[Mesa-dev] [PATCH] radv: move to using nir clip/cull merge pass.
Edward O'Callaghan
funfunctor at folklore1984.net
Wed Mar 29 14:09:33 UTC 2017
Reviewed-by: Edward O'Callaghan <funfunctor at folklore1984.net>
On 03/29/2017 04:14 PM, Dave Airlie wrote:
> From: Dave Airlie <airlied at redhat.com>
>
> Doing this before tessellation makes doing some bits of
> tessellation a bit cleaner. It also cleans up a bit of the
> llvm generator code.
>
> Signed-off-by: Dave Airlie <airlied at redhat.com>
> ---
> src/amd/common/ac_nir_to_llvm.c | 144 ++++++++++------------------------------
> src/amd/vulkan/radv_pipeline.c | 1 +
> 2 files changed, 36 insertions(+), 109 deletions(-)
>
> diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
> index f164d8f..78602fd 100644
> --- a/src/amd/common/ac_nir_to_llvm.c
> +++ b/src/amd/common/ac_nir_to_llvm.c
> @@ -144,8 +144,6 @@ struct nir_to_llvm_context {
> int num_locals;
> LLVMValueRef *locals;
> bool has_ddxy;
> - uint8_t num_input_clips;
> - uint8_t num_input_culls;
> uint8_t num_output_clips;
> uint8_t num_output_culls;
>
> @@ -170,12 +168,9 @@ static unsigned shader_io_get_unique_index(gl_varying_slot slot)
> return 0;
> if (slot == VARYING_SLOT_PSIZ)
> return 1;
> - if (slot == VARYING_SLOT_CLIP_DIST0 ||
> - slot == VARYING_SLOT_CULL_DIST0)
> + if (slot == VARYING_SLOT_CLIP_DIST0)
> return 2;
> - if (slot == VARYING_SLOT_CLIP_DIST1 ||
> - slot == VARYING_SLOT_CULL_DIST1)
> - return 3;
> + /* 3 is reserved for clip dist as well */
> if (slot >= VARYING_SLOT_VAR0 && slot <= VARYING_SLOT_VAR31)
> return 4 + (slot - VARYING_SLOT_VAR0);
> unreachable("illegal slot in get unique index\n");
> @@ -2195,7 +2190,6 @@ load_gs_input(struct nir_to_llvm_context *ctx,
> unsigned param, vtx_offset_param;
> LLVMValueRef value[4], result;
> unsigned vertex_index;
> - unsigned cull_offset = 0;
> radv_get_deref_offset(ctx, &instr->variables[0]->deref,
> false, &vertex_index,
> &const_index, &indir_index);
> @@ -2205,13 +2199,11 @@ load_gs_input(struct nir_to_llvm_context *ctx,
> LLVMConstInt(ctx->i32, 4, false), "");
>
> param = shader_io_get_unique_index(instr->variables[0]->var->data.location);
> - if (instr->variables[0]->var->data.location == VARYING_SLOT_CULL_DIST0)
> - cull_offset += ctx->num_input_clips;
> for (unsigned i = 0; i < instr->num_components; i++) {
>
> args[0] = ctx->esgs_ring;
> args[1] = vtx_offset;
> - args[2] = LLVMConstInt(ctx->i32, (param * 4 + i + const_index + cull_offset) * 256, false);
> + args[2] = LLVMConstInt(ctx->i32, (param * 4 + i + const_index) * 256, false);
> args[3] = ctx->i32zero;
> args[4] = ctx->i32one; /* OFFEN */
> args[5] = ctx->i32zero; /* IDXEN */
> @@ -2366,8 +2358,7 @@ visit_store_var(struct nir_to_llvm_context *ctx,
>
> value = llvm_extract_elem(ctx, src, chan);
>
> - if (instr->variables[0]->var->data.location == VARYING_SLOT_CLIP_DIST0 ||
> - instr->variables[0]->var->data.location == VARYING_SLOT_CULL_DIST0)
