static int rs6000_sched_reorder2 (FILE *, int, rtx *, int *, int);
static int rs6000_use_sched_lookahead (void);
static int rs6000_use_sched_lookahead_guard (rtx);
+static tree rs6000_builtin_reciprocal (unsigned int, bool, bool);
static tree rs6000_builtin_mask_for_load (void);
static tree rs6000_builtin_mul_widen_even (tree);
static tree rs6000_builtin_mul_widen_odd (tree);
#undef TARGET_USE_BLOCKS_FOR_CONSTANT_P
#define TARGET_USE_BLOCKS_FOR_CONSTANT_P rs6000_use_blocks_for_constant_p
+#undef TARGET_BUILTIN_RECIPROCAL
+#define TARGET_BUILTIN_RECIPROCAL rs6000_builtin_reciprocal
+
struct gcc_target targetm = TARGET_INITIALIZER;
\f
return "li %0,%1\n\tevmergelo %0,%0,%0\n\tli %0,%2";
}
+/* Initialize TARGET of vector PAIRED to VALS. */
+
+void
+paired_expand_vector_init (rtx target, rtx vals)
+{
+ enum machine_mode mode = GET_MODE (target);
+ int n_elts = GET_MODE_NUNITS (mode);
+ int n_var = 0;
+ rtx x, new, tmp, constant_op, op1, op2;
+ int i;
+
+ for (i = 0; i < n_elts; ++i)
+ {
+ x = XVECEXP (vals, 0, i);
+ if (!CONSTANT_P (x))
+ ++n_var;
+ }
+ if (n_var == 0)
+ {
+ /* Load from constant pool. */
+ emit_move_insn (target, gen_rtx_CONST_VECTOR (mode, XVEC (vals, 0)));
+ return;
+ }
+
+ if (n_var == 2)
+ {
+ /* The vector is initialized only with non-constants. */
+ new = gen_rtx_VEC_CONCAT (V2SFmode, XVECEXP (vals, 0, 0),
+ XVECEXP (vals, 0, 1));
+
+ emit_move_insn (target, new);
+ return;
+ }
+
+ /* One field is non-constant and the other one is a constant. Load the
+ constant from the constant pool and use ps_merge instruction to
+ construct the whole vector. */
+ op1 = XVECEXP (vals, 0, 0);
+ op2 = XVECEXP (vals, 0, 1);
+
+ constant_op = (CONSTANT_P (op1)) ? op1 : op2;
+
+ tmp = gen_reg_rtx (GET_MODE (constant_op));
+ emit_move_insn (tmp, constant_op);
+
+ if (CONSTANT_P (op1))
+ new = gen_rtx_VEC_CONCAT (V2SFmode, tmp, op2);
+ else
+ new = gen_rtx_VEC_CONCAT (V2SFmode, op1, tmp);
+
+ emit_move_insn (target, new);
+}
+
/* Initialize vector TARGET to VALS. */
void
case V2SImode:
case V1DImode:
case V2SFmode:
+ /* Paired vector modes. Only reg+reg addressing is valid and
+ constant offset zero should not occur due to canonicalization.
