From: Iago Toral Quiroga <ito...@igalia.com> This is pretty much the same we do with SSBOs. --- src/mesa/drivers/dri/i965/brw_fs_nir.cpp | 32 +++++++++++++++++++++++++++----- 1 file changed, 27 insertions(+), 5 deletions(-)
diff --git a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp index 9479850..bcf9288 100644 --- a/src/mesa/drivers/dri/i965/brw_fs_nir.cpp +++ b/src/mesa/drivers/dri/i965/brw_fs_nir.cpp @@ -2662,6 +2662,21 @@ fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld, /* Writemask */ unsigned writemask = instr->const_index[1]; + /* get_nir_src() retypes to integer. Be wary of 64-bit types though + * since the untyped writes below operate in units of 32-bits, which + * means that we need to write twice as many components each time. + * Also, we have to suffle 64-bit data to be in the appropriate layout + * expected by our 32-bit write messages. + */ + unsigned type_size = 4; + if (instr->src[0].ssa->bit_size == 64) { + type_size = 8; + SHUFFLE_32BIT_DATA_FOR_64BIT_WRITE(bld, val_reg, val_reg, + instr->num_components); + } + + unsigned type_slots = type_size / 4; + /* Combine groups of consecutive enabled channels in one write * message. We use ffs to find the first enabled channel and then ffs on * the bit-inverse, down-shifted writemask to determine the length of @@ -2670,22 +2685,29 @@ fs_visitor::nir_emit_cs_intrinsic(const fs_builder &bld, while (writemask) { unsigned first_component = ffs(writemask) - 1; unsigned length = ffs(~(writemask >> first_component)) - 1; - fs_reg offset_reg; + /* We can't write more than 2 64-bit components at once. Limit the + * length of the write to what we can do and let the next iteration + * handle the rest + */ + if (type_size > 4) + length = MIN2(2, length); + + fs_reg offset_reg; nir_const_value *const_offset = nir_src_as_const_value(instr->src[1]); if (const_offset) { offset_reg = brw_imm_ud(instr->const_index[0] + const_offset->u32[0] + - 4 * first_component); + type_size * first_component); } else { offset_reg = vgrf(glsl_type::uint_type); bld.ADD(offset_reg, retype(get_nir_src(instr->src[1]), BRW_REGISTER_TYPE_UD), - brw_imm_ud(instr->const_index[0] + 4 * first_component)); + brw_imm_ud(instr->const_index[0] + type_size * first_component)); } emit_untyped_write(bld, surf_index, offset_reg, - offset(val_reg, bld, first_component), - 1 /* dims */, length, + offset(val_reg, bld, first_component * type_slots), + 1 /* dims */, length * type_slots, BRW_PREDICATE_NONE); /* Clear the bits in the writemask that we just wrote, then try -- 2.5.0 _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev