<div dir="ltr"><div>I do not think this patch does what you think it does. The old opcode allowed you to shift any bit size integer by a 32-bit integer. The new version allows you to shift N bits by N bits. In particular, you can't shift a 16-bit by a 32-bit value. </div>I'm not sure what the best thing is to do here. Really, the size of src1 doesn't really matter as 8-bit is enough to do any shifting needed for a 64-bit src0. You can always compose with a u2u32 to get any src1 bit size you want. We picked 32 because it's been the GL default for a long time. </div><div class="gmail_extra"> <div class="gmail_quote">On Wed, Feb 28, 2018 at 11:51 AM, Rob Clark <<a href="mailto:robdclark@gmail.com" target="_blank">robdclark@gmail.com</a>> wrote: <blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">From: Karol Herbst <<a href="mailto:kherbst@redhat.com">kherbst@redhat.com</a>> This is a thing for OpenCL kernels. Signed-off-by: Rob Clark <<a href="mailto:robdclark@gmail.com">robdclark@gmail.com</a>> --- src/compiler/nir/nir_opcodes.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/src/compiler/nir/nir_opcodes.py b/src/compiler/nir/nir_opcodes.py index 278562b2bd1..c4d2c7805eb 100644 --- a/src/compiler/nir/nir_opcodes.py +++ b/src/compiler/nir/nir_opcodes.py @@ -479,9 +479,9 @@ binop("seq", tfloat32, commutative, "(src0 == src1) ? 1.0f : 0.0f") # Set on Equ binop("sne", tfloat32, commutative, "(src0 != src1) ? 1.0f : 0.0f") # Set on Not Equal -opcode("ishl", 0, tint, [0, 0], [tint, tuint32], "", "src0 << src1") -opcode("ishr", 0, tint, [0, 0], [tint, tuint32], "", "src0 >> src1") -opcode("ushr", 0, tuint, [0, 0], [tuint, tuint32], "", "src0 >> src1") +opcode("ishl", 0, tint, [0, 0], [tint, tuint], "", "src0 << src1") +opcode("ishr", 0, tint, [0, 0], [tint, tuint], "", "src0 >> src1") +opcode("ushr", 0, tuint, [0, 0], [tuint, tuint], "", "src0 >> src1") # bitwise logic operators # -- 2.14.3 </blockquote></div> </div>