+ Allow any offset when not strict before reload. */
+ if (TARGET_PAIRED_FLOAT)
+ return !strict;
/* SPE vector modes. */
return SPE_CONST_OFFSET_OK (offset);
if (GET_CODE (x) == PLUS
&& GET_CODE (XEXP (x, 0)) == REG
&& GET_CODE (XEXP (x, 1)) == CONST_INT
- && (unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 1)) + 0x8000) >= 0x10000)
+ && (unsigned HOST_WIDE_INT) (INTVAL (XEXP (x, 1)) + 0x8000) >= 0x10000
+ && !(SPE_VECTOR_MODE (mode)
+ || (TARGET_E500_DOUBLE && (mode == DFmode || mode == TFmode
+ || mode == DImode))))
{
HOST_WIDE_INT high_int, low_int;
rtx sum;
{ 0, CODE_FOR_paired_nmadd, "__builtin_paired_nmadd", PAIRED_BUILTIN_NMADD },
{ 0, CODE_FOR_paired_sum0, "__builtin_paired_sum0", PAIRED_BUILTIN_SUM0 },
{ 0, CODE_FOR_paired_sum1, "__builtin_paired_sum1", PAIRED_BUILTIN_SUM1 },
+ { 0, CODE_FOR_selv2sf4, "__builtin_paired_selv2sf4", PAIRED_BUILTIN_SELV2SF4 },
};
/* DST operations: void foo (void *, const int, const char). */
if (! (*insn_data[icode].operand[3].predicate) (op2, mode2))
op2 = copy_to_mode_reg (mode2, op2);
- pat = GEN_FCN (icode) (target, op0, op1, op2);
+ if (TARGET_PAIRED_FLOAT && icode == CODE_FOR_selv2sf4)
+ pat = GEN_FCN (icode) (target, op0, op1, op2, CONST0_RTX (SFmode));
+ else
+ pat = GEN_FCN (icode) (target, op0, op1, op2);
if (! pat)
return 0;
emit_insn (pat);
{
tree fndecl = TREE_OPERAND (CALL_EXPR_FN (exp), 0);
unsigned int fcode = DECL_FUNCTION_CODE (fndecl);
- struct builtin_description *d;
+ const struct builtin_description *d;
size_t i;
*expandedp = true;
}
/* Expand the paired predicates. */
- d = (struct builtin_description *) bdesc_paired_preds;
+ d = bdesc_paired_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_paired_preds); i++, d++)
if (d->code == fcode)
return paired_expand_predicate_builtin (d->icode, exp, target);
rtx ret;
bool success;
+ if (fcode == RS6000_BUILTIN_RECIP)
+ return rs6000_expand_binop_builtin (CODE_FOR_recipdf3, exp, target);
+
+ if (fcode == RS6000_BUILTIN_RECIPF)
+ return rs6000_expand_binop_builtin (CODE_FOR_recipsf3, exp, target);
+
+ if (fcode == RS6000_BUILTIN_RSQRTF)
+ return rs6000_expand_unop_builtin (CODE_FOR_rsqrtsf2, exp, target);
+
if (fcode == ALTIVEC_BUILTIN_MASK_FOR_LOAD
|| fcode == ALTIVEC_BUILTIN_MASK_FOR_STORE)
{
altivec_init_builtins ();
if (TARGET_ALTIVEC || TARGET_SPE || TARGET_PAIRED_FLOAT)
rs6000_common_init_builtins ();
+ if (TARGET_PPC_GFXOPT)
+ {
+ tree ftype = build_function_type_list (float_type_node,
+ float_type_node,
+ float_type_node,
+ NULL_TREE);
+ def_builtin (MASK_PPC_GFXOPT, "__builtin_recipdivf", ftype,
+ RS6000_BUILTIN_RECIPF);
+
+ ftype = build_function_type_list (float_type_node,
+ float_type_node,
+ NULL_TREE);
+ def_builtin (MASK_PPC_GFXOPT, "__builtin_rsqrtf", ftype,
+ RS6000_BUILTIN_RSQRTF);
+ }
+ if (TARGET_POPCNTB)
+ {
+ tree ftype = build_function_type_list (double_type_node,
+ double_type_node,
+ double_type_node,
+ NULL_TREE);
+ def_builtin (MASK_POPCNTB, "__builtin_recipdiv", ftype,
+ RS6000_BUILTIN_RECIP);
+
+ }
#if TARGET_XCOFF
/* AIX libm provides clog as __clog. */
static void
paired_init_builtins (void)
{
- struct builtin_description *d;
+ const struct builtin_description *d;
size_t i;
tree endlink = void_list_node;
PAIRED_BUILTIN_STX);
/* Predicates. */
- d = (struct builtin_description *) bdesc_paired_preds;
+ d = bdesc_paired_preds;
for (i = 0; i < ARRAY_SIZE (bdesc_paired_preds); ++i, d++)
{
tree type;
int reg_size = TARGET_32BIT ? 4 : 8;
int ehrd_size;
int save_align;
+ int first_gp;
HOST_WIDE_INT non_fixed_size;
memset (&info, 0, sizeof (info));
/* Calculate which registers need to be saved & save area size. */
info_ptr->first_gp_reg_save = first_reg_to_save ();
/* Assume that we will have to save RS6000_PIC_OFFSET_TABLE_REGNUM,
- even if it currently looks like we won't. */
+ even if it currently looks like we won't. Reload may need it to
+ get at a constant; if so, it will have already created a constant
+ pool entry for it. */
if (((TARGET_TOC && TARGET_MINIMAL_TOC)
|| (flag_pic == 1 && DEFAULT_ABI == ABI_V4)
|| (flag_pic && DEFAULT_ABI == ABI_DARWIN))
+ && current_function_uses_const_pool
&& info_ptr->first_gp_reg_save > RS6000_PIC_OFFSET_TABLE_REGNUM)
- info_ptr->gp_size = reg_size * (32 - RS6000_PIC_OFFSET_TABLE_REGNUM);
+ first_gp = RS6000_PIC_OFFSET_TABLE_REGNUM;
else
- info_ptr->gp_size = reg_size * (32 - info_ptr->first_gp_reg_save);
+ first_gp = info_ptr->first_gp_reg_save;
+
+ info_ptr->gp_size = reg_size * (32 - first_gp);
/* For the SPE, we have an additional upper 32-bits on each GPR.
