 4, 4, 4, 4, 8, 4,
+ 4, 4, 4, 4")])
+
;; Here is how to move condition codes around. When we store CC data in
;; an integer register or memory, we store just the high-order 4 bits.
@@ -8134,7 +8219,7 @@ (define_insn "*movdi_internal64"
xxlor %x0,%x1,%x1
xxspltib %x0,0
xxspltib %x0,255
- vspltisw %0,%1
+ #
xxlxor %x0,%x0,%x0
xxlorc %x0,%x0,%x0
#
@@ -8228,9 +8313,11 @@ (define_split
DONE;
})
+;; Split integer constants that can be loaded with XXSPLTIB and a
+;; sign extend operation.
(define_split
- [(set (match_operand:DI 0 "altivec_register_operand" "")
- (match_operand:DI 1 "xxspltib_constant_split" ""))]
+ [(set (match_operand:INT_ISA3 0 "altivec_register_operand" "")
+ (match_operand:INT_ISA3 1 "xxspltib_constant_split" ""))]
"TARGET_UPPER_REGS_DI && TARGET_P9_VECTOR && reload_completed"
[(const_int 0)]
{
@@ -8240,7 +8327,15 @@ (define_split
rtx op0_v16qi = gen_rtx_REG (V16QImode, r);
emit_insn (gen_xxspltib_v16qi (op0_v16qi, op1));
- emit_insn (gen_vsx_sign_extend_qi_di (operands[0], op0_v16qi));
+ if (<MODE>mode == DImode)
+ emit_insn (gen_vsx_sign_extend_qi_di (operands[0], op0_v16qi));
+ else if (<MODE>mode == SImode)
+ emit_insn (gen_vsx_sign_extend_qi_si (operands[0], op0_v16qi));
+ else if (<MODE>mode == HImode)
+ {
+ rtx op0_v8hi = gen_rtx_REG (V8HImode, r);
+ emit_insn (gen_altivec_vupkhsb (op0_v8hi, op0_v16qi));
+ }
DONE;
})
Index: gcc/config/rs6000/rs6000.c
===================================================================
--- gcc/config/rs6000/rs6000.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/config/rs6000) (revision 241924)
+++ gcc/config/rs6000/rs6000.c (.../gcc/config/rs6000) (working copy)
@@ -2019,8 +2019,14 @@ rs6000_hard_regno_mode_ok (int regno, ma
if(GET_MODE_SIZE (mode) == UNITS_PER_FP_WORD)
return 1;
- if (TARGET_VSX_SMALL_INTEGER && mode == SImode)
- return 1;
+ if (TARGET_VSX_SMALL_INTEGER)
+ {
+ if (mode == SImode)
+ return 1;
+
+ if (TARGET_P9_VECTOR && (mode == HImode || mode == QImode))
+ return 1;
+ }
}
if (PAIRED_SIMD_REGNO_P (regno) && TARGET_PAIRED_FLOAT
@@ -3403,7 +3409,14 @@ rs6000_init_hard_regno_mode_ok (bool glo
reg_addr[SFmode].scalar_in_vmx_p = true;
if (TARGET_VSX_SMALL_INTEGER)
- reg_addr[SImode].scalar_in_vmx_p = true;
+ {
+ reg_addr[SImode].scalar_in_vmx_p = true;
+ if (TARGET_P9_VECTOR)
+ {
+ reg_addr[HImode].scalar_in_vmx_p = true;
+ reg_addr[QImode].scalar_in_vmx_p = true;
+ }
+ }
}
/* Setup the fusion operations. */
@@ -20607,8 +20620,14 @@ rs6000_secondary_reload_simple_move (enu
}
/* ISA 2.07: MTVSRWZ or MFVSRWZ. */
- if (TARGET_VSX_SMALL_INTEGER && mode == SImode)
- return true;
+ if (TARGET_VSX_SMALL_INTEGER)
+ {
+ if (mode == SImode)
+ return true;
+
+ if (TARGET_P9_VECTOR && (mode == HImode || mode == QImode))
+ return true;
+ }
/* ISA 2.07: MTVSRWZ or MFVSRWZ. */
if (mode == SDmode)
@@ -21413,6 +21432,33 @@ rs6000_preferred_reload_class (rtx x, en
if (GET_CODE (x) == CONST_VECTOR && easy_vector_constant (x, mode))
return ALTIVEC_REGS;
+ /* If this is an integer constant that can easily be loaded into
+ vector registers, allow it. */
+ if (CONST_INT_P (x))
+ {
+ HOST_WIDE_INT value = INTVAL (x);
+
+ /* ISA 2.07 can generate -1 in all registers with XXLORC. ISA
+ 2.06 can generate it in the Altivec registers with
+ VSPLTI<x>. */
+ if (value == -1)
+ {
+ if (TARGET_P8_VECTOR)
+ return rclass;
+ else if (rclass == ALTIVEC_REGS || rclass == VSX_REGS)
+ return ALTIVEC_REGS;
+ else
