summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorEric Anholt <eric@anholt.net>2011-07-07 17:08:04 -0700
committerEric Anholt <eric@anholt.net>2011-07-11 08:58:37 -0700
commit46a7639174d2c55c30ec24b179cbef059fb3ca43 (patch)
tree89f6dcab412efbe34e2c3e65d7f32f8dc35a69ef
parente3ea5bc08e32119d05bce543c07c61ce93869e60 (diff)
i965: Fix fp-dst-aliasing-[12].vpfp.
There's no pretty way to avoid the overwriting of the src operands, so just use a temporary destination and rely on the MOV optimization. Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
-rw-r--r--src/mesa/drivers/dri/i965/brw_wm_fp.c23
1 files changed, 19 insertions, 4 deletions
diff --git a/src/mesa/drivers/dri/i965/brw_wm_fp.c b/src/mesa/drivers/dri/i965/brw_wm_fp.c
index 87eabf1fba1..7cd3edad235 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_fp.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_fp.c
@@ -563,13 +563,14 @@ static void precalc_dst( struct brw_wm_compile *c,
struct prog_src_register src0 = inst->SrcReg[0];
struct prog_src_register src1 = inst->SrcReg[1];
struct prog_dst_register dst = inst->DstReg;
-
+ struct prog_dst_register temp = get_temp(c);
+
if (dst.WriteMask & WRITEMASK_Y) {
/* dst.y = mul src0.y, src1.y
*/
emit_op(c,
OPCODE_MUL,
- dst_mask(dst, WRITEMASK_Y),
+ dst_mask(temp, WRITEMASK_Y),
inst->SaturateMode,
src0,
src1,
@@ -584,7 +585,7 @@ static void precalc_dst( struct brw_wm_compile *c,
*/
swz = emit_op(c,
OPCODE_SWZ,
- dst_mask(dst, WRITEMASK_XZ),
+ dst_mask(temp, WRITEMASK_XZ),
inst->SaturateMode,
src_swizzle(src0, SWIZZLE_ONE, z, z, z),
src_undef(),
@@ -597,12 +598,26 @@ static void precalc_dst( struct brw_wm_compile *c,
*/
emit_op(c,
OPCODE_MOV,
- dst_mask(dst, WRITEMASK_W),
+ dst_mask(temp, WRITEMASK_W),
inst->SaturateMode,
src1,
src_undef(),
src_undef());
}
+
+ /* This will get optimized out in general, but it ensures that we
+ * don't overwrite src operands in our channel-wise splitting
+ * above. See piglit fp-dst-aliasing-[12].
+ */
+ emit_op(c,
+ OPCODE_MOV,
+ dst,
+ 0,
+ src_reg_from_dst(temp),
+ src_undef(),
+ src_undef());
+
+ release_temp(c, temp);
}