Ideally we should save the entire 64-bits only when the upper
+ info_ptr->parm_size);
if (TARGET_SPE_ABI && info_ptr->spe_64bit_regs_used != 0)
- info_ptr->spe_gp_size = 8 * (32 - info_ptr->first_gp_reg_save);
+ info_ptr->spe_gp_size = 8 * (32 - first_gp);
else
info_ptr->spe_gp_size = 0;
return;
}
- /* Set sp_offset based on the stack push from the prologue. */
- if (info->total_size < 32767)
+ /* frame_reg_rtx + sp_offset points to the top of this stack frame. */
+ if (info->push_p)
sp_offset = info->total_size;
/* Restore AltiVec registers if needed. */
emit_insn (generate_set_vrsave (reg, info, 1));
}
- sp_offset = 0;
-
/* If we have a frame pointer, a call to alloca, or a large stack
frame, restore the old stack pointer using the backchain. Otherwise,
we know what size to update it with. */
emit_move_insn (frame_reg_rtx,
gen_rtx_MEM (Pmode, sp_reg_rtx));
+ sp_offset = 0;
}
- else if (info->push_p)
+ else if (info->push_p
+ && DEFAULT_ABI != ABI_V4
+ && !current_function_calls_eh_return)
{
- if (DEFAULT_ABI == ABI_V4
- || current_function_calls_eh_return)
- sp_offset = info->total_size;
- else
- {
- emit_insn (TARGET_32BIT
- ? gen_addsi3 (sp_reg_rtx, sp_reg_rtx,
- GEN_INT (info->total_size))
- : gen_adddi3 (sp_reg_rtx, sp_reg_rtx,
- GEN_INT (info->total_size)));
- }
+ emit_insn (TARGET_32BIT
+ ? gen_addsi3 (sp_reg_rtx, sp_reg_rtx,
+ GEN_INT (info->total_size))
+ : gen_adddi3 (sp_reg_rtx, sp_reg_rtx,
+ GEN_INT (info->total_size)));
+ sp_offset = 0;
}
/* Get the old lr if we saved it. */
&& info->spe_64bit_regs_used != 0
&& info->first_gp_reg_save != 32)
{
- rtx spe_save_area_ptr;
/* Determine whether we can address all of the registers that need
to be saved with an offset from the stack pointer that fits in
the small const field for SPE memory instructions. */
int spe_offset;
if (spe_regs_addressable_via_sp)
- {
- spe_save_area_ptr = frame_reg_rtx;
- spe_offset = info->spe_gp_save_offset + sp_offset;
- }
+ spe_offset = info->spe_gp_save_offset + sp_offset;
else
{
+ rtx old_frame_reg_rtx = frame_reg_rtx;
/* Make r11 point to the start of the SPE save area. We worried about
not clobbering it when we were saving registers in the prologue.
