diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 27dcc96db763bb4a22bb0ba1367ae77a66eba449..825208bf99a4eb834579e1ef315834ae1b3343c6 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,12 @@
+2004-06-12 Roger Sayle <roger@eyesopen.com>
+
+ * expmed.c (shift_cost, shiftadd_cost, shiftsub_cost): Additionally
+ index by machine mode.
+ (init_expmed): Initialize shift_cost, shiftadd_cost and shiftsub_cost
+ tables inside the loop over machine modes.
+ (synth_mult, expand_mult_highpart_optab, expand_mult_highpart,
+ expand_divmod): Index shift*_cost by the appropriate machine mode.
+
2004-06-12 Eric Christopher <echristo@redhat.com>
* config/s390/s390.h: Rename TARGET_TPF to TARGET_TPF_PROFILING.
diff --git a/gcc/expmed.c b/gcc/expmed.c
index 00df7ce0c7be3c97a660ea1ab350a141e8b4ee30..f45de5d56cc9a0baaf6dcab5765b5c1617040bac 100644
--- a/gcc/expmed.c
+++ b/gcc/expmed.c
@@ -94,9 +94,9 @@ static int smod_pow2_cheap[NUM_MACHINE_MODES];
static int zero_cost;
static int add_cost[NUM_MACHINE_MODES];
static int neg_cost[NUM_MACHINE_MODES];
-static int shift_cost[MAX_BITS_PER_WORD];
-static int shiftadd_cost[MAX_BITS_PER_WORD];
-static int shiftsub_cost[MAX_BITS_PER_WORD];
+static int shift_cost[NUM_MACHINE_MODES][MAX_BITS_PER_WORD];
+static int shiftadd_cost[NUM_MACHINE_MODES][MAX_BITS_PER_WORD];
+static int shiftsub_cost[NUM_MACHINE_MODES][MAX_BITS_PER_WORD];
static int mul_cost[NUM_MACHINE_MODES];
static int div_cost[NUM_MACHINE_MODES];
static int mul_widen_cost[NUM_MACHINE_MODES];
@@ -106,38 +106,24 @@ void
init_expmed (void)
{
rtx reg, shift_insn, shiftadd_insn, shiftsub_insn;
+ rtx shift_pat, shiftadd_pat, shiftsub_pat;
+ rtx pow2[MAX_BITS_PER_WORD];
+ rtx cint[MAX_BITS_PER_WORD];
int dummy;
- int m;
+ int m, n;
enum machine_mode mode, wider_mode;
start_sequence ();
- /* This is "some random pseudo register" for purposes of calling recog
- to see what insns exist. */
- reg = gen_rtx_REG (word_mode, 10000);
-
zero_cost = rtx_cost (const0_rtx, 0);
- shift_insn = emit_insn (gen_rtx_SET (VOIDmode, reg,
- gen_rtx_ASHIFT (word_mode, reg,
- const0_rtx)));
-
- shiftadd_insn
- = emit_insn (gen_rtx_SET (VOIDmode, reg,
- gen_rtx_PLUS (word_mode,
- gen_rtx_MULT (word_mode,
- reg, const0_rtx),
- reg)));
-
- shiftsub_insn
- = emit_insn (gen_rtx_SET (VOIDmode, reg,
- gen_rtx_MINUS (word_mode,
- gen_rtx_MULT (word_mode,
- reg, const0_rtx),
- reg)));
-
init_recog ();
+ for (m = 1; m < MAX_BITS_PER_WORD; m++)
+ {
+ pow2[m] = GEN_INT ((HOST_WIDE_INT) 1 << m);
+ cint[m] = GEN_INT (m);
+ }
for (mode = GET_CLASS_NARROWEST_MODE (MODE_INT);
mode != VOIDmode;
@@ -176,27 +162,52 @@ init_expmed (void)
