On Fri, Jun 1, 2018 at 3:57 AM, Bas Nieuwenhuizen <[email protected]> wrote:
> On Fri, Jun 1, 2018 at 7:07 AM, Jason Ekstrand <[email protected]> > wrote: > > From: Bas Nieuwenhuizen <[email protected]> > > > > --- > > src/amd/common/ac_nir_to_llvm.c | 404 +++++------------------------- > ---------- > > 1 file changed, 50 insertions(+), 354 deletions(-) > > > > diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_ > llvm.c > > index 60ac8c4..38d6b31 100644 > > --- a/src/amd/common/ac_nir_to_llvm.c > > +++ b/src/amd/common/ac_nir_to_llvm.c > > @@ -54,7 +54,6 @@ struct ac_nir_context { > > }; > > > > static LLVMValueRef get_sampler_desc(struct ac_nir_context *ctx, > > - const nir_deref_var *deref, > > nir_deref_instr *deref_instr, > > enum ac_descriptor_type desc_type, > > const nir_tex_instr *instr, > > @@ -1323,7 +1322,7 @@ static LLVMValueRef build_tex_intrinsic(struct > ac_nir_context *ctx, > > > > if (instr->op == nir_texop_tg4 && ctx->ac.chip_class <= VI) { > > nir_deref_instr *texture_deref_instr = > get_tex_texture_deref(instr); > > - nir_variable *var = texture_deref_instr ? > nir_deref_instr_get_variable(texture_deref_instr) : instr->texture->var; > > + nir_variable *var = nir_deref_instr_get_variable( > texture_deref_instr); > > enum glsl_base_type stype = glsl_get_sampler_result_type( > var->type); > > if (stype == GLSL_TYPE_UINT || stype == GLSL_TYPE_INT) { > > return lower_gather4_integer(&ctx->ac, var, > args, instr); > > @@ -1629,10 +1628,10 @@ static LLVMValueRef visit_load_ubo_buffer(struct > ac_nir_context *ctx, > > } > > > > static void > > -get_deref_instr_offset(struct ac_nir_context *ctx, nir_deref_instr > *instr, > > - bool vs_in, unsigned *vertex_index_out, > > - LLVMValueRef *vertex_index_ref, > > - unsigned *const_out, LLVMValueRef *indir_out) > > +get_deref_offset(struct ac_nir_context *ctx, nir_deref_instr *instr, > > + bool vs_in, unsigned *vertex_index_out, > > + LLVMValueRef *vertex_index_ref, > > + unsigned *const_out, LLVMValueRef *indir_out) > > { > > nir_variable *var = nir_deref_instr_get_variable(instr); > > nir_deref_path path; > > @@ -1697,124 +1696,6 @@ out: > > *indir_out = offset; > > } > > > > -static void > > -get_deref_offset(struct ac_nir_context *ctx, nir_deref_var *deref, > > - bool vs_in, unsigned *vertex_index_out, > > - LLVMValueRef *vertex_index_ref, > > - unsigned *const_out, LLVMValueRef *indir_out) > > -{ > > - unsigned const_offset = 0; > > - nir_deref *tail = &deref->deref; > > - LLVMValueRef offset = NULL; > > - > > - if (vertex_index_out != NULL || vertex_index_ref != NULL) { > > - tail = tail->child; > > - nir_deref_array *deref_array = nir_deref_as_array(tail); > > - if (vertex_index_out) > > - *vertex_index_out = deref_array->base_offset; > > - > > - if (vertex_index_ref) { > > - LLVMValueRef vtx = LLVMConstInt(ctx->ac.i32, > deref_array->base_offset, false); > > - if (deref_array->deref_array_type == > nir_deref_array_type_indirect) { > > - vtx = LLVMBuildAdd(ctx->ac.builder, vtx, > get_src(ctx, deref_array->indirect), ""); > > - } > > - *vertex_index_ref = vtx; > > - } > > - } > > - > > - if (deref->var->data.compact) { > > - assert(tail->child->deref_type == nir_deref_type_array); > > - assert(glsl_type_is_scalar( > glsl_without_array(deref->var->type))); > > - nir_deref_array *deref_array = nir_deref_as_array(tail-> > child); > > - /* We always lower indirect dereferences for "compact" > array vars. */ > > - assert(deref_array->deref_array_type == > nir_deref_array_type_direct); > > - > > - const_offset = deref_array->base_offset; > > - goto out; > > - } > > - > > - while (tail->child != NULL) { > > - const struct glsl_type *parent_type = tail->type; > > - tail = tail->child; > > - > > - if (tail->deref_type == nir_deref_type_array) { > > - nir_deref_array *deref_array = > nir_deref_as_array(tail); > > - LLVMValueRef index, stride, local_offset; > > - unsigned size = > > glsl_count_attribute_slots(tail->type, > vs_in); > > - > > - const_offset += size * deref_array->base_offset; > > - if (deref_array->deref_array_type == > nir_deref_array_type_direct) > > - continue; > > - > > - assert(deref_array->deref_array_type == > nir_deref_array_type_indirect); > > - index = get_src(ctx, deref_array->indirect); > > - stride = LLVMConstInt(ctx->ac.i32, size, 0); > > - local_offset = LLVMBuildMul(ctx->ac.builder, > stride, index, ""); > > - > > - if (offset) > > - offset = LLVMBuildAdd(ctx->ac.builder, > offset, local_offset, ""); > > - else > > - offset = local_offset; > > - } else if (tail->deref_type == nir_deref_type_struct) { > > - nir_deref_struct *deref_struct = > nir_deref_as_struct(tail); > > - > > - for (unsigned i = 0; i < deref_struct->index; > i++) { > > - const struct glsl_type *ft = > glsl_get_struct_field(parent_type, i); > > - const_offset += > glsl_count_attribute_slots(ft, vs_in); > > - } > > - } else > > - unreachable("unsupported deref type"); > > - > > - } > > -out: > > - if (const_offset && offset) > > - offset = LLVMBuildAdd(ctx->ac.builder, offset, > > - LLVMConstInt(ctx->ac.i32, > const_offset, 0), > > - ""); > > - > > - *const_out = const_offset; > > - *indir_out = offset; > > -} > > - > > -static LLVMValueRef > > -build_gep_for_deref(struct ac_nir_context *ctx, > > - nir_deref_var *deref) > > -{ > > - struct hash_entry *entry = _mesa_hash_table_search(ctx->vars, > deref->var); > > - assert(entry->data); > > - LLVMValueRef val = entry->data; > > - nir_deref *tail = deref->deref.child; > > - while (tail != NULL) { > > - LLVMValueRef offset; > > - switch (tail->deref_type) { > > - case nir_deref_type_array: { > > - nir_deref_array *array = > nir_deref_as_array(tail); > > - offset = LLVMConstInt(ctx->ac.i32, > array->base_offset, 0); > > - if (array->deref_array_type == > > - nir_deref_array_type_indirect) { > > - offset = LLVMBuildAdd(ctx->ac.builder, > offset, > > - get_src(ctx, > > - > array->indirect), > > - ""); > > - } > > - break; > > - } > > - case nir_deref_type_struct: { > > - nir_deref_struct *deref_struct = > > - nir_deref_as_struct(tail); > > - offset = LLVMConstInt(ctx->ac.i32, > > - deref_struct->index, 0); > > - break; > > - } > > - default: > > - unreachable("bad deref type"); > > - } > > - val = ac_build_gep0(&ctx->ac, val, offset); > > - tail = tail->child; > > - } > > - return val; > > -} > > - > > static LLVMValueRef load_tess_varyings(struct ac_nir_context *ctx, > > nir_intrinsic_instr *instr, > > bool load_inputs) > > @@ -1824,24 +1705,16 @@ static LLVMValueRef load_tess_varyings(struct > ac_nir_context *ctx, > > LLVMValueRef indir_index = NULL; > > unsigned const_index = 0; > > > > - bool uses_deref_chain = instr->intrinsic == > nir_intrinsic_load_var; > > - nir_variable *var = uses_deref_chain ? instr->variables[0]->var : > > - nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > + nir_variable *var = nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > > > unsigned location = var->data.location; > > unsigned driver_location = var->data.driver_location; > > const bool is_patch = var->data.patch; > > const bool is_compact = var->data.compact; > > > > - if (uses_deref_chain) { > > - get_deref_offset(ctx, instr->variables[0], > > - false, NULL, is_patch ? NULL : > &vertex_index, > > - &const_index, &indir_index); > > - } else { > > - get_deref_instr_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > > - false, NULL, is_patch ? NULL : > &vertex_index, > > - &const_index, &indir_index); > > - } > > + get_deref_offset(ctx, nir_instr_as_deref(instr->src[ > 0].ssa->parent_instr), > > + false, NULL, is_patch ? NULL : &vertex_index, > > + &const_index, &indir_index); > > > > LLVMTypeRef dest_type = get_def_type(ctx, &instr->dest.ssa); > > > > @@ -1863,9 +1736,7 @@ static LLVMValueRef load_tess_varyings(struct > ac_nir_context *ctx, > > static LLVMValueRef visit_load_var(struct ac_nir_context *ctx, > > nir_intrinsic_instr *instr) > > { > > - bool uses_deref_chain = instr->intrinsic == > nir_intrinsic_load_var; > > - nir_variable *var = uses_deref_chain ? instr->variables[0]->var : > > - nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > + nir_variable *var = nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > > > LLVMValueRef values[8]; > > int idx = var->data.driver_location; > > @@ -1878,13 +1749,8 @@ static LLVMValueRef visit_load_var(struct > ac_nir_context *ctx, > > bool vs_in = ctx->stage == MESA_SHADER_VERTEX && > > var->data.mode == nir_var_shader_in; > > > > - if (uses_deref_chain) { > > - get_deref_offset(ctx, instr->variables[0], vs_in, NULL, > NULL, > > - &const_index, &indir_index); > > - } else { > > - get_deref_instr_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), vs_in, NULL, NULL, > > - &const_index, &indir_index); > > - } > > + get_deref_offset(ctx, > > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > vs_in, NULL, NULL, > > + &const_index, &indir_index); > > > > if (instr->dest.ssa.bit_size == 64) > > ve *= 2; > > @@ -1900,14 +1766,8 @@ static LLVMValueRef visit_load_var(struct > ac_nir_context *ctx, > > LLVMTypeRef type = > > LLVMIntTypeInContext(ctx->ac.context, > instr->dest.ssa.bit_size); > > LLVMValueRef indir_index; > > unsigned const_index, vertex_index; > > - if (uses_deref_chain) { > > - get_deref_offset(ctx, > instr->variables[0], > > - false, &vertex_index, > NULL, > > - &const_index, > &indir_index); > > - } else { > > - get_deref_instr_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > > - false, > &vertex_index, NULL, &const_index, &indir_index); > > - } > > + get_deref_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > > + false, &vertex_index, NULL, > &const_index, &indir_index); > > > > return ctx->abi->load_inputs(ctx->abi, > var->data.location, > > > var->data.driver_location, > > @@ -1951,9 +1811,7 @@ static LLVMValueRef visit_load_var(struct > ac_nir_context *ctx, > > } > > break; > > case nir_var_shared: { > > - LLVMValueRef address = uses_deref_chain ? > > - build_gep_for_deref(ctx, > instr->variables[0]) > > - : get_src(ctx, instr->src[0]); > > + LLVMValueRef address = get_src(ctx, instr->src[0]); > > LLVMValueRef val = LLVMBuildLoad(ctx->ac.builder, > address, ""); > > return LLVMBuildBitCast(ctx->ac.builder, val, > > get_def_type(ctx, > &instr->dest.ssa), > > @@ -1994,25 +1852,18 @@ static void > > visit_store_var(struct ac_nir_context *ctx, > > nir_intrinsic_instr *instr) > > { > > - bool uses_deref_chain = instr->intrinsic == > nir_intrinsic_store_var; > > - nir_variable *var = uses_deref_chain ? instr->variables[0]->var > : > > - nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > + nir_variable *var = nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > > > LLVMValueRef temp_ptr, value; > > int idx = var->data.driver_location; > > unsigned comp = var->data.location_frac; > > - LLVMValueRef src = ac_to_float(&ctx->ac, get_src(ctx, > instr->src[uses_deref_chain ? 0 : 1])); > > + LLVMValueRef src = ac_to_float(&ctx->ac, get_src(ctx, > instr->src[1])); > > int writemask = instr->const_index[0]; > > LLVMValueRef indir_index; > > unsigned const_index; > > > > - if (uses_deref_chain) { > > - get_deref_offset(ctx, instr->variables[0], false, > > - NULL, NULL, &const_index, &indir_index); > > - } else { > > - get_deref_instr_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), false, > > - NULL, NULL, &const_index, > &indir_index); > > - } > > + get_deref_offset(ctx, > > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > false, > > + NULL, NULL, &const_index, &indir_index); > > > > if (ac_get_elem_bits(&ctx->ac, LLVMTypeOf(src)) == 64) { > > > > @@ -2034,15 +1885,9 @@ visit_store_var(struct ac_nir_context *ctx, > > unsigned const_index = 0; > > const bool is_patch = var->data.patch; > > > > - if (uses_deref_chain) { > > - get_deref_offset(ctx, > instr->variables[0], > > - false, NULL, is_patch ? > NULL : &vertex_index, > > - &const_index, > &indir_index); > > - } else { > > - get_deref_instr_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > > - false, NULL, is_patch ? > NULL : &vertex_index, > > - &const_index, > &indir_index); > > - } > > + get_deref_offset(ctx, > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > > + false, NULL, is_patch ? NULL : > &vertex_index, > > + &const_index, &indir_index); > > > > ctx->abi->store_tcs_outputs(ctx->abi, var, > > vertex_index, > indir_index, > > @@ -2106,10 +1951,8 @@ visit_store_var(struct ac_nir_context *ctx, > > break; > > case nir_var_shared: { > > int writemask = instr->const_index[0]; > > - LLVMValueRef address = uses_deref_chain ? > > - build_gep_for_deref(ctx, > instr->variables[0]) > > - : get_src(ctx, instr->src[0]); > > - LLVMValueRef val = get_src(ctx, > instr->src[uses_deref_chain ? 