+2007-06-05 Uros Bizjak <ubizjak@gmail.com>
+
+ * config/i386/predicates/md (reg_not_xmm0_operand): New predicate.
+ (nonimm_not_xmm0_operand): Ditto.
+ * config/i386/sse.md ("sse4_1_blendvpd"): Use "reg_not_xmm0_operand"
+ as operand[0] and operand[1] predicate. Use "nonimm_not_xmm0_operand"
+ as operand[2] predicate. Require "z" class XMM register for
+ operand[3]. Adjust asm template.
+ ("sse4_1_blendvpd"): Ditto.
+ ("sse4_1_pblendvb"): Ditto.
+ * config/i386/i386.c (ix86_expand_sse_4_operands_builtin): Do not
+ force op2 into xmm0 register for variable blend instructions.
+
2007-06-04 Tom Tromey <tromey@redhat.com>
* c-tree.h (start_enum): Update.
enum machine_mode tmode = insn_data[icode].operand[0].mode;
enum machine_mode mode1 = insn_data[icode].operand[1].mode;
enum machine_mode mode2 = insn_data[icode].operand[2].mode;
- enum machine_mode mode3;
- rtx xmm0;
+ enum machine_mode mode3 = insn_data[icode].operand[3].mode;
if (VECTOR_MODE_P (mode1))
op0 = safe_vector_operand (op0, mode1);
if (VECTOR_MODE_P (mode2))
op1 = safe_vector_operand (op1, mode2);
+ if (VECTOR_MODE_P (mode3))
+ op2 = safe_vector_operand (op2, mode3);
if (optimize
|| target == 0
|| !(*insn_data[icode].operand[2].predicate) (op1, mode2))
op1 = copy_to_mode_reg (mode2, op1);
- switch (icode)
- {
- case CODE_FOR_sse4_1_blendvpd:
- case CODE_FOR_sse4_1_blendvps:
- case CODE_FOR_sse4_1_pblendvb:
- mode3 = tmode;
- op2 = safe_vector_operand (op2, mode3);
-
- if (!register_operand (op2, mode3))
+ if (! (*insn_data[icode].operand[3].predicate) (op2, mode3))
+ switch (icode)
+ {
+ case CODE_FOR_sse4_1_blendvpd:
+ case CODE_FOR_sse4_1_blendvps:
+ case CODE_FOR_sse4_1_pblendvb:
op2 = copy_to_mode_reg (mode3, op2);
+ break;
- /* ??? The third argument of variable blends must be xmm0. */
- xmm0 = gen_rtx_REG (mode3, XMM0_REG);
+ case CODE_FOR_sse4_1_roundsd:
+ case CODE_FOR_sse4_1_roundss:
+ error ("the third argument must be a 4-bit immediate");
+ return const0_rtx;
- emit_move_insn (xmm0, op2);
- op2 = xmm0;
- break;
-
- default:
- mode3 = insn_data[icode].operand[3].mode;
- if (! (*insn_data[icode].operand[3].predicate) (op2, mode3))
- {
- switch (icode)
- {
- case CODE_FOR_sse4_1_roundsd:
- case CODE_FOR_sse4_1_roundss:
- error ("the third argument must be a 4-bit immediate");
- break;
- default:
- error ("the third argument must be an 8-bit immediate");
- break;
- }
- return const0_rtx;
- }
- break;
- }
+ default:
+ error ("the third argument must be an 8-bit immediate");
+ return const0_rtx;
+ }
pat = GEN_FCN (icode) (target, op0, op1, op2);
if (! pat)
"GENERAL_REGS", \
"FP_TOP_REG", "FP_SECOND_REG", \
"FLOAT_REGS", \
- "FIRST_SSE_REG", \
+ "SSE_FIRST_REG", \
"SSE_REGS", \
"MMX_REGS", \
"FP_TOP_SSE_REGS", \
{ 0x1100ff, 0x1fe0 }, /* GENERAL_REGS */ \
{ 0x100, 0x0 }, { 0x0200, 0x0 },/* FP_TOP_REG, FP_SECOND_REG */\
{ 0xff00, 0x0 }, /* FLOAT_REGS */ \
- { 0x200000, 0x0 }, /* FIRST_SSE_REG */ \
+ { 0x200000, 0x0 }, /* SSE_FIRST_REG */ \
{ 0x1fe00000,0x1fe000 }, /* SSE_REGS */ \
{ 0xe0000000, 0x1f }, /* MMX_REGS */ \
{ 0x1fe00100,0x1fe000 }, /* FP_TOP_SSE_REG */ \
(FLAGS_REG 17)
(FPSR_REG 18)
(FPCR_REG 19)
- (XMM0_REG 21)
(R10_REG 39)
(R11_REG 40)
])
(and (match_code "reg")
(match_test "REGNO (op) == FLAGS_REG")))
+;; Return true if op is not xmm0 register.
