Mesa (main): zink: fix 32bit bo rewriting
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Fri Jun 10 02:49:40 UTC 2022
Module: Mesa
Branch: main
Commit: bbe5136658f048b6403f31f0d01dff856dd3ff81
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=bbe5136658f048b6403f31f0d01dff856dd3ff81
Author: Mike Blumenkrantz <michael.blumenkrantz at gmail.com>
Date: Fri May 20 11:04:11 2022 -0400
zink: fix 32bit bo rewriting
this was correct for 64bit loads and manually converted 32bit loads (e.g., bindless),
but it was broken for the case where 64bit was not supported, as the offset wasn't
being correctly adjusted
break out the offset division to hopefully make this a little clearer
Fixes: 150d6ee97e3 ("zink: move all 64-32bit shader load rewriting to nir pass")
Reviewed-by: Adam Jackson <ajax at redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16669>
---
src/gallium/drivers/zink/ci/zink-tu-a630-fails.txt | 153 ---------------------
src/gallium/drivers/zink/zink_compiler.c | 29 ++--
2 files changed, 19 insertions(+), 163 deletions(-)
diff --git a/src/gallium/drivers/zink/ci/zink-tu-a630-fails.txt b/src/gallium/drivers/zink/ci/zink-tu-a630-fails.txt
index 0b90047a3fb..1c4ceaba8be 100644
--- a/src/gallium/drivers/zink/ci/zink-tu-a630-fails.txt
+++ b/src/gallium/drivers/zink/ci/zink-tu-a630-fails.txt
@@ -19,160 +19,13 @@ GTF-GL46.gtf40.GL3Tests.transform_feedback3.transform_feedback3_streams_overflow
GTF-GL46.gtf40.GL3Tests.transform_feedback3.transform_feedback3_streams_queried,Fail
KHR-GL46.buffer_storage.map_persistent_draw,Fail
KHR-GL46.compute_shader.fp64-case1,Crash
-KHR-GL46.compute_shader.fp64-case2,Fail
KHR-GL46.compute_shader.fp64-case3,Crash
KHR-GL46.copy_image.functional,Fail
KHR-GL46.direct_state_access.buffers_functional,Fail
KHR-GL46.geometry_shader.api.max_atomic_counters,Fail
-KHR-GL46.gpu_shader_fp64.builtin.abs_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.abs_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.abs_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.ceil_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.ceil_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.ceil_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_against_scalar_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_against_scalar_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_against_scalar_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.clamp_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.cross_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.determinant_dmat2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.determinant_dmat3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.distance_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.distance_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.distance_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.dot_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.dot_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.dot_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.equal_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.equal_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.equal_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.faceforward_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.faceforward_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.faceforward_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.floor_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.floor_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.floor_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fma_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fma_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fma_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fma_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fract_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fract_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.fract_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.frexp_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.frexp_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.frexp_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.greaterthan_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.greaterthan_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.greaterthan_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.greaterthanequal_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.greaterthanequal_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.greaterthanequal_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.inverse_dmat2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.inverse_dmat3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.inversesqrt_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.inversesqrt_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.inversesqrt_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.ldexp_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.ldexp_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.ldexp_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.length_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.length_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.length_