> + if (instr->variables[0]->var->data.compact)
> stride = 1;
> if (indir_index) {
> unsigned count = glsl_count_attribute_slots(
> @@ -3143,7 +3134,7 @@ visit_emit_vertex(struct nir_to_llvm_context *ctx,
> LLVMValueRef gs_next_vertex;
> LLVMValueRef can_emit, kill;
> int idx;
> - int clip_cull_slot = -1;
> +
> assert(instr->const_index[0] == 0);
> /* Write vertex attribute values to GSVS ring */
> gs_next_vertex = LLVMBuildLoad(ctx->builder,
> @@ -3175,27 +3166,11 @@ visit_emit_vertex(struct nir_to_llvm_context *ctx,
> if (!(ctx->output_mask & (1ull << i)))
> continue;
>
> - if (i == VARYING_SLOT_CLIP_DIST1 ||
> - i == VARYING_SLOT_CULL_DIST1)
> - continue;
> -
> - if (i == VARYING_SLOT_CLIP_DIST0 ||
> - i == VARYING_SLOT_CULL_DIST0) {
> + if (i == VARYING_SLOT_CLIP_DIST0) {
> /* pack clip and cull into a single set of slots */
> - if (clip_cull_slot == -1) {
> - clip_cull_slot = idx;
> - if (ctx->num_output_clips + ctx->num_output_culls > 4)
> - slot_inc = 2;
> - } else {
> - slot = clip_cull_slot;
> - slot_inc = 0;
> - }
> - if (i == VARYING_SLOT_CLIP_DIST0)
> - length = ctx->num_output_clips;
> - if (i == VARYING_SLOT_CULL_DIST0) {
> - start = ctx->num_output_clips;
> - length = ctx->num_output_culls;
> - }
> + length = ctx->num_output_clips + ctx->num_output_culls;
> + if (length > 4)
> + slot_inc = 2;
> }
> for (unsigned j = 0; j < length; j++) {
> LLVMValueRef out_val = LLVMBuildLoad(ctx->builder,
> @@ -4083,22 +4058,6 @@ handle_vs_input_decl(struct nir_to_llvm_context *ctx,
> }
> }
>
> -static void
> -handle_gs_input_decl(struct nir_to_llvm_context *ctx,
> - struct nir_variable *variable)
> -{
> - int idx = variable->data.location;
> -
> - if (idx == VARYING_SLOT_CLIP_DIST0 ||
> - idx == VARYING_SLOT_CULL_DIST0) {
> - int length = glsl_get_length(glsl_get_array_element(variable->type));
> - if (idx == VARYING_SLOT_CLIP_DIST0)
> - ctx->num_input_clips = length;
> - else
> - ctx->num_input_culls = length;
> - }
> -}
> -
> static void interp_fs_input(struct nir_to_llvm_context *ctx,
> unsigned attr,
> LLVMValueRef interp_param,
> @@ -4191,9 +4150,6 @@ handle_shader_input_decl(struct nir_to_llvm_context *ctx,
> case MESA_SHADER_FRAGMENT:
> handle_fs_input_decl(ctx, variable);
> break;
> - case MESA_SHADER_GEOMETRY:
> - handle_gs_input_decl(ctx, variable);
> - break;
> default:
> break;
> }
> @@ -4276,33 +4232,33 @@ static LLVMValueRef si_build_alloca_undef(struct nir_to_llvm_context *ctx,
>
> static void
> handle_shader_output_decl(struct nir_to_llvm_context *ctx,
> + struct nir_shader *nir,
> struct nir_variable *variable)
> {
> int idx = variable->data.location + variable->data.index;
> unsigned attrib_count = glsl_count_attribute_slots(variable->type, false);
> -
> + unsigned mask_attribs;
> variable->data.driver_location = idx * 4;
>
> if (ctx->stage == MESA_SHADER_VERTEX ||
> ctx->stage == MESA_SHADER_GEOMETRY) {