+(define_predicate "reg_not_xmm0_operand"
+ (and (match_operand 0 "register_operand")
+ (match_test "GET_CODE (op) != REG
+ || REGNO (op) != FIRST_SSE_REG")))
+
+;; As above, but allow nonimmediate operands.
+(define_predicate "nonimm_not_xmm0_operand"
+ (and (match_operand 0 "nonimmediate_operand")
+ (match_test "GET_CODE (op) != REG
+ || REGNO (op) != FIRST_SSE_REG")))
+
;; Return 1 if VALUE can be stored in a sign extended immediate field.
(define_predicate "x86_64_immediate_operand"
(match_code "const_int,symbol_ref,label_ref,const")
(set_attr "mode" "V4SF")])
(define_insn "sse4_1_blendvpd"
- [(set (match_operand:V2DF 0 "register_operand" "=x")
- (unspec:V2DF [(match_operand:V2DF 1 "register_operand" "0")
- (match_operand:V2DF 2 "nonimmediate_operand" "xm")
- (reg:V2DF XMM0_REG)]
+ [(set (match_operand:V2DF 0 "reg_not_xmm0_operand" "=x")
+ (unspec:V2DF [(match_operand:V2DF 1 "reg_not_xmm0_operand" "0")
+ (match_operand:V2DF 2 "nonimm_not_xmm0_operand" "xm")
+ (match_operand:V2DF 3 "register_operand" "z")]
UNSPEC_BLENDV))]
"TARGET_SSE4_1"
- "blendvpd\t{%%xmm0, %2, %0|%0, %2, %%xmm0}"
+ "blendvpd\t{%3, %2, %0|%0, %2, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix_extra" "1")
(set_attr "mode" "V2DF")])
(define_insn "sse4_1_blendvps"
- [(set (match_operand:V4SF 0 "register_operand" "=x")
- (unspec:V4SF [(match_operand:V4SF 1 "register_operand" "0")
- (match_operand:V4SF 2 "nonimmediate_operand" "xm")
- (reg:V4SF XMM0_REG)]
+ [(set (match_operand:V4SF 0 "reg_not_xmm0_operand" "=x")
+ (unspec:V4SF [(match_operand:V4SF 1 "reg_not_xmm0_operand" "0")
+ (match_operand:V4SF 2 "nonimm_not_xmm0_operand" "xm")
+ (match_operand:V4SF 3 "register_operand" "z")]
UNSPEC_BLENDV))]
"TARGET_SSE4_1"
- "blendvps\t{%%xmm0, %2, %0|%0, %2, %%xmm0}"
+ "blendvps\t{%3, %2, %0|%0, %2, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix_extra" "1")
(set_attr "mode" "V4SF")])
(set_attr "mode" "TI")])
(define_insn "sse4_1_pblendvb"
- [(set (match_operand:V16QI 0 "register_operand" "=x")
- (unspec:V16QI [(match_operand:V16QI 1 "register_operand" "0")
- (match_operand:V16QI 2 "nonimmediate_operand" "xm")
- (reg:V16QI XMM0_REG)]
+ [(set (match_operand:V16QI 0 "reg_not_xmm0_operand" "=x")
+ (unspec:V16QI [(match_operand:V16QI 1 "reg_not_xmm0_operand" "0")
+ (match_operand:V16QI 2 "nonimm_not_xmm0_operand" "xm")
+ (match_operand:V16QI 3 "register_operand" "z")]
UNSPEC_BLENDV))]
"TARGET_SSE4_1"
- "pblendvb\t{%%xmm0, %2, %0|%0, %2, %%xmm0}"
+ "pblendvb\t{%3, %2, %0|%0, %2, %3}"
[(set_attr "type" "ssemov")
(set_attr "prefix_extra" "1")
(set_attr "mode" "TI")])