GEN_INT (GET_MODE_BITSIZE (mode)))),
SET);
}
- }
- shift_cost[0] = 0;
- shiftadd_cost[0] = shiftsub_cost[0] = add_cost[word_mode];
-
- for (m = 1; m < MAX_BITS_PER_WORD; m++)
- {
- rtx c_int = GEN_INT ((HOST_WIDE_INT) 1 << m);
- shift_cost[m] = shiftadd_cost[m] = shiftsub_cost[m] = 32000;
-
- XEXP (SET_SRC (PATTERN (shift_insn)), 1) = GEN_INT (m);
- if (recog (PATTERN (shift_insn), shift_insn, &dummy) >= 0)
- shift_cost[m] = rtx_cost (SET_SRC (PATTERN (shift_insn)), SET);
-
- XEXP (XEXP (SET_SRC (PATTERN (shiftadd_insn)), 0), 1) = c_int;
- if (recog (PATTERN (shiftadd_insn), shiftadd_insn, &dummy) >= 0)
- shiftadd_cost[m] = rtx_cost (SET_SRC (PATTERN (shiftadd_insn)), SET);
-
- XEXP (XEXP (SET_SRC (PATTERN (shiftsub_insn)), 0), 1) = c_int;
- if (recog (PATTERN (shiftsub_insn), shiftsub_insn, &dummy) >= 0)
- shiftsub_cost[m] = rtx_cost (SET_SRC (PATTERN (shiftsub_insn)), SET);
+ shift_insn = emit_insn (gen_rtx_SET (VOIDmode, reg,
+ gen_rtx_ASHIFT (mode, reg,
+ const0_rtx)));
+
+ shiftadd_insn
+ = emit_insn (gen_rtx_SET (VOIDmode, reg,
+ gen_rtx_PLUS (mode,
+ gen_rtx_MULT (mode,
+ reg,
+ const0_rtx),
+ reg)));
+
+ shiftsub_insn
+ = emit_insn (gen_rtx_SET (VOIDmode, reg,
+ gen_rtx_MINUS (mode,
+ gen_rtx_MULT (mode,
+ reg,
+ const0_rtx),
+ reg)));
+
+ shift_pat = PATTERN (shift_insn);
+ shiftadd_pat = PATTERN (shiftadd_insn);
+ shiftsub_pat = PATTERN (shiftsub_insn);
+
+ shift_cost[mode][0] = 0;
+ shiftadd_cost[mode][0] = shiftsub_cost[mode][0] = add_cost[mode];
+
+ n = MIN (MAX_BITS_PER_WORD, GET_MODE_BITSIZE (mode));
+ for (m = 1; m < n; m++)
+ {
+ shift_cost[mode][m] = 32000;
+ XEXP (SET_SRC (shift_pat), 1) = cint[m];
+ if (recog (shift_pat, shift_insn, &dummy) >= 0)
+ shift_cost[mode][m] = rtx_cost (SET_SRC (shift_pat), SET);
+
+ shiftadd_cost[mode][m] = 32000;
+ XEXP (XEXP (SET_SRC (shiftadd_pat), 0), 1) = pow2[m];
+ if (recog (shiftadd_pat, shiftadd_insn, &dummy) >= 0)
+ shiftadd_cost[mode][m] = rtx_cost (SET_SRC (shiftadd_pat), SET);
+
+ shiftsub_cost[mode][m] = 32000;
+ XEXP (XEXP (SET_SRC (shiftsub_pat), 0), 1) = pow2[m];
+ if (recog (shiftsub_pat, shiftsub_insn, &dummy) >= 0)
+ shiftsub_cost[mode][m] = rtx_cost (SET_SRC (shiftsub_pat), SET);
+ }
}
end_sequence ();
@@ -2226,7 +2237,7 @@ synth_mult (struct algorithm *alg_out, unsigned HOST_WIDE_INT t,
if (m < BITS_PER_WORD)
{
q = t >> m;
- cost = shift_cost[m];
+ cost = shift_cost[mode][m];
synth_mult (alg_in, q, cost_limit - cost, mode);
cost += alg_in->cost;
@@ -2310,9 +2321,9 @@ synth_mult (struct algorithm *alg_out, unsigned HOST_WIDE_INT t,
d = ((unsigned HOST_WIDE_INT) 1 << m) + 1;
if (t % d == 0 && t > d && m < BITS_PER_WORD)
{
- cost = add_cost[mode] + shift_cost[m];
- if (shiftadd_cost[m] < cost)
- cost = shiftadd_cost[m];
+ cost = add_cost[mode] + shift_cost[mode][m];
+ if (shiftadd_cost[mode][m] < cost)
+ cost = shiftadd_cost[mode][m];
synth_mult (alg_in, t / d, cost_limit - cost, mode);
cost += alg_in->cost;
@@ -2331,9 +2342,9 @@ synth_mult (struct algorithm *alg_out, unsigned HOST_WIDE_INT t,
d = ((unsigned HOST_WIDE_INT) 1 << m) - 1;
if (t % d == 0 && t > d && m < BITS_PER_WORD)
{
- cost = add_cost[mode] + shift_cost[m];
- if (shiftsub_cost[m] < cost)
- cost = shiftsub_cost[m];
+ cost = add_cost[mode] + shift_cost[mode][m];
+ if (shiftsub_cost[mode][m] < cost)
+ cost = shiftsub_cost[mode][m];
synth_mult (alg_in, t / d, cost_limit - cost, mode);
cost += alg_in->cost;
@@ -2358,7 +2369,7 @@ synth_mult (struct algorithm *alg_out, unsigned HOST_WIDE_INT t,
m = exact_log2 (q);
if (m >= 0 && m < BITS_PER_WORD)
{
- cost = shiftadd_cost[m];
+ cost = shiftadd_cost[mode][m];
synth_mult (alg_in, (t - 1) >> m, cost_limit - cost, mode);
cost += alg_in->cost;
@@ -2377,7 +2388,7 @@ synth_mult (struct algorithm *alg_out, unsigned HOST_WIDE_INT t,
m = exact_log2 (q);
if (m >= 0 && m < BITS_PER_WORD)
{
- cost = shiftsub_cost[m];
+ cost = shiftsub_cost[mode][m];
synth_mult (alg_in, (t + 1) >> m, cost_limit - cost, mode);
cost += alg_in->cost;
@@ -2911,7 +2922,7 @@ expand_mult_highpart_optab (enum machine_mode mode, rtx op0, rtx op1,
/* Secondly, same as above, but use sign flavor opposite of unsignedp.
Need to adjust the result after the multiplication. */
if (size - 1 < BITS_PER_WORD
- && (mul_highpart_cost[mode] + 2 * shift_cost[size-1]
+ && (mul_highpart_cost[mode] + 2 * shift_cost[mode][size-1]
+ 4 * add_cost[mode] < max_cost))
{
moptab = unsignedp ? smul_highpart_optab : umul_highpart_optab;
@@ -2938,7 +2949,7 @@ expand_mult_highpart_optab (enum machine_mode mode, rtx op0, rtx op1,
moptab = smul_optab;
if (smul_optab->handlers[wider_mode].insn_code != CODE_FOR_nothing
&& size - 1 < BITS_PER_WORD
- && mul_cost[wider_mode] + shift_cost[size-1] < max_cost)
+ && mul_cost[wider_mode] + shift_cost[mode][size-1] < max_cost)
{
tem = expand_binop (wider_mode, moptab, op0, op1, 0,
unsignedp, OPTAB_WIDEN);
@@ -2950,7 +2961,7 @@ expand_mult_highpart_optab (enum machine_mode mode, rtx op0, rtx op1,
moptab = unsignedp ? smul_widen_optab : umul_widen_optab;
if (moptab->handlers[wider_mode].insn_code != CODE_FOR_nothing
&& size - 1 < BITS_PER_WORD
- && (mul_widen_cost[wider_mode] + 2 * shift_cost[size-1]
+ && (mul_widen_cost[wider_mode] + 2 * shift_cost[mode][size-1]