+ return NO_REGS;
+ }
+
+ /* ISA 3.0 can load -128..127 using the XXSPLTIB instruction and
+ a sign extend in the Altivec registers. */
+ if (IN_RANGE (value, -128, 127) && TARGET_P9_VECTOR
+ && TARGET_VSX_SMALL_INTEGER
+ && (rclass == ALTIVEC_REGS || rclass == VSX_REGS))
+ return ALTIVEC_REGS;
+ }
+
/* Force constant to memory. */
return NO_REGS;
}
Index: gcc/config/rs6000/vsx.md
===================================================================
--- gcc/config/rs6000/vsx.md (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/config/rs6000) (revision 241924)
+++ gcc/config/rs6000/vsx.md (.../gcc/config/rs6000) (working copy)
@@ -338,7 +338,6 @@ (define_c_enum "unspec"
UNSPEC_VSX_XVCVDPSXDS
UNSPEC_VSX_XVCVDPUXDS
UNSPEC_VSX_SIGN_EXTEND
- UNSPEC_P9_MEMORY
UNSPEC_VSX_VSLO
UNSPEC_VSX_EXTRACT
UNSPEC_VSX_SXEXPDP
@@ -2519,72 +2518,29 @@ (define_expand "vec_perm_const<mode>"
;; types are currently allowed in a vector register, so we extract to a DImode
;; and either do a direct move or store.
(define_expand "vsx_extract_<mode>"
- [(parallel [(set (match_operand:<VS_scalar> 0 "nonimmediate_operand")
+ [(parallel [(set (match_operand:<VS_scalar> 0 "gpc_reg_operand")
(vec_select:<VS_scalar>
(match_operand:VSX_EXTRACT_I 1 "gpc_reg_operand")
(parallel [(match_operand:QI 2 "const_int_operand")])))
- (clobber (match_dup 3))])]
+ (clobber (match_scratch:VSX_EXTRACT_I 3))])]
"VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_DIRECT_MOVE_64BIT"
{
- machine_mode smode = ((<MODE>mode != V4SImode && TARGET_VEXTRACTUB)
- ? DImode : <MODE>mode);
- operands[3] = gen_rtx_SCRATCH (smode);
-})
-
-;; Under ISA 3.0, we can use the byte/half-word/word integer stores if we are
-;; extracting a vector element and storing it to memory, rather than using
-;; direct move to a GPR and a GPR store.
-(define_insn_and_split "*vsx_extract_<mode>_p9"
- [(set (match_operand:<VS_scalar> 0 "nonimmediate_operand" "=r,Z")
- (vec_select:<VS_scalar>
- (match_operand:VSX_EXTRACT_I2 1 "gpc_reg_operand" "v,v")
- (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "n,n")])))
- (clobber (match_scratch:DI 3 "=v,v"))]
- "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_VEXTRACTUB"
- "#"
- "&& (reload_completed || MEM_P (operands[0]))"
- [(const_int 0)]
-{
- rtx dest = operands[0];
- rtx src = operands[1];
- rtx element = operands[2];
- rtx di_tmp = operands[3];
-
- if (GET_CODE (di_tmp) == SCRATCH)
- di_tmp = gen_reg_rtx (DImode);
-
- emit_insn (gen_vsx_extract_<mode>_di (di_tmp, src, element));
-
- if (REG_P (dest))
- emit_move_insn (gen_rtx_REG (DImode, REGNO (dest)), di_tmp);
- else if (SUBREG_P (dest))
- emit_move_insn (gen_rtx_REG (DImode, subreg_regno (dest)), di_tmp);
- else if (MEM_P (operands[0]))
+ /* If we have ISA 3.0, we can do a xxextractuw/vextractu{b,h}. */
+ if (TARGET_VSX_SMALL_INTEGER && TARGET_P9_VECTOR)
{
- if (can_create_pseudo_p ())
- dest = rs6000_address_for_fpconvert (dest);
-
- if (<MODE>mode == V16QImode)
- emit_insn (gen_p9_stxsibx (dest, di_tmp));
- else if (<MODE>mode == V8HImode)
- emit_insn (gen_p9_stxsihx (dest, di_tmp));
- else
- gcc_unreachable ();
+ emit_insn (gen_vsx_extract_<mode>_p9 (operands[0], operands[1],
+ operands[2]));
+ DONE;
}
- else
- gcc_unreachable ();
-
- DONE;
-}
- [(set_attr "type" "vecsimple,fpstore")])
+})
-(define_insn "vsx_extract_<mode>_di"
- [(set (match_operand:DI 0 "gpc_reg_operand" "=<VSX_EX>")
- (zero_extend:DI
- (vec_select:<VS_scalar>