There's no need to worry here because the static chain is passed
anew to every function. */
- spe_save_area_ptr = gen_rtx_REG (Pmode, 11);
-
- emit_insn (gen_addsi3 (spe_save_area_ptr, frame_reg_rtx,
+ if (frame_reg_rtx == sp_reg_rtx)
+ frame_reg_rtx = gen_rtx_REG (Pmode, 11);
+ emit_insn (gen_addsi3 (frame_reg_rtx, old_frame_reg_rtx,
GEN_INT (info->spe_gp_save_offset + sp_offset)));
+ /* Keep the invariant that frame_reg_rtx + sp_offset points
+ at the top of the stack frame. */
+ sp_offset = -info->spe_gp_save_offset;
spe_offset = 0;
}
gcc_assert (SPE_CONST_OFFSET_OK (spe_offset + reg_size * i));
offset = GEN_INT (spe_offset + reg_size * i);
- addr = gen_rtx_PLUS (Pmode, spe_save_area_ptr, offset);
+ addr = gen_rtx_PLUS (Pmode, frame_reg_rtx, offset);
mem = gen_rtx_MEM (V2SImode, addr);
emit_move_insn (gen_rtx_REG (reg_mode, info->first_gp_reg_save + i),
/* This blockage is needed so that sched doesn't decide to move
the sp change before the register restores. */
rs6000_emit_stack_tie ();
- if (TARGET_SPE_ABI
- && info->spe_64bit_regs_used != 0
- && info->first_gp_reg_save != 32)
- emit_insn (gen_addsi3 (sp_reg_rtx, gen_rtx_REG (Pmode, 11),
- GEN_INT (-(info->spe_gp_save_offset + sp_offset))));
+ if (sp_offset != 0)
+ emit_insn (gen_addsi3 (sp_reg_rtx, frame_reg_rtx,
+ GEN_INT (sp_offset)));
else
emit_move_insn (sp_reg_rtx, frame_reg_rtx);
}
return 4 + rs6000_register_move_cost (mode, class, GENERAL_REGS);
}
+/* Returns a code for a target-specific builtin that implements
+ reciprocal of the function, or NULL_TREE if not available. */
+
+static tree
+rs6000_builtin_reciprocal (unsigned int fn, bool md_fn,
+ bool sqrt ATTRIBUTE_UNUSED)
+{
+ if (! (TARGET_RECIP && TARGET_PPC_GFXOPT && !optimize_size
+ && flag_finite_math_only && !flag_trapping_math
+ && flag_unsafe_math_optimizations))
+ return NULL_TREE;
+
+ if (md_fn)
+ return NULL_TREE;
+ else
+ switch (fn)
+ {
+ case BUILT_IN_SQRTF:
+ return rs6000_builtin_decls[RS6000_BUILTIN_RSQRTF];
+
+ default:
+ return NULL_TREE;
+ }
+}
+
/* Newton-Raphson approximation of single-precision floating point divide n/d.
Assumes no trapping math and finite arguments. */
void
-rs6000_emit_swdivsf (rtx res, rtx n, rtx d)
+rs6000_emit_swdivsf (rtx dst, rtx n, rtx d)
{
rtx x0, e0, e1, y1, u0, v0, one;
emit_insn (gen_rtx_SET (VOIDmode, v0,
gen_rtx_MINUS (SFmode, n,
gen_rtx_MULT (SFmode, d, u0))));
- /* res = u0 + v0 * y1 */
- emit_insn (gen_rtx_SET (VOIDmode, res,
+ /* dst = u0 + v0 * y1 */
+ emit_insn (gen_rtx_SET (VOIDmode, dst,
gen_rtx_PLUS (SFmode,
gen_rtx_MULT (SFmode, v0, y1), u0)));
}
Assumes no trapping math and finite arguments. */
void
-rs6000_emit_swdivdf (rtx res, rtx n, rtx d)
+rs6000_emit_swdivdf (rtx dst, rtx n, rtx d)
{
rtx x0, e0, e1, e2, y1, y2, y3, u0, v0, one;
emit_insn (gen_rtx_SET (VOIDmode, v0,
gen_rtx_MINUS (DFmode, n,
gen_rtx_MULT (DFmode, d, u0))));
- /* res = u0 + v0 * y3 */
- emit_insn (gen_rtx_SET (VOIDmode, res,
+ /* dst = u0 + v0 * y3 */
+ emit_insn (gen_rtx_SET (VOIDmode, dst,
gen_rtx_PLUS (DFmode,
gen_rtx_MULT (DFmode, v0, y3), u0)));
}
+/* Newton-Raphson approximation of single-precision floating point rsqrt.