+ 4 * add_cost[mode] < max_cost))
{
tem = expand_binop (wider_mode, moptab, op0, narrow_op1,
@@ -3004,7 +3015,7 @@ expand_mult_highpart (enum machine_mode mode, rtx op0,
return expand_mult_highpart_optab (mode, op0, op1, target,
unsignedp, max_cost);
- extra_cost = shift_cost[GET_MODE_BITSIZE (mode) - 1];
+ extra_cost = shift_cost[mode][GET_MODE_BITSIZE (mode) - 1];
/* Check whether we try to multiply by a negative constant. */
if (!unsignedp && ((cnst1 >> (GET_MODE_BITSIZE (mode) - 1)) & 1))
@@ -3344,9 +3355,10 @@ expand_divmod (int rem_flag, enum tree_code code, enum machine_mode mode,
if (post_shift - 1 >= BITS_PER_WORD)
goto fail1;
- extra_cost = (shift_cost[post_shift - 1]
- + shift_cost[1]
- + 2 * add_cost[compute_mode]);
+ extra_cost
+ = (shift_cost[compute_mode][post_shift - 1]
+ + shift_cost[compute_mode][1]
+ + 2 * add_cost[compute_mode]);
t1 = expand_mult_highpart (compute_mode, op0, ml,
NULL_RTX, 1,
max_cost - extra_cost);
@@ -3376,8 +3388,9 @@ expand_divmod (int rem_flag, enum tree_code code, enum machine_mode mode,
t1 = expand_shift (RSHIFT_EXPR, compute_mode, op0,
build_int_2 (pre_shift, 0),
NULL_RTX, 1);
- extra_cost = (shift_cost[pre_shift]
- + shift_cost[post_shift]);
+ extra_cost
+ = (shift_cost[compute_mode][pre_shift]
+ + shift_cost[compute_mode][post_shift]);
t2 = expand_mult_highpart (compute_mode, t1, ml,
NULL_RTX, 1,
max_cost - extra_cost);
@@ -3511,8 +3524,8 @@ expand_divmod (int rem_flag, enum tree_code code, enum machine_mode mode,
|| size - 1 >= BITS_PER_WORD)
goto fail1;
- extra_cost = (shift_cost[post_shift]
- + shift_cost[size - 1]
+ extra_cost = (shift_cost[compute_mode][post_shift]
+ + shift_cost[compute_mode][size - 1]
+ add_cost[compute_mode]);
t1 = expand_mult_highpart (compute_mode, op0, ml,
NULL_RTX, 0,
@@ -3543,8 +3556,8 @@ expand_divmod (int rem_flag, enum tree_code code, enum machine_mode mode,
goto fail1;
ml |= (~(unsigned HOST_WIDE_INT) 0) << (size - 1);
- extra_cost = (shift_cost[post_shift]
- + shift_cost[size - 1]
+ extra_cost = (shift_cost[compute_mode][post_shift]
+ + shift_cost[compute_mode][size - 1]
+ 2 * add_cost[compute_mode]);
t1 = expand_mult_highpart (compute_mode, op0, ml,
NULL_RTX, 0,
@@ -3634,8 +3647,8 @@ expand_divmod (int rem_flag, enum tree_code code, enum machine_mode mode,
NULL_RTX, 0);
t2 = expand_binop (compute_mode, xor_optab, op0, t1,
NULL_RTX, 0, OPTAB_WIDEN);
- extra_cost = (shift_cost[post_shift]
- + shift_cost[size - 1]
+ extra_cost = (shift_cost[compute_mode][post_shift]
+ + shift_cost[compute_mode][size - 1]
+ 2 * add_cost[compute_mode]);
t3 = expand_mult_highpart (compute_mode, t2, ml,
NULL_RTX, 1,