- (match_operand:VSX_EXTRACT_I 1 "gpc_reg_operand" "<VSX_EX>")
- (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "n")]))))]
- "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_VEXTRACTUB"
+(define_insn "vsx_extract_<mode>_p9"
+ [(set (match_operand:<VS_scalar> 0 "gpc_reg_operand" "=<VSX_EX>")
+ (vec_select:<VS_scalar>
+ (match_operand:VSX_EXTRACT_I 1 "gpc_reg_operand" "<VSX_EX>")
+ (parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "n")])))]
+ "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_VEXTRACTUB
+ && TARGET_VSX_SMALL_INTEGER"
{
/* Note, the element number has already been adjusted for endianness, so we
don't have to adjust it here. */
@@ -2599,13 +2555,51 @@ (define_insn "vsx_extract_<mode>_di"
}
[(set_attr "type" "vecsimple")])
+;; Optimize zero extracts to eliminate the AND after the extract.
+(define_insn_and_split "*vsx_extract_<mode>_di_p9"
+ [(set (match_operand:DI 0 "gpc_reg_operand" "=<VSX_EX>")
+ (zero_extend:DI
+ (vec_select:<VS_scalar>
+ (match_operand:VSX_EXTRACT_I 1 "gpc_reg_operand" "<VSX_EX>")
+ (parallel [(match_operand:QI 2 "const_int_operand" "n")]))))]
+ "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_VEXTRACTUB
+ && TARGET_VSX_SMALL_INTEGER"
+ "#"
+ "&& reload_completed"
+ [(set (match_dup 3)
+ (vec_select:<VS_scalar>
+ (match_dup 1)
+ (parallel [(match_dup 2)])))]
+{
+ operands[3] = gen_rtx_REG (<VS_scalar>mode, REGNO (operands[0]));
+})
+
+;; Optimize stores to use the ISA 3.0 scalar store instructions
+(define_insn_and_split "*vsx_extract_<mode>_store_p9"
+ [(set (match_operand:<VS_scalar> 0 "memory_operand" "=Z")
+ (vec_select:<VS_scalar>
+ (match_operand:VSX_EXTRACT_I 1 "gpc_reg_operand" "<VSX_EX>")
+ (parallel [(match_operand:QI 2 "const_int_operand" "n")])))
+ (clobber (match_scratch:<VS_scalar> 3 "=<VSX_EX>"))]
+ "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_VEXTRACTUB
+ && TARGET_VSX_SMALL_INTEGER"
+ "#"
+ "&& reload_completed"
+ [(set (match_dup 3)
+ (vec_select:<VS_scalar>
+ (match_dup 1)
+ (parallel [(match_dup 2)])))
+ (set (match_dup 0)
+ (match_dup 3))])
+
(define_insn_and_split "*vsx_extract_si"
[(set (match_operand:SI 0 "nonimmediate_operand" "=r,wHwI,Z")
(vec_select:SI
(match_operand:V4SI 1 "gpc_reg_operand" "wJv,wJv,wJv")
(parallel [(match_operand:QI 2 "const_0_to_3_operand" "n,n,n")])))
(clobber (match_scratch:V4SI 3 "=wJv,wJv,wJv"))]
- "VECTOR_MEM_VSX_P (V4SImode) && TARGET_DIRECT_MOVE_64BIT"
+ "VECTOR_MEM_VSX_P (V4SImode) && TARGET_DIRECT_MOVE_64BIT
+ && (!TARGET_P9_VECTOR || !TARGET_VSX_SMALL_INTEGER)"
"#"
"&& reload_completed"
[(const_int 0)]
@@ -2624,10 +2618,10 @@ (define_insn_and_split "*vsx_extract_si
value = INTVAL (element);
if (value != 1)
{
- if (TARGET_VEXTRACTUB)
+ if (TARGET_P9_VECTOR && TARGET_VSX_SMALL_INTEGER)
{
- rtx di_tmp = gen_rtx_REG (DImode, REGNO (vec_tmp));
- emit_insn (gen_vsx_extract_v4si_di (di_tmp,src, element));
+ rtx si_tmp = gen_rtx_REG (SImode, REGNO (vec_tmp));
+ emit_insn (gen_vsx_extract_v4si_p9 (si_tmp,src, element));
}
else
emit_insn (gen_altivec_vspltw_direct (vec_tmp, src, element));
@@ -2663,7 +2657,8 @@ (define_insn_and_split "*vsx_extract_<m
(match_operand:VSX_EXTRACT_I2 1 "gpc_reg_operand" "v")
(parallel [(match_operand:QI 2 "<VSX_EXTRACT_PREDICATE>" "n")])))
(clobber (match_scratch:VSX_EXTRACT_I2 3 "=v"))]
- "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_DIRECT_MOVE_64BIT"