> - if (idx == VARYING_SLOT_CLIP_DIST0 ||
> - idx == VARYING_SLOT_CULL_DIST0) {
> + if (idx == VARYING_SLOT_CLIP_DIST0) {
> int length = glsl_get_length(variable->type);
> - if (idx == VARYING_SLOT_CLIP_DIST0) {
> - if (ctx->stage == MESA_SHADER_VERTEX)
> - ctx->shader_info->vs.outinfo.clip_dist_mask = (1 << length) - 1;
> - ctx->num_output_clips = length;
> - } else if (idx == VARYING_SLOT_CULL_DIST0) {
> - if (ctx->stage == MESA_SHADER_VERTEX)
> - ctx->shader_info->vs.outinfo.cull_dist_mask = (1 << length) - 1;
> - ctx->num_output_culls = length;
> + if (ctx->stage == MESA_SHADER_VERTEX) {
> + ctx->shader_info->vs.outinfo.clip_dist_mask = (1 << nir->info->clip_distance_array_size) - 1;
> + ctx->shader_info->vs.outinfo.cull_dist_mask = (1 << nir->info->cull_distance_array_size) - 1;
> }
> + ctx->num_output_clips = nir->info->clip_distance_array_size;
> + ctx->num_output_culls = nir->info->cull_distance_array_size;
> +
> if (length > 4)
> attrib_count = 2;
> else
> attrib_count = 1;
> }
> - }
> + mask_attribs = ((1ull << 1) - 1) << idx;
> + } else
> + mask_attribs = ((1ull << attrib_count) - 1) << idx;
>
> for (unsigned i = 0; i < attrib_count; ++i) {
> for (unsigned chan = 0; chan < 4; chan++) {
> @@ -4310,7 +4266,7 @@ handle_shader_output_decl(struct nir_to_llvm_context *ctx,
> si_build_alloca_undef(ctx, ctx->f32, "");
> }
> }
> - ctx->output_mask |= ((1ull << attrib_count) - 1) << idx;
> + ctx->output_mask |= mask_attribs;
> }
>
> static void
> @@ -4519,14 +4475,10 @@ handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
> struct ac_export_args args, pos_args[4] = {};
> LLVMValueRef psize_value = NULL, layer_value = NULL, viewport_index_value = NULL;
> int i;
> - const uint64_t clip_mask = ctx->output_mask & ((1ull << VARYING_SLOT_CLIP_DIST0) |
> - (1ull << VARYING_SLOT_CLIP_DIST1) |
> - (1ull << VARYING_SLOT_CULL_DIST0) |
> - (1ull << VARYING_SLOT_CULL_DIST1));
>
> outinfo->prim_id_output = 0xffffffff;
> outinfo->layer_output = 0xffffffff;
> - if (clip_mask) {
> + if (ctx->output_mask & (1ull << VARYING_SLOT_CLIP_DIST0)) {
> LLVMValueRef slots[8];
> unsigned j;
>
> @@ -4534,13 +4486,9 @@ handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
> outinfo->cull_dist_mask <<= ctx->num_output_clips;
>
> i = VARYING_SLOT_CLIP_DIST0;
> - for (j = 0; j < ctx->num_output_clips; j++)
> + for (j = 0; j < ctx->num_output_clips + ctx->num_output_culls; j++)
> slots[j] = to_float(ctx, LLVMBuildLoad(ctx->builder,
> ctx->outputs[radeon_llvm_reg_index_soa(i, j)], ""));
> - i = VARYING_SLOT_CULL_DIST0;
> - for (j = 0; j < ctx->num_output_culls; j++)
> - slots[ctx->num_output_clips + j] = to_float(ctx, LLVMBuildLoad(ctx->builder,
> - ctx->outputs[radeon_llvm_reg_index_soa(i, j)], ""));
>
> for (i = ctx->num_output_clips + ctx->num_output_culls; i < 8; i++)
> slots[i] = LLVMGetUndef(ctx->f32);
> @@ -4570,10 +4518,7 @@ handle_vs_outputs_post(struct nir_to_llvm_context *ctx,