0 : 1]); > > + LLVMValueRef address = get_src(ctx, instr->src[0]); > > + LLVMValueRef val = get_src(ctx, instr->src[1]); > > if (util_is_power_of_two_nonzero(writemask)) { > > val = LLVMBuildBitCast( > > ctx->ac.builder, val, > > @@ -2236,32 +2079,8 @@ static LLVMValueRef > > adjust_sample_index_using_fmask(struct > ac_llvm_context *ctx, > > return sample_index; > > } > > > > -static bool is_var_image_intrinsic(const nir_intrinsic_instr *instr) > > -{ > > - switch(instr->intrinsic) { > > - case nir_intrinsic_image_var_samples: > > - case nir_intrinsic_image_var_load: > > - case nir_intrinsic_image_var_store: > > - case nir_intrinsic_image_var_atomic_add: > > - case nir_intrinsic_image_var_atomic_min: > > - case nir_intrinsic_image_var_atomic_max: > > - case nir_intrinsic_image_var_atomic_and: > > - case nir_intrinsic_image_var_atomic_or: > > - case nir_intrinsic_image_var_atomic_xor: > > - case nir_intrinsic_image_var_atomic_exchange: > > - case nir_intrinsic_image_var_atomic_comp_swap: > > - case nir_intrinsic_image_var_size: > > - return true; > > - default: > > - return false; > > - } > > -} > > - > > static nir_variable *get_image_variable(const nir_intrinsic_instr > *instr) > > { > > - if (is_var_image_intrinsic(instr)) > > - return instr->variables[0]->var; > > - > > assert(instr->src[0].is_ssa); > > return nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > } > > @@ -2271,26 +2090,21 @@ static LLVMValueRef get_image_descriptor(struct > ac_nir_context *ctx, > > enum ac_descriptor_type > desc_type, > > bool write) > > { > > - if (is_var_image_intrinsic(instr)) > > - return get_sampler_desc(ctx, instr->variables[0], NULL, > desc_type, NULL, true, true); > > - > > - return get_sampler_desc(ctx, NULL, > > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > desc_type, NULL, true, true); > > + return get_sampler_desc(ctx, > > nir_instr_as_deref(instr->src[0].ssa->parent_instr), > desc_type, NULL, true, true); > > } > > > > static void get_image_coords(struct ac_nir_context *ctx, > > const nir_intrinsic_instr *instr, > > struct ac_image_args *args) > > { > > - /* As the deref instrinsics have the deref as src 0, everything > is shifted. */ > > - int src_shift = is_var_image_intrinsic(instr) ? 0 : 1; > > const struct glsl_type *type = glsl_without_array(get_image_ > variable(instr)->type); > > > > - LLVMValueRef src0 = get_src(ctx, instr->src[src_shift]); > > + LLVMValueRef src0 = get_src(ctx, instr->src[1]); > > LLVMValueRef masks[] = { > > LLVMConstInt(ctx->ac.i32, 0, false), > LLVMConstInt(ctx->ac.i32, 1, false), > > LLVMConstInt(ctx->ac.i32, 2, false), > LLVMConstInt(ctx->ac.i32, 3, false), > > }; > > - LLVMValueRef sample_index = ac_llvm_extract_elem(&ctx->ac, > get_src(ctx, instr->src[src_shift + 1]), 0); > > + LLVMValueRef sample_index = ac_llvm_extract_elem(&ctx->ac, > get_src(ctx, instr->src[2]), 0); > > > > int count; > > enum glsl_sampler_dim dim = glsl_get_sampler_dim(type); > > @@ -2328,7 +2142,7 @@ static void get_image_coords(struct ac_nir_context > *ctx, > > > get_image_descriptor(ctx, instr, AC_DESC_FMASK, false)); > > } > > if (count == 1 && !gfx9_1d) { > > - if (instr->src[src_shift].ssa->num_components) > > + if (instr->src[1].ssa->num_components) > > args->coords[0] = > > LLVMBuildExtractElement(ctx->ac.builder, > src0, masks[0], ""); > > else > > args->coords[0] = src0; > > @@ -2398,15 +2212,12 @@ static LLVMValueRef visit_image_load(struct > ac_nir_context *ctx, > > > > const enum glsl_sampler_dim dim = glsl_get_sampler_dim(type); > > if (dim == GLSL_SAMPLER_DIM_BUF) { > > - /* As the deref instrinsics have the deref as src 0, > everything is shifted. */ > > - int src_shift = is_var_image_intrinsic(instr) ? 0 : 1; > > - > > unsigned mask = nir_ssa_def_components_read(& > instr->dest.ssa); > > unsigned num_channels = util_last_bit(mask); > > LLVMValueRef rsrc, vindex; > > > > rsrc = get_image_buffer_descriptor(ctx, instr, false); > > - vindex = LLVMBuildExtractElement(ctx->ac.builder, > get_src(ctx, instr->src[src_shift]), > > + vindex = LLVMBuildExtractElement(ctx->ac.builder, > get_src(ctx, instr->src[1]), > > ctx->ac.i32_0, ""); > > > > /* TODO: set "glc" and "can_speculate" when OpenGL needs > it. */ > > @@ -2446,15 +2257,12 @@ static void visit_image_store(struct > ac_nir_context *ctx, > > if (force_glc) > > glc = ctx->ac.i1true; > > > > - /* As the deref instrinsics have the deref as src 0, everything > is shifted. */ > > - int src_shift = is_var_image_intrinsic(instr) ? 