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.lessthan_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.lessthan_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.lessthan_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.lessthanequal_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.lessthanequal_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.lessthanequal_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat2x3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat2x4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat3x2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat3x4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat4x2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.matrixcompmult_dmat4x3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_against_scalar_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_against_scalar_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_against_scalar_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.max_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_against_scalar_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_against_scalar_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_against_scalar_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.min_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mix_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mix_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mix_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mix_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mod_against_scalar_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mod_against_scalar_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mod_against_scalar_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.mod_double,Fail
KHR-GL46.gpu_shader_fp64.builtin.mod_dvec2,Fail
KHR-GL46.gpu_shader_fp64.builtin.mod_dvec3,Fail
KHR-GL46.gpu_shader_fp64.builtin.mod_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.modf_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.modf_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.modf_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.normalize_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.normalize_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.normalize_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.notequal_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.notequal_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.notequal_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat2x3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat2x4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat3x2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat3x4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat4x2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.outerproduct_dmat4x3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.reflect_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.reflect_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.reflect_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.refract_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.refract_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.refract_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.round_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.round_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.round_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.roundeven_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.roundeven_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.roundeven_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_against_scalar_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_against_scalar_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_against_scalar_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.smoothstep_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.sqrt_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.sqrt_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.sqrt_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_against_scalar_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_against_scalar_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_against_scalar_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_double,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.step_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.builtin.trunc_dvec2,Fail
-KHR-GL46.gpu_shader_fp64.builtin.trunc_dvec3,Fail
-KHR-GL46.gpu_shader_fp64.builtin.trunc_dvec4,Fail
-KHR-GL46.gpu_shader_fp64.fp64.max_uniform_components,Fail
KHR-GL46.limits.max_fragment_input_components,Fail
KHR-GL46.packed_pixels.pbo_rectangle.r11f_g11f_b10f,Fail
KHR-GL46.packed_pixels.pbo_rectangle.r16,Fail
@@ -235,14 +88,9 @@ KHR-GL46.texture_view.view_classes,Fail
KHR-GL46.vertex_attrib_64bit.vao,Fail
KHR-Single-GL46.arrays_of_arrays_gl.AtomicUsage,Fail
KHR-Single-GL46.arrays_of_arrays_gl.SubroutineFunctionCalls2,Crash
-KHR-Single-GL46.enhanced_layouts.ssb_member_offset_and_align,Fail
-KHR-Single-GL46.enhanced_layouts.uniform_block_member_offset_and_align,Fail
KHR-Single-GL46.enhanced_layouts.varying_structure_locations,Crash
