nv50/ir: fix integer mul lowering for u32 x u32 -> high u32

UNION appears to expect that all of its sources are conditionally
defined. Otherwise it inserts an unpredicated mov instruction which
overwrites the desired result. This fixes tests that use UMUL_HI, and
much less directly, unsigned integer division by a constant, which uses
this functionality in a peephole pass.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: "10.1 10.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Ben Skeggs <bskeggs@redhat.com>

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
index 63db1d7554ced6bc98277e42277209b7f1efc45e..b17d57d0bfd2cb51a8ff129184855a81853f224d 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
@@ -75,16 +75,17 @@ expandIntegerMUL(BuildUtil *bld, Instruction *mul)
    i[4] = bld->mkOp3(OP_MAD, fTy, t[3], a[0], b[0], t[2]);
 
    if (highResult) {
-      Value *r[3];
+      Value *r[4];
       Value *imm = bld->loadImm(NULL, 1 << (halfSize * 8));
       c[0] = bld->getSSA(1, FILE_FLAGS);
       c[1] = bld->getSSA(1, FILE_FLAGS);
-      for (int j = 0; j < 3; ++j)
+      for (int j = 0; j < 4; ++j)
          r[j] = bld->getSSA(fullSize);
 
       i[8] = bld->mkOp2(OP_SHR, fTy, r[0], t[1], bld->mkImm(halfSize * 8));
       i[6] = bld->mkOp2(OP_ADD, fTy, r[1], r[0], imm);
-      bld->mkOp2(OP_UNION, TYPE_U32, r[2], r[1], r[0]);
+      bld->mkMov(r[3], r[0])->setPredicate(CC_NC, c[0]);
+      bld->mkOp2(OP_UNION, TYPE_U32, r[2], r[1], r[3]);
       i[5] = bld->mkOp3(OP_MAD, fTy, mul->getDef(0), a[1], b[1], r[2]);
 
       // set carry defs / sources