0 : 1; > > - > > if (dim == GLSL_SAMPLER_DIM_BUF) { > > LLVMValueRef rsrc = get_image_buffer_descriptor(ctx, > instr, true); > > > > - params[0] = ac_to_float(&ctx->ac, get_src(ctx, > instr->src[src_shift + 2])); /* data */ > > + params[0] = ac_to_float(&ctx->ac, get_src(ctx, > instr->src[3])); /* data */ > > params[1] = rsrc; > > - params[2] = LLVMBuildExtractElement(ctx->ac.builder, > get_src(ctx, instr->src[src_shift]), > > + params[2] = LLVMBuildExtractElement(ctx->ac.builder, > get_src(ctx, instr->src[1]), > > ctx->ac.i32_0, ""); > /* vindex */ > > params[3] = ctx->ac.i32_0; /* voffset */ > > params[4] = glc; /* glc */ > > @@ -2464,9 +2272,9 @@ static void visit_image_store(struct > ac_nir_context *ctx, > > } else { > > struct ac_image_args args = {}; > > args.opcode = ac_image_store; > > - args.data[0] = ac_to_float(&ctx->ac, get_src(ctx, > instr->src[2])); > > + args.data[0] = ac_to_float(&ctx->ac, get_src(ctx, > instr->src[3])); > > get_image_coords(ctx, instr, &args); > > - args.resource = get_image_descriptor(ctx, instr, > AC_DESC_IMAGE, true);; > > + args.resource = get_image_descriptor(ctx, instr, > AC_DESC_IMAGE, true); > > args.dim = get_ac_image_dim(&ctx->ac, > glsl_get_sampler_dim(type), > > glsl_sampler_type_is_array( > type)); > > args.dmask = 15; > > @@ -2485,8 +2293,7 @@ static LLVMValueRef visit_image_atomic(struct > ac_nir_context *ctx, > > int param_count = 0; > > const nir_variable *var = get_image_variable(instr); > > > > - bool cmpswap = instr->intrinsic == > > nir_intrinsic_image_var_atomic_comp_swap > || > > - instr->intrinsic == nir_intrinsic_image_deref_ > atomic_comp_swap; > > + bool cmpswap = instr->intrinsic == nir_intrinsic_image_deref_ > atomic_comp_swap; > > const char *atomic_name; > > char intrinsic_name[41]; > > enum ac_atomic_op atomic_subop; > > @@ -2496,42 +2303,34 @@ static LLVMValueRef visit_image_atomic(struct > ac_nir_context *ctx, > > bool is_unsigned = glsl_get_sampler_result_type(type) == > GLSL_TYPE_UINT; > > > > switch (instr->intrinsic) { > > - case nir_intrinsic_image_var_atomic_add: > > case nir_intrinsic_image_deref_atomic_add: > > atomic_name = "add"; > > atomic_subop = ac_atomic_add; > > break; > > - case nir_intrinsic_image_var_atomic_min: > > case nir_intrinsic_image_deref_atomic_min: > > atomic_name = is_unsigned ? "umin" : "smin"; > > atomic_subop = is_unsigned ? ac_atomic_umin : > ac_atomic_smin; > > break; > > - case nir_intrinsic_image_var_atomic_max: > > case nir_intrinsic_image_deref_atomic_max: > > atomic_name = is_unsigned ? "umax" : "smax"; > > atomic_subop = is_unsigned ? ac_atomic_umax : > ac_atomic_smax; > > break; > > - case nir_intrinsic_image_var_atomic_and: > > case nir_intrinsic_image_deref_atomic_and: > > atomic_name = "and"; > > atomic_subop = ac_atomic_and; > > break; > > - case nir_intrinsic_image_var_atomic_or: > > case nir_intrinsic_image_deref_atomic_or: > > atomic_name = "or"; > > atomic_subop = ac_atomic_or; > > break; > > - case nir_intrinsic_image_var_atomic_xor: > > case nir_intrinsic_image_deref_atomic_xor: > > atomic_name = "xor"; > > atomic_subop = ac_atomic_xor; > > break; > > - case nir_intrinsic_image_var_atomic_exchange: > > case nir_intrinsic_image_deref_atomic_exchange: > > atomic_name = "swap"; > > atomic_subop = ac_atomic_swap; > > break; > > - case nir_intrinsic_image_var_atomic_comp_swap: > > case nir_intrinsic_image_deref_atomic_comp_swap: > > atomic_name = "cmpswap"; > > atomic_subop = 0; /* not used */ > > @@ -2540,16 +2339,13 @@ static LLVMValueRef visit_image_atomic(struct > ac_nir_context *ctx, > > abort(); > > } > > > > - /* As the deref instrinsics have the deref as src 0, everything > is shifted. */ > > - int src_shift = is_var_image_intrinsic(instr) ? 0 : 1; > > - > > if (cmpswap) > > - params[param_count++] = get_src(ctx, > instr->src[src_shift + 3]); > > - params[param_count++] = get_src(ctx, instr->src[src_shift + 2]); > > + params[param_count++] = get_src(ctx, instr->src[4]); > > + params[param_count++] = get_src(ctx, instr->src[3]); > > > > if (glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_BUF) { > > params[param_count++] = get_image_buffer_descriptor(ctx, > instr, true); > > - params[param_count++] = > > LLVMBuildExtractElement(ctx->ac.builder, > get_src(ctx, instr->src[src_shift]), > > + params[param_count++] = > > LLVMBuildExtractElement(ctx->ac.builder, > get_src(ctx, instr->src[1]), > > > ctx->ac.i32_0, ""); /* vindex */ > > params[param_count++] = ctx->ac.i32_0; /* voffset */ > > params[param_count++] = ctx->ac.i1false; /* slc */ > > @@ -2805,8 +2601,7 @@ static LLVMValueRef visit_var_atomic(struct > ac_nir_context *ctx, > > LLVMValueRef result; > > LLVMValueRef src = get_src(ctx, instr->src[src_idx]); > > > > - if (instr->intrinsic == nir_intrinsic_var_atomic_comp_swap || > > - instr->intrinsic == nir_intrinsic_shared_atomic_comp_swap || > > + if (instr->intrinsic == nir_intrinsic_shared_atomic_comp_swap || > > instr->intrinsic == nir_intrinsic_deref_atomic_comp_swap) { > > LLVMValueRef src1 = get_src(ctx, instr->src[src_idx + > 1]); > > result = LLVMBuildAtomicCmpXchg(ctx->ac.builder, > > @@ -2818,47 +2613,38 @@ static LLVMValueRef visit_var_atomic(struct > ac_nir_context *ctx, > > } else { > > LLVMAtomicRMWBinOp op; > > switch (instr->intrinsic) { > > - case nir_intrinsic_var_atomic_add: > > case nir_intrinsic_shared_atomic_add: > > case nir_intrinsic_deref_atomic_add: > > op = LLVMAtomicRMWBinOpAdd; > > break; > > - case nir_intrinsic_var_atomic_umin: > > case nir_intrinsic_shared_atomic_umin: > > case nir_intrinsic_deref_atomic_umin: > > op = LLVMAtomicRMWBinOpUMin; > > break; > > - case nir_intrinsic_var_atomic_umax: > > case nir_intrinsic_shared_atomic_umax: > > case nir_intrinsic_deref_atomic_umax: > > op = LLVMAtomicRMWBinOpUMax; > > break; > > - case nir_intrinsic_var_atomic_imin: > > case nir_intrinsic_shared_atomic_imin: > > case nir_intrinsic_deref_atomic_imin: > > op = LLVMAtomicRMWBinOpMin; > > break; > > - case nir_intrinsic_var_atomic_imax: > > case nir_intrinsic_shared_atomic_imax: > > case nir_intrinsic_deref_atomic_imax: > > op = LLVMAtomicRMWBinOpMax; > > break; > > - case nir_intrinsic_var_atomic_and: > > case nir_intrinsic_shared_atomic_and: > > case nir_intrinsic_deref_atomic_and: > > op = LLVMAtomicRMWBinOpAnd; > > break; > > - case nir_intrinsic_var_atomic_or: > > case nir_intrinsic_shared_atomic_or: > > case nir_intrinsic_deref_atomic_or: > > op = LLVMAtomicRMWBinOpOr; > > break; > > - case nir_intrinsic_var_atomic_xor: > > case nir_intrinsic_shared_atomic_xor: > > case nir_intrinsic_deref_atomic_xor: > > op = LLVMAtomicRMWBinOpXor; > > break; > > - case nir_intrinsic_var_atomic_exchange: > > case nir_intrinsic_shared_atomic_exchange: > > case nir_intrinsic_deref_atomic_exchange: > > op = LLVMAtomicRMWBinOpXchg; > > @@ -2897,42 +2683,26 @@ static LLVMValueRef visit_interp(struct > ac_nir_context *ctx, > > LLVMValueRef src_c0 = NULL; > > LLVMValueRef src_c1 = NULL; > > LLVMValueRef src0 = NULL; > > - bool uses_deref_chain; > > - switch (instr->intrinsic) { > > - case nir_intrinsic_interp_var_at_centroid: > > - case nir_intrinsic_interp_var_at_sample: > > - case nir_intrinsic_interp_var_at_offset: > > - uses_deref_chain = true; > > - break; > > - default: > > - uses_deref_chain = false; > > - break; > > - } > > > > - nir_variable *var = uses_deref_chain ? instr->variables[0]->var > : nir_deref_instr_get_variable(nir_instr_as_deref(instr->src[ > 0].ssa->parent_instr)); > > + nir_variable *var = nir_deref_instr_get_variable( > nir_instr_as_deref(instr->src[0].ssa->parent_instr)); > > int input_index = var->data.location - VARYING_SLOT_VAR0; > > switch (instr->intrinsic) { > > - case nir_intrinsic_interp_var_at_centroid: > > case nir_intrinsic_interp_deref_at_centroid: > > location = INTERP_CENTROID; > > break; > > - case nir_intrinsic_interp_var_at_sample: > > - case nir_intrinsic_interp_var_at_offset: > > case nir_intrinsic_interp_deref_at_sample: > > case nir_intrinsic_interp_deref_at_offset: > > location = INTERP_CENTER; > > - src0 = get_src(ctx, instr->src[uses_deref_chain ? 