+ "VECTOR_MEM_VSX_P (<MODE>mode) && TARGET_DIRECT_MOVE_64BIT
+ && (!TARGET_P9_VECTOR || !TARGET_VSX_SMALL_INTEGER)"
"#"
"&& reload_completed"
[(const_int 0)]
@@ -3253,26 +3248,6 @@ (define_insn "*vsx_sign_extend_si_v2di"
[(set_attr "type" "vecexts")])
-;; ISA 3.0 memory operations
-(define_insn "p9_lxsi<wd>zx"
- [(set (match_operand:DI 0 "vsx_register_operand" "=wi")
- (unspec:DI [(zero_extend:DI
- (match_operand:QHI 1 "indexed_or_indirect_operand" "Z"))]
- UNSPEC_P9_MEMORY))]
- "TARGET_P9_VECTOR"
- "lxsi<wd>zx %x0,%y1"
- [(set_attr "type" "fpload")])
-
-(define_insn "p9_stxsi<wd>x"
- [(set (match_operand:QHI 0 "reg_or_indexed_operand" "=r,Z")
- (unspec:QHI [(match_operand:DI 1 "vsx_register_operand" "wi,wi")]
- UNSPEC_P9_MEMORY))]
- "TARGET_P9_VECTOR"
- "@
- mfvsrd %0,%x1
- stxsi<wd>x %x1,%y0"
- [(set_attr "type" "mffgpr,fpstore")])
-
;; ISA 3.0 Binary Floating-Point Support
;; VSX Scalar Extract Exponent Double-Precision
Index: gcc/testsuite/gcc.target/powerpc/vsx-qimode.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/vsx-qimode.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/vsx-qimode.c (.../gcc/testsuite/gcc.target/powerpc) (revision 242012)
@@ -0,0 +1,22 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
+/* { dg-options "-mcpu=power9 -O2 -mvsx-small-integer" } */
+
+double load_asm_d_constraint (signed char *p)
+{
+ double ret;
+ __asm__ ("xxlor %x0,%x1,%x1\t# load d constraint" : "=d" (ret) : "d" (*p));
+ return ret;
+}
+
+void store_asm_d_constraint (signed char *p, double x)
+{
+ signed char i;
+ __asm__ ("xxlor %x0,%x1,%x1\t# store d constraint" : "=d" (i) : "d" (x));
+ *p = i;
+}
+
+/* { dg-final { scan-assembler "lxsibzx" } } */
+/* { dg-final { scan-assembler "stxsibx" } } */
Index: gcc/testsuite/gcc.target/powerpc/vsx-himode.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/vsx-himode.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/vsx-himode.c (.../gcc/testsuite/gcc.target/powerpc) (revision 242012)
@@ -0,0 +1,22 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
+/* { dg-options "-mcpu=power9 -O2 -mvsx-small-integer" } */
+
+double load_asm_d_constraint (short *p)
+{
+ double ret;
+ __asm__ ("xxlor %x0,%x1,%x1\t# load d constraint" : "=d" (ret) : "d" (*p));
+ return ret;
+}
+
+void store_asm_d_constraint (short *p, double x)
+{
+ short i;
+ __asm__ ("xxlor %x0,%x1,%x1\t# store d constraint" : "=d" (i) : "d" (x));
+ *p = i;
+}
+
+/* { dg-final { scan-assembler "lxsihzx" } } */
+/* { dg-final { scan-assembler "stxsihx" } } */
Index: gcc/testsuite/gcc.target/powerpc/vsx-himode2.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/vsx-himode2.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/vsx-himode2.c (.../gcc/testsuite/gcc.target/powerpc) (revision 242012)
@@ -0,0 +1,15 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
+/* { dg-options "-mcpu=power9 -O2 -mvsx-small-integer" } */
+
+unsigned int foo (unsigned short u)
+{
+ unsigned int ret;
+ __asm__ ("xxlor %x0,%x1,%x1\t# v, v constraints" : "=v" (ret) : "v" (u));
+ return ret;
+}
+
+/* { dg-final { scan-assembler "mtvsrwz" } } */
+/* { dg-final { scan-assembler "mfvsrwz" } } */
Index: gcc/testsuite/gcc.target/powerpc/vsx-himode3.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/vsx-himode3.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/vsx-himode3.c (.../gcc/testsuite/gcc.target/powerpc) (revision 242012)
@@ -0,0 +1,22 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
