From 3e294ab8931296999e2cabdd7b1817f9a461beca Mon Sep 17 00:00:00 2001 From: Iago Toral Quiroga Date: Thu, 30 Jun 2016 11:03:17 +0200 Subject: i965/vec4/tcs: fix outputs for 64-bit data v2: use byte_offset() instead of offset() Reviewed-by: Matt Turner --- src/mesa/drivers/dri/i965/brw_vec4_tcs.cpp | 31 ++++++++++++++++++++++++++++-- 1 file changed, 29 insertions(+), 2 deletions(-) (limited to 'src') diff --git a/src/mesa/drivers/dri/i965/brw_vec4_tcs.cpp b/src/mesa/drivers/dri/i965/brw_vec4_tcs.cpp index 66cc837748c..bfa22458f0e 100644 --- a/src/mesa/drivers/dri/i965/brw_vec4_tcs.cpp +++ b/src/mesa/drivers/dri/i965/brw_vec4_tcs.cpp @@ -444,13 +444,40 @@ vec4_tcs_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr) unsigned first_component = nir_intrinsic_component(instr); if (first_component) { + if (nir_src_bit_size(instr->src[0]) == 64) + first_component /= 2; assert(swiz == BRW_SWIZZLE_XYZW); swiz = BRW_SWZ_COMP_OUTPUT(first_component); mask = mask << first_component; } - emit_urb_write(swizzle(value, swiz), mask, - imm_offset, indirect_offset); + if (nir_src_bit_size(instr->src[0]) == 64) { + /* For 64-bit data we need to shuffle the data before we write and + * emit two messages. Also, since each channel is twice as large we + * need to fix the writemask in each 32-bit message to account for it. + */ + value = swizzle(retype(value, BRW_REGISTER_TYPE_DF), swiz); + dst_reg shuffled = dst_reg(this, glsl_type::dvec4_type); + shuffle_64bit_data(shuffled, value, true); + src_reg shuffled_float = src_reg(retype(shuffled, BRW_REGISTER_TYPE_F)); + + for (int n = 0; n < 2; n++) { + unsigned fixed_mask = 0; + if (mask & WRITEMASK_X) + fixed_mask |= WRITEMASK_XY; + if (mask & WRITEMASK_Y) + fixed_mask |= WRITEMASK_ZW; + emit_urb_write(shuffled_float, fixed_mask, + imm_offset, indirect_offset); + + shuffled_float = byte_offset(shuffled_float, REG_SIZE); + mask >>= 2; + imm_offset++; + } + } else { + emit_urb_write(swizzle(value, swiz), mask, + imm_offset, indirect_offset); + } break; } -- cgit v1.2.3