+ Assumes no trapping math and finite arguments. */
+
+void
+rs6000_emit_swrsqrtsf (rtx dst, rtx src)
+{
+ rtx x0, x1, x2, y1, u0, u1, u2, v0, v1, v2, t0,
+ half, one, halfthree, c1, cond, label;
+
+ x0 = gen_reg_rtx (SFmode);
+ x1 = gen_reg_rtx (SFmode);
+ x2 = gen_reg_rtx (SFmode);
+ y1 = gen_reg_rtx (SFmode);
+ u0 = gen_reg_rtx (SFmode);
+ u1 = gen_reg_rtx (SFmode);
+ u2 = gen_reg_rtx (SFmode);
+ v0 = gen_reg_rtx (SFmode);
+ v1 = gen_reg_rtx (SFmode);
+ v2 = gen_reg_rtx (SFmode);
+ t0 = gen_reg_rtx (SFmode);
+ halfthree = gen_reg_rtx (SFmode);
+ cond = gen_rtx_REG (CCFPmode, CR1_REGNO);
+ label = gen_rtx_LABEL_REF (VOIDmode, gen_label_rtx ());
+
+ /* check 0.0, 1.0, NaN, Inf by testing src * src = src */
+ emit_insn (gen_rtx_SET (VOIDmode, t0,
+ gen_rtx_MULT (SFmode, src, src)));
+
+ emit_insn (gen_rtx_SET (VOIDmode, cond,
+ gen_rtx_COMPARE (CCFPmode, t0, src)));
+ c1 = gen_rtx_EQ (VOIDmode, cond, const0_rtx);
+ emit_unlikely_jump (c1, label);
+
+ half = force_reg (SFmode, CONST_DOUBLE_FROM_REAL_VALUE (dconsthalf, SFmode));
+ one = force_reg (SFmode, CONST_DOUBLE_FROM_REAL_VALUE (dconst1, SFmode));
+
+ /* halfthree = 1.5 = 1.0 + 0.5 */
+ emit_insn (gen_rtx_SET (VOIDmode, halfthree,
+ gen_rtx_PLUS (SFmode, one, half)));
+
+ /* x0 = rsqrt estimate */
+ emit_insn (gen_rtx_SET (VOIDmode, x0,
+ gen_rtx_UNSPEC (SFmode, gen_rtvec (1, src),
+ UNSPEC_RSQRT)));
+
+ /* y1 = 0.5 * src = 1.5 * src - src -> fewer constants */
+ emit_insn (gen_rtx_SET (VOIDmode, y1,
+ gen_rtx_MINUS (SFmode,
+ gen_rtx_MULT (SFmode, src, halfthree),
+ src)));
+
+ /* x1 = x0 * (1.5 - y1 * (x0 * x0)) */
+ emit_insn (gen_rtx_SET (VOIDmode, u0,
+ gen_rtx_MULT (SFmode, x0, x0)));
+ emit_insn (gen_rtx_SET (VOIDmode, v0,
+ gen_rtx_MINUS (SFmode,
+ halfthree,
+ gen_rtx_MULT (SFmode, y1, u0))));
+ emit_insn (gen_rtx_SET (VOIDmode, x1,
+ gen_rtx_MULT (SFmode, x0, v0)));
+
+ /* x2 = x1 * (1.5 - y1 * (x1 * x1)) */
+ emit_insn (gen_rtx_SET (VOIDmode, u1,
+ gen_rtx_MULT (SFmode, x1, x1)));
+ emit_insn (gen_rtx_SET (VOIDmode, v1,
+ gen_rtx_MINUS (SFmode,
+ halfthree,
+ gen_rtx_MULT (SFmode, y1, u1))));
+ emit_insn (gen_rtx_SET (VOIDmode, x2,
+ gen_rtx_MULT (SFmode, x1, v1)));
+
+ /* dst = x2 * (1.5 - y1 * (x2 * x2)) */
+ emit_insn (gen_rtx_SET (VOIDmode, u2,
+ gen_rtx_MULT (SFmode, x2, x2)));
+ emit_insn (gen_rtx_SET (VOIDmode, v2,
+ gen_rtx_MINUS (SFmode,
+ halfthree,
+ gen_rtx_MULT (SFmode, y1, u2))));
+ emit_insn (gen_rtx_SET (VOIDmode, dst,
+ gen_rtx_MULT (SFmode, x2, v2)));
+
+ emit_label (XEXP (label, 0));
+}
+
/* Emit popcount intrinsic on TARGET_POPCNTB targets. DST is the
target, and SRC is the argument operand. */
OSDN Git Service