+/* { dg-options "-mcpu=power9 -O2 -mvsx-small-integer" } */
+
+double load_asm_v_constraint (short *p)
+{
+ double ret;
+ __asm__ ("xxlor %x0,%x1,%x1\t# load v constraint" : "=d" (ret) : "v" (*p));
+ return ret;
+}
+
+void store_asm_v_constraint (short *p, double x)
+{
+ short i;
+ __asm__ ("xxlor %x0,%x1,%x1\t# store v constraint" : "=v" (i) : "d" (x));
+ *p = i;
+}
+
+/* { dg-final { scan-assembler "lxsihzx" } } */
+/* { dg-final { scan-assembler "stxsihx" } } */
Index: gcc/testsuite/gcc.target/powerpc/vsx-qimode2.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/vsx-qimode2.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/vsx-qimode2.c (.../gcc/testsuite/gcc.target/powerpc) (revision 242012)
@@ -0,0 +1,15 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
+/* { dg-options "-mcpu=power9 -O2 -mvsx-small-integer" } */
+
+unsigned int foo (unsigned char u)
+{
+ unsigned int ret;
+ __asm__ ("xxlor %x0,%x1,%x1\t# v, v constraints" : "=v" (ret) : "v" (u));
+ return ret;
+}
+
+/* { dg-final { scan-assembler "mtvsrwz" } } */
+/* { dg-final { scan-assembler "mfvsrwz" } } */
Index: gcc/testsuite/gcc.target/powerpc/vsx-qimode3.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/vsx-qimode3.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/vsx-qimode3.c (.../gcc/testsuite/gcc.target/powerpc) (revision 242012)
@@ -0,0 +1,22 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-skip-if "" { powerpc*-*-darwin* } { "*" } { "" } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-skip-if "do not override -mcpu" { powerpc*-*-* } { "-mcpu=*" } { "-mcpu=power9" } } */
+/* { dg-options "-mcpu=power9 -O2 -mvsx-small-integer" } */
+
+double load_asm_v_constraint (signed char *p)
+{
+ double ret;
+ __asm__ ("xxlor %x0,%x1,%x1\t# load v constraint" : "=d" (ret) : "v" (*p));
+ return ret;
+}
+
+void store_asm_v_constraint (signed char *p, double x)
+{
+ signed char i;
+ __asm__ ("xxlor %x0,%x1,%x1\t# store v constraint" : "=v" (i) : "d" (x));
+ *p = i;
+}
+
+/* { dg-final { scan-assembler "lxsibzx" } } */
+/* { dg-final { scan-assembler "stxsibx" } } */
Property changes on: gcc/testsuite/gcc.target/powerpc/p9-minmax-1.c
___________________________________________________________________
Modified: svn:mergeinfo
Merged /trunk/gcc/testsuite/gcc.target/powerpc/p9-minmax-1.c:r241733-241924
Property changes on: gcc/testsuite/gcc.target/powerpc/p9-minmax-2.c
___________________________________________________________________
Modified: svn:mergeinfo
Merged /trunk/gcc/testsuite/gcc.target/powerpc/p9-minmax-2.c:r241733-241924
Index: gcc/testsuite/gcc.target/powerpc/p9-extract-1.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/p9-extract-1.c (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk/gcc/testsuite/gcc.target/powerpc) (revision 241924)
+++ gcc/testsuite/gcc.target/powerpc/p9-extract-1.c (.../gcc/testsuite/gcc.target/powerpc) (working copy)
@@ -17,7 +17,7 @@ int extract_schar_3 (vector signed char
/* { dg-final { scan-assembler "vextractub" } } */
/* { dg-final { scan-assembler "vextractuh" } } */
/* { dg-final { scan-assembler "xxextractuw" } } */
-/* { dg-final { scan-assembler "mfvsrd" } } */
+/* { dg-final { scan-assembler "mfvsr" } } */
/* { dg-final { scan-assembler-not "stxvd2x" } } */
/* { dg-final { scan-assembler-not "stxv" } } */
/* { dg-final { scan-assembler-not "lwa" } } */
More information about the Gcc-patches
mailing list