KHR-Single-GL46.enhanced_layouts.xfb_capture_inactive_output_block_member,Fail
KHR-Single-GL46.enhanced_layouts.xfb_capture_struct,Fail
-KHR-Single-GL46.enhanced_layouts.xfb_global_buffer,Fail
-KHR-Single-GL46.enhanced_layouts.xfb_override_qualifiers_with_api,Fail
-KHR-Single-GL46.enhanced_layouts.xfb_stride,Fail
KHR-Single-GL46.enhanced_layouts.xfb_struct_explicit_location,Crash
KHR-Single-GL46.enhanced_layouts.xfb_vertex_streams,Fail
dEQP-GLES3.functional.occlusion_query.depth_clear,Fail
@@ -338,7 +186,6 @@ dEQP-GLES31.functional.texture.border_clamp.formats.depth24_stencil8_sample_dept
dEQP-GLES31.functional.texture.border_clamp.formats.depth24_stencil8_sample_depth.nearest_size_pot,Fail
dEQP-GLES31.functional.texture.border_clamp.formats.depth24_stencil8_sample_stencil.nearest_size_npot,Fail
dEQP-GLES31.functional.texture.border_clamp.formats.depth24_stencil8_sample_stencil.nearest_size_pot,Fail
-dEQP-GLES31.functional.texture.border_clamp.formats.depth32f_stencil8_sample_depth.nearest_size_pot,Fail
dEQP-GLES31.functional.texture.border_clamp.formats.depth32f_stencil8_sample_stencil.nearest_size_npot,Fail
dEQP-GLES31.functional.texture.border_clamp.formats.depth32f_stencil8_sample_stencil.nearest_size_pot,Fail
dEQP-GLES31.functional.texture.border_clamp.formats.depth_component24.gather_size_npot,Fail
diff --git a/src/gallium/drivers/zink/zink_compiler.c b/src/gallium/drivers/zink/zink_compiler.c
index 403e801dca0..ff0ed5ba851 100644
--- a/src/gallium/drivers/zink/zink_compiler.c
+++ b/src/gallium/drivers/zink/zink_compiler.c
@@ -948,10 +948,11 @@ rewrite_bo_access_instr(nir_builder *b, nir_instr *instr, void *data)
nir_src_as_uint(intr->src[0]) == 0 &&
nir_dest_bit_size(intr->dest) == 64 &&
nir_intrinsic_align_offset(intr) % 8 != 0;
- nir_instr_rewrite_src_ssa(instr, &intr->src[1], nir_udiv_imm(b, intr->src[1].ssa,
- (force_2x32 ? 32 : nir_dest_bit_size(intr->dest)) / 8));
+ force_2x32 |= nir_dest_bit_size(intr->dest) == 64 && !has_int64;
+ nir_ssa_def *offset = nir_udiv_imm(b, intr->src[1].ssa, (force_2x32 ? 32 : nir_dest_bit_size(intr->dest)) / 8);
+ nir_instr_rewrite_src_ssa(instr, &intr->src[1], offset);
/* if 64bit isn't supported, 64bit loads definitely aren't supported, so rewrite as 2x32 with cast and pray */
- if (force_2x32 || (nir_dest_bit_size(intr->dest) == 64 && !has_int64)) {
+ if (force_2x32) {
/* this is always scalarized */
assert(intr->dest.ssa.num_components == 1);
/* rewrite as 2x32 */
@@ -972,9 +973,11 @@ rewrite_bo_access_instr(nir_builder *b, nir_instr *instr, void *data)
}
case nir_intrinsic_load_shared:
b->cursor = nir_before_instr(instr);
- nir_instr_rewrite_src_ssa(instr, &intr->src[0], nir_udiv_imm(b, intr->src[0].ssa, nir_dest_bit_size(intr->dest) / 8));
+ bool force_2x32 = nir_dest_bit_size(intr->dest) == 64 && !has_int64;
+ nir_ssa_def *offset = nir_udiv_imm(b, intr->src[0].ssa, (force_2x32 ? 32 : nir_dest_bit_size(intr->dest)) / 8);
+ nir_instr_rewrite_src_ssa(instr, &intr->src[0], offset);
/* if 64bit isn't supported, 64bit loads definitely aren't supported, so rewrite as 2x32 with cast and pray */
- if (nir_dest_bit_size(intr->dest) == 64 && !has_int64) {
+ if (force_2x32) {
/* this is always scalarized */
assert(intr->dest.ssa.num_components == 1);
/* rewrite as 2x32 */
@@ -988,11 +991,13 @@ rewrite_bo_access_instr(nir_builder *b, nir_instr *instr, void *data)
return true;
}
break;
- case nir_intrinsic_store_ssbo:
+ case nir_intrinsic_store_ssbo: {
b->cursor = nir_before_instr(instr);
- nir_instr_rewrite_src_ssa(instr, &intr->src[2], nir_udiv_imm(b, intr->src[2].ssa, nir_src_bit_size(intr->src[0]) / 8));
+ bool force_2x32 = nir_src_bit_size(intr->src[0]) == 64 && !has_int64;
+ nir_ssa_def *offset = nir_udiv_imm(b, intr->src[2].ssa, (force_2x32 ? 32 : nir_src_bit_size(intr->src[0])) / 8);
+ nir_instr_rewrite_src_ssa(instr, &intr->src[2], offset);
/* if 64bit isn't supported, 64bit loads definitely aren't supported, so rewrite as 2x32 with cast and pray */
- if (nir_src_bit_size(intr->src[0]) == 64 && !has_int64) {
+ if (force_2x32) {
/* this is always scalarized */
assert(intr->src[0].ssa->num_components == 1);
nir_ssa_def *vals[2] = {nir_unpack_64_2x32_split_x(b, intr->src[0].ssa), nir_unpack_64_2x32_split_y(b, intr->src[0].ssa)};
@@ -1001,9 +1006,12 @@ rewrite_bo_access_instr(nir_builder *b, nir_instr *instr, void *data)
nir_instr_remove(instr);
}
return true;
- case nir_intrinsic_store_shared:
+ }
+ case nir_intrinsic_store_shared: {
b->cursor = nir_before_instr(instr);
- nir_instr_rewrite_src_ssa(instr, &intr->src[1], nir_udiv_imm(b, intr->src[1].ssa, nir_src_bit_size(intr->src[0]) / 8));
+ bool force_2x32 = nir_src_bit_size(intr->src[0]) == 64 && !has_int64;
+ nir_ssa_def *offset = nir_udiv_imm(b, intr->src[1].ssa, (force_2x32 ? 32 : nir_src_bit_size(intr->src[0])) / 8);
+ nir_instr_rewrite_src_ssa(instr, &intr->src[1], offset);
/* if 64bit isn't supported, 64bit loads definitely aren't supported, so rewrite as 2x32 with cast and pray */
if (nir_src_bit_size(intr->src[0]) == 64 && !has_int64) {
/* this is always scalarized */
@@ -1014,6 +1022,7 @@ rewrite_bo_access_instr(nir_builder *b, nir_instr *instr, void *data)
nir_instr_remove(instr);
}
return true;
+ }
default:
break;
}
More information about the mesa-commit
mailing list