0 : > 1]); > > + src0 = get_src(ctx, instr->src[11]); > > Stupid mistake here, 11 -> 1 > > Not sure why I did not catch this before :( > Fixed. I also force-pushed the branch with the change. --Jason > > break; > > default: > > break; > > } > > > > - if (instr->intrinsic == nir_intrinsic_interp_var_at_offset || > > - instr->intrinsic == nir_intrinsic_interp_deref_at_offset) { > > + if (instr->intrinsic == nir_intrinsic_interp_deref_at_offset) { > > src_c0 = ac_to_float(&ctx->ac, > LLVMBuildExtractElement(ctx->ac.builder, src0, ctx->ac.i32_0, "")); > > src_c1 = ac_to_float(&ctx->ac, > LLVMBuildExtractElement(ctx->ac.builder, src0, ctx->ac.i32_1, "")); > > - } else if (instr->intrinsic == nir_intrinsic_interp_var_at_sample > || > > - instr->intrinsic == nir_intrinsic_interp_deref_at_sample) > { > > + } else if (instr->intrinsic == nir_intrinsic_interp_deref_at_sample) > { > > LLVMValueRef sample_position; > > LLVMValueRef halfval = LLVMConstReal(ctx->ac.f32, 0.5f); > > > > @@ -3168,11 +2938,9 @@ static void visit_intrinsic(struct ac_nir_context > *ctx, > > result = visit_get_buffer_size(ctx, instr); > > break; > > case nir_intrinsic_load_deref: > > - case nir_intrinsic_load_var: > > result = visit_load_var(ctx, instr); > > break; > > case nir_intrinsic_store_deref: > > - case nir_intrinsic_store_var: > > visit_store_var(ctx, instr); > > break; > > case nir_intrinsic_load_shared: > > @@ -3181,26 +2949,15 @@ static void visit_intrinsic(struct > ac_nir_context *ctx, > > case nir_intrinsic_store_shared: > > visit_store_shared(ctx, instr); > > break; > > - case nir_intrinsic_image_var_samples: > > case nir_intrinsic_image_deref_samples: > > result = visit_image_samples(ctx, instr); > > break; > > - case nir_intrinsic_image_var_load: > > case nir_intrinsic_image_deref_load: > > result = visit_image_load(ctx, instr); > > break; > > - case nir_intrinsic_image_var_store: > > case nir_intrinsic_image_deref_store: > > visit_image_store(ctx, instr); > > break; > > - case nir_intrinsic_image_var_atomic_add: > > - case nir_intrinsic_image_var_atomic_min: > > - case nir_intrinsic_image_var_atomic_max: > > - case nir_intrinsic_image_var_atomic_and: > > - case nir_intrinsic_image_var_atomic_or: > > - case nir_intrinsic_image_var_atomic_xor: > > - case nir_intrinsic_image_var_atomic_exchange: > > - case nir_intrinsic_image_var_atomic_comp_swap: > > case nir_intrinsic_image_deref_atomic_add: > > case nir_intrinsic_image_deref_atomic_min: > > case nir_intrinsic_image_deref_atomic_max: > > @@ -3211,7 +2968,6 @@ static void visit_intrinsic(struct ac_nir_context > *ctx, > > case nir_intrinsic_image_deref_atomic_comp_swap: > > result = visit_image_atomic(ctx, instr); > > break; > > - case nir_intrinsic_image_var_size: > > case nir_intrinsic_image_deref_size: > > result = visit_image_size(ctx, instr); > > break; > > @@ -3247,20 +3003,6 @@ static void visit_intrinsic(struct ac_nir_context > *ctx, > > result = visit_var_atomic(ctx, instr, ptr, 1); > > break; > > } > > - case nir_intrinsic_var_atomic_add: > > - case nir_intrinsic_var_atomic_imin: > > - case nir_intrinsic_var_atomic_umin: > > - case nir_intrinsic_var_atomic_imax: > > - case nir_intrinsic_var_atomic_umax: > > - case nir_intrinsic_var_atomic_and: > > - case nir_intrinsic_var_atomic_or: > > - case nir_intrinsic_var_atomic_xor: > > - case nir_intrinsic_var_atomic_exchange: > > - case nir_intrinsic_var_atomic_comp_swap: { > > - LLVMValueRef ptr = build_gep_for_deref(ctx, > instr->variables[0]); > > - result = visit_var_atomic(ctx, instr, ptr, 0); > > - break; > > - } > > case nir_intrinsic_deref_atomic_add: > > case nir_intrinsic_deref_atomic_imin: > > case nir_intrinsic_deref_atomic_umin: > > @@ -3275,9 +3017,6 @@ static void visit_intrinsic(struct ac_nir_context > *ctx, > > result = visit_var_atomic(ctx, instr, ptr, 1); > > break; > > } > > - case nir_intrinsic_interp_var_at_centroid: > > - case nir_intrinsic_interp_var_at_sample: > > - case nir_intrinsic_interp_var_at_offset: > > case nir_intrinsic_interp_deref_at_centroid: > > case nir_intrinsic_interp_deref_at_sample: > > case nir_intrinsic_interp_deref_at_offset: > > @@ -3358,7 +3097,6 @@ static void visit_intrinsic(struct ac_nir_context > *ctx, > > } > > > > static LLVMValueRef get_sampler_desc(struct ac_nir_context *ctx, > > - const nir_deref_var *deref, > > nir_deref_instr *deref_instr, > > enum ac_descriptor_type