[Mesa-dev] [PATCH] prog_to_nir: Make RSQ properly take the absolute value of its argument.

Kenneth Graunke kenneth at whitecape.org
Thu Jun 4 02:00:39 PDT 2015


I just botched this when writing the original code.

>From the ARB_vertex_program specification:
"The RSQ instruction approximates the reciprocal of the square root of
 the absolute value of the scalar operand and replicates it to all four
 components of the result vector."

Fixes a Glean vertProg1 subtest:
RSQ test 2 (reciprocal square root of negative value)

Cc: mesa-stable at lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90547
Signed-off-by: Kenneth Graunke <kenneth at whitecape.org>
---
 src/mesa/program/prog_to_nir.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/src/mesa/program/prog_to_nir.c b/src/mesa/program/prog_to_nir.c
index e986e51..d3e3f15 100644
--- a/src/mesa/program/prog_to_nir.c
+++ b/src/mesa/program/prog_to_nir.c
@@ -764,7 +764,8 @@ ptn_emit_instruction(struct ptn_compile *c, struct prog_instruction *prog_inst)
 
    switch (op) {
    case OPCODE_RSQ:
-      ptn_move_dest(b, dest, nir_frsq(b, ptn_channel(b, src[0], X)));
+      ptn_move_dest(b, dest,
+                    nir_frsq(b, nir_fabs(b, ptn_channel(b, src[0], X))));
       break;
 
    case OPCODE_RCP:
-- 
2.4.2



More information about the mesa-dev mailing list