summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/vc4/vc4_opt_algebraic.c
diff options
context:
space:
mode:
authorEric Anholt <[email protected]>2014-12-10 14:56:46 -0800
committerEric Anholt <[email protected]>2014-12-17 19:35:13 -0800
commite473fbe4690b5cbe3769042a4917f22559e2ba8d (patch)
treed2c2a467d69a4713651b40bf269db9691544baab /src/gallium/drivers/vc4/vc4_opt_algebraic.c
parentff266483fb61fd69775daf5c931ca7a56a26f4ac (diff)
vc4: Add support for turning constant uniforms into small immediates.
Small immediates have the downside of taking over the raddr B field, so you might have less chance to pack instructions together thanks to raddr B conflicts. However, it also reduces some register pressure since it lets you load 2 "uniform" values in one instruction (avoiding a previous load of the constant value to a register), and increases some pairing for the same reason. total uniforms in shared programs: 16231 -> 13374 (-17.60%) uniforms in affected programs: 10280 -> 7423 (-27.79%) total instructions in shared programs: 40795 -> 41168 (0.91%) instructions in affected programs: 25551 -> 25924 (1.46%) In a previous version of this patch I had a reduction in instruction count by forcing the other args alongside a SMALL_IMM to be in the A file or accumulators, but that increases register pressure and had a bug in handling FRAG_Z. In this patch is I just use raddr conflict resolution, which is more expensive. I think I'd rather tweak allocation to have some way to slightly prefer good choices for files in general, rather than risk failing to register allocate by forcing things into register classes.
Diffstat (limited to 'src/gallium/drivers/vc4/vc4_opt_algebraic.c')
-rw-r--r--src/gallium/drivers/vc4/vc4_opt_algebraic.c26
1 files changed, 18 insertions, 8 deletions
diff --git a/src/gallium/drivers/vc4/vc4_opt_algebraic.c b/src/gallium/drivers/vc4/vc4_opt_algebraic.c
index 4376c7ba08f..d36bb2d6596 100644
--- a/src/gallium/drivers/vc4/vc4_opt_algebraic.c
+++ b/src/gallium/drivers/vc4/vc4_opt_algebraic.c
@@ -60,23 +60,33 @@ dump_to(struct vc4_compile *c, struct qinst *inst)
}
static bool
+is_constant_value(struct vc4_compile *c, struct qinst **defs, struct qreg reg,
+ uint32_t val)
+{
+ if (reg.file == QFILE_UNIF &&
+ c->uniform_contents[reg.index] == QUNIFORM_CONSTANT &&
+ c->uniform_data[reg.index] == val) {
+ return true;
+ }
+
+ if (reg.file == QFILE_SMALL_IMM && reg.index == val)
+ return true;
+
+ return false;
+}
+
+static bool
is_zero(struct vc4_compile *c, struct qinst **defs, struct qreg reg)
{
reg = qir_follow_movs(defs, reg);
-
- return (reg.file == QFILE_UNIF &&
- c->uniform_contents[reg.index] == QUNIFORM_CONSTANT &&
- c->uniform_data[reg.index] == 0);
+ return is_constant_value(c, defs, reg, 0);
}
static bool
is_1f(struct vc4_compile *c, struct qinst **defs, struct qreg reg)
{
reg = qir_follow_movs(defs, reg);
-
- return (reg.file == QFILE_UNIF &&
- c->uniform_contents[reg.index] == QUNIFORM_CONSTANT &&
- c->uniform_data[reg.index] == fui(1.0));
+ return is_constant_value(c, defs, reg, fui(1.0));
}
static void