desc_type, > > const nir_tex_instr *tex_instr, > > @@ -3370,11 +3108,11 @@ static LLVMValueRef get_sampler_desc(struct > ac_nir_context *ctx, > > unsigned base_index; > > bool bindless = false; > > > > - if (!deref && !deref_instr) { > > + if (!deref_instr) { > > assert(tex_instr && !image); > > descriptor_set = 0; > > base_index = tex_instr->sampler_index; > > - } else if(deref_instr) { > > + } else { > > while(deref_instr->deref_type != nir_deref_type_var) { > > unsigned array_size = > glsl_get_aoa_size(deref_instr->type); > > if (!array_size) > > @@ -3400,41 +3138,6 @@ static LLVMValueRef get_sampler_desc(struct > ac_nir_context *ctx, > > } > > descriptor_set = deref_instr->var->data.descriptor_set; > > base_index = deref_instr->var->data.binding; > > - } else { > > - const nir_deref *tail = &deref->deref; > > - while (tail->child) { > > - const nir_deref_array *child = > nir_deref_as_array(tail->child); > > - unsigned array_size = > glsl_get_aoa_size(tail->child->type); > > - > > - if (!array_size) > > - array_size = 1; > > - > > - assert(child->deref_array_type != > nir_deref_array_type_wildcard); > > - > > - if (child->deref_array_type == > nir_deref_array_type_indirect) { > > - LLVMValueRef indirect = get_src(ctx, > child->indirect); > > - > > - indirect = LLVMBuildMul(ctx->ac.builder, > indirect, > > - LLVMConstInt(ctx->ac.i32, > array_size, false), ""); > > - > > - if (!index) > > - index = indirect; > > - else > > - index = > LLVMBuildAdd(ctx->ac.builder, index, indirect, ""); > > - } > > - > > - constant_index += child->base_offset * > array_size; > > - > > - tail = &child->deref; > > - } > > - descriptor_set = deref->var->data.descriptor_set; > > - > > - if (deref->var->data.bindless) { > > - bindless = deref->var->data.bindless; > > - base_index = deref->var->data.driver_location; > > - } else { > > - base_index = deref->var->data.binding; > > - } > > } > > > > return ctx->abi->load_sampler_desc(ctx->abi, > > @@ -3480,8 +3183,6 @@ static void tex_fetch_ptrs(struct ac_nir_context > *ctx, > > { > > nir_deref_instr *texture_deref_instr = NULL; > > nir_deref_instr *sampler_deref_instr = NULL; > > - nir_deref_var *texture_deref_var = NULL; > > - nir_deref_var *sampler_deref_var = NULL; > > > > for (unsigned i = 0; i < instr->num_srcs; i++) { > > switch (instr->src[i].src_type) { > > @@ -3499,23 +3200,18 @@ static void tex_fetch_ptrs(struct ac_nir_context > *ctx, > > if (!sampler_deref_instr) > > sampler_deref_instr = texture_deref_instr; > > > > - if (!texture_deref_instr) { > > - texture_deref_var = instr->texture; > > - sampler_deref_var = instr->sampler ? instr->sampler : > instr->texture; > > - } > > - > > if (instr->sampler_dim == GLSL_SAMPLER_DIM_BUF) > > - *res_ptr = get_sampler_desc(ctx, texture_deref_var, > texture_deref_instr, AC_DESC_BUFFER, instr, false, false); > > + *res_ptr = get_sampler_desc(ctx, texture_deref_instr, > AC_DESC_BUFFER, instr, false, false); > > else > > - *res_ptr = get_sampler_desc(ctx, texture_deref_var, > texture_deref_instr, AC_DESC_IMAGE, instr, false, false); > > + *res_ptr = get_sampler_desc(ctx, texture_deref_instr, > AC_DESC_IMAGE, instr, false, false); > > if (samp_ptr) { > > - *samp_ptr = get_sampler_desc(ctx, sampler_deref_var, > sampler_deref_instr, AC_DESC_SAMPLER, instr, false, false); > > + *samp_ptr = get_sampler_desc(ctx, sampler_deref_instr, > AC_DESC_SAMPLER, instr, false, false); > > if (instr->sampler_dim < GLSL_SAMPLER_DIM_RECT) > > *samp_ptr = sici_fix_sampler_aniso(ctx, > *res_ptr, *samp_ptr); > > } > > - if (fmask_ptr && !instr->sampler && (instr->op == > nir_texop_txf_ms || > > - instr->op == > nir_texop_samples_identical)) > > - *fmask_ptr = get_sampler_desc(ctx, instr->texture, > texture_deref_instr, AC_DESC_FMASK, instr, false, false); > > + if (fmask_ptr && (instr->op == nir_texop_txf_ms || > > + instr->op == nir_texop_samples_identical)) > > + *fmask_ptr = get_sampler_desc(ctx, texture_deref_instr, > AC_DESC_FMASK, instr, false, false); > > } > > > > static LLVMValueRef apply_round_slice(struct ac_llvm_context *ctx, > > -- > > 2.5.0.400.gff86faf > > >
_______________________________________________ mesa-dev mailing list [email protected] https://lists.freedesktop.org/mailman/listinfo/mesa-dev
