<div dir="ltr"><div dir="ltr">commit dd553bc67f8ab1513fd196b6ffb7c4a76723adfd Author: Jason Ekstrand <<a href="mailto:jason.ekstrand@intel.com">jason.ekstrand@intel.com</a>> Date: Wed Oct 3 12:14:20 2018 -0500 nir/alu_to_scalar: Use ssa_for_alu_src in hand-rolled expansions The ssa_for_alu_src helper will correctly handle swizzles and other source modifiers for you. The expansions for unpack_half_2x16, pack_uvec2_to_uint, and pack_uvec4_to_uint were all broken with regards to swizzles. The brokenness of unpack_half_2x16 was causing rendering errors in Rise of the Tomb Raider on Intel ever since c11833ab24dcba26 which added an extra copy propagation to the optimization pipeline and caused us to start seeing swizzles where we hadn't seen any before. Bugzilla: <a href="https://bugs.freedesktop.org/show_bug.cgi?id=107926">https://bugs.freedesktop.org/show_bug.cgi?id=107926</a> Fixes: 9ce901058f3d "nir: Add lowering of nir_op_unpack_half_2x16." Fixes: 9b8786eba955 "nir: Add lowering support for packing opcodes." Tested-by: Alex Smith <<a href="mailto:asmith@feralinteractive.com">asmith@feralinteractive.com</a>> Tested-by: Józef Kucia <<a href="mailto:joseph.kucia@gmail.com">joseph.kucia@gmail.com</a>> Reviewed-by: Matt Turner <<a href="mailto:mattst88@gmail.com">mattst88@gmail.com</a>> <div class="gmail_quote"><div dir="ltr">On Wed, Oct 10, 2018 at 1:45 PM Ian Romanick <<a href="mailto:idr@freedesktop.org">idr@freedesktop.org</a>> wrote: </div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">These changes look good, but I'm wondering why. :) Did you find some bug, or ... ? On 10/04/2018 10:42 AM, Jason Ekstrand wrote: > --- > .../fs_pack.shader_test.mako | 14 +++++++++++--- > .../fs_unpack.shader_test.mako | 6 +++--- > .../vs_pack.shader_test.mako | 14 +++++++++++--- > .../vs_unpack.shader_test.mako | 6 +++--- > 4 files changed, 28 insertions(+), 12 deletions(-) > > diff --git a/generated_tests/templates/gen_builtin_packing_tests/fs_pack.shader_test.mako b/generated_tests/templates/gen_builtin_packing_tests/fs_pack.shader_test.mako > index 4c053bf89..86c63061e 100644 > --- a/generated_tests/templates/gen_builtin_packing_tests/fs_pack.shader_test.mako > +++ b/generated_tests/templates/gen_builtin_packing_tests/fs_pack.shader_test.mako > @@ -21,7 +21,7 @@ precision highp float; > const vec4 red = vec4(1, 0, 0, 1); > const vec4 green = vec4(0, 1, 0, 1); > > -uniform ${func.vector_type} func_input; > +uniform vec4 func_input; > > % for i in range(func.num_valid_outputs): > uniform ${func.result_precision} uint expect${i}; > @@ -31,7 +31,11 @@ out vec4 frag_color; > > void main() > { > - ${func.result_precision} uint actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input); > +% if func.vector_type == 'vec2': > + ${func.result_precision} uint actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input.wx); > +% else: > + ${func.result_precision} uint actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input.wxzy); > +% endif > > if (false > % for i in range(func.num_valid_outputs): > @@ -53,7 +57,11 @@ vertex/float/2 > > [test] > % for io in func.inout_seq: > -uniform ${func.vector_type} func_input ${" ".join(io.input)} > +% if func.vector_type == 'vec2': > +uniform vec4 func_input ${io.input[1]} 0.0 0.0 ${io.input[0]} > +% else: > +uniform vec4 func_input ${io.input[1]} ${io.input[3]} ${io.input[2]} ${io.input[0]} > +% endif > % for i in range(func.num_valid_outputs): > uniform uint expect${i} ${io.valid_outputs[i]} > % endfor > diff --git a/generated_tests/templates/gen_builtin_packing_tests/fs_unpack.shader_test.mako b/generated_tests/templates/gen_builtin_packing_tests/fs_unpack.shader_test.mako > index ee610bbfb..8e3f12a58 100644 > --- a/generated_tests/templates/gen_builtin_packing_tests/fs_unpack.shader_test.mako > +++ b/generated_tests/templates/gen_builtin_packing_tests/fs_unpack.shader_test.mako > @@ -21,7 +21,7 @@ precision highp float; > const vec4 red = vec4(1, 0, 0, 1); > const vec4 green = vec4(0, 1, 0, 1); > > -uniform highp uint func_input; > +uniform highp uvec4 func_input; > > uniform bool exact; > > @@ -33,7 +33,7 @@ out vec4 frag_color; > > void main() > { > - ${func.result_precision} ${func.vector_type} actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input); > + ${func.result_precision} ${func.vector_type} actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input.y); > > if (false > % for i in range(func.num_valid_outputs): > @@ -56,7 +56,7 @@ vertex/float/2 > > [test] > % for io in func.inout_seq: > -uniform uint func_input ${io.input} > +uniform uvec4 func_input 0 ${io.input} 0 0 > % if func.exact: > uniform int exact 1 > % else: > diff --git a/generated_tests/templates/gen_builtin_packing_tests/vs_pack.shader_test.mako b/generated_tests/templates/gen_builtin_packing_tests/vs_pack.shader_test.mako > index 5f37d7eaf..81c725846 100644 > --- a/generated_tests/templates/gen_builtin_packing_tests/vs_pack.shader_test.mako > +++ b/generated_tests/templates/gen_builtin_packing_tests/vs_pack.shader_test.mako > @@ -11,7 +11,7 @@ ${func.requirements} > const vec4 red = vec4(1, 0, 0, 1); > const vec4 green = vec4(0, 1, 0, 1); > > -uniform ${func.vector_type} func_input; > +uniform vec4 func_input; > > % for j in range(func.num_valid_outputs): > uniform ${func.result_precision} uint expect${j}; > @@ -23,7 +23,11 @@ out vec4 vert_color; > void main() > { > gl_Position = vertex; > - ${func.result_precision} uint actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input); > +% if func.vector_type == 'vec2': > + ${func.result_precision} uint actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input.zy); > +% else: > + ${func.result_precision} uint actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input.wyzx); > +% endif > > if (false > % for j in range(func.num_valid_outputs): > @@ -58,7 +62,11 @@ vertex/float/2 > > [test] > % for io in func.inout_seq: > -uniform ${func.vector_type} func_input ${" ".join(io.input)} > +% if func.vector_type == 'vec2': > +uniform vec4 func_input 0.0 ${io.input[1]} ${io.input[0]} 0.0 > +% else: > +uniform vec4 func_input ${io.input[3]} ${io.input[1]} ${io.input[2]} ${io.input[0]} > +% endif > % for j in range(func.num_valid_outputs): > uniform uint expect${j} ${io.valid_outputs[j]} > % endfor > diff --git a/generated_tests/templates/gen_builtin_packing_tests/vs_unpack.shader_test.mako b/generated_tests/templates/gen_builtin_packing_tests/vs_unpack.shader_test.mako > index 5f2a39125..c14fd35a0 100644 > --- a/generated_tests/templates/gen_builtin_packing_tests/vs_unpack.shader_test.mako > +++ b/generated_tests/templates/gen_builtin_packing_tests/vs_unpack.shader_test.mako > @@ -11,7 +11,7 @@ ${func.requirements} > const vec4 red = vec4(1, 0, 0, 1); > const vec4 green = vec4(0, 1, 0, 1); > > -uniform highp uint func_input; > +uniform highp uvec4 func_input; > > uniform bool exact; > > @@ -26,7 +26,7 @@ void main() > { > gl_Position = vertex; > > - ${func.result_precision} ${func.vector_type} actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input); > + ${func.result_precision} ${func.vector_type} actual = ${<a href="http://func.name" rel="noreferrer" target="_blank">func.name</a>}(func_input.z); > > if (false > % for i in range(func.num_valid_outputs): > @@ -61,7 +61,7 @@ vertex/float/2 > > [test] > % for io in func.inout_seq: > -uniform uint func_input ${io.input} > +uniform uvec4 func_input 0 0 ${io.input} 0 > % if func.exact: > uniform int exact 1 > % else: > </blockquote></div></div></div>