>
> if (i == VARYING_SLOT_POS) {
> target = V_008DFC_SQ_EXP_POS;
> - } else if (i == VARYING_SLOT_CLIP_DIST0 ||
> - i == VARYING_SLOT_CLIP_DIST1 ||
> - i == VARYING_SLOT_CULL_DIST0 ||
> - i == VARYING_SLOT_CULL_DIST1) {
> + } else if (i == VARYING_SLOT_CLIP_DIST0) {
> continue;
> } else if (i == VARYING_SLOT_PSIZ) {
> outinfo->writes_pointsize = true;
> @@ -4679,12 +4624,9 @@ handle_es_outputs_post(struct nir_to_llvm_context *ctx,
> if (!(ctx->output_mask & (1ull << i)))
> continue;
>
> - if (i == VARYING_SLOT_CLIP_DIST0) {
> - length = ctx->num_output_clips;
> - } else if (i == VARYING_SLOT_CULL_DIST0) {
> - start = ctx->num_output_clips;
> - length = ctx->num_output_culls;
> - }
> + if (i == VARYING_SLOT_CLIP_DIST0)
> + length = ctx->num_output_clips + ctx->num_output_culls;
> +
> param_index = shader_io_get_unique_index(i);
>
> if (param_index > max_output_written)
> @@ -4980,7 +4922,7 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
> handle_fs_inputs_pre(&ctx, nir);
>
> nir_foreach_variable(variable, &nir->outputs)
> - handle_shader_output_decl(&ctx, variable);
> + handle_shader_output_decl(&ctx, nir, variable);
>
> ctx.defs = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
> _mesa_key_pointer_equal);
> @@ -5185,7 +5127,7 @@ ac_gs_copy_shader_emit(struct nir_to_llvm_context *ctx)
> args[8] = ctx->i32zero; /* TFE */
>
> int idx = 0;
> - int clip_cull_slot = -1;
> +
> for (unsigned i = 0; i < RADEON_LLVM_MAX_OUTPUTS; ++i) {
> int length = 4;
> int start = 0;
> @@ -5194,27 +5136,11 @@ ac_gs_copy_shader_emit(struct nir_to_llvm_context *ctx)
> if (!(ctx->output_mask & (1ull << i)))
> continue;
>
> - if (i == VARYING_SLOT_CLIP_DIST1 ||
> - i == VARYING_SLOT_CULL_DIST1)
> - continue;
> -
> - if (i == VARYING_SLOT_CLIP_DIST0 ||
> - i == VARYING_SLOT_CULL_DIST0) {
> + if (i == VARYING_SLOT_CLIP_DIST0) {
> /* unpack clip and cull from a single set of slots */
> - if (clip_cull_slot == -1) {
> - clip_cull_slot = idx;
> - if (ctx->num_output_clips + ctx->num_output_culls > 4)
> - slot_inc = 2;
> - } else {
> - slot = clip_cull_slot;
> - slot_inc = 0;
> - }
> - if (i == VARYING_SLOT_CLIP_DIST0)
> - length = ctx->num_output_clips;
> - if (i == VARYING_SLOT_CULL_DIST0) {
> - start = ctx->num_output_clips;
> - length = ctx->num_output_culls;
> - }
> + length = ctx->num_output_clips + ctx->num_output_culls;
> + if (length > 4)
> + slot_inc = 2;
> }
>
> for (unsigned j = 0; j < length; j++) {
> @@ -5268,7 +5194,7 @@ void ac_create_gs_copy_shader(LLVMTargetMachineRef tm,
> ac_setup_rings(&ctx);
>
> nir_foreach_variable(variable, &geom_shader->outputs)
> - handle_shader_output_decl(&ctx, variable);
> + handle_shader_output_decl(&ctx, geom_shader, variable);
>
> ac_gs_copy_shader_emit(&ctx);
>
> diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
> index 07020e8..df46276 100644
> --- a/src/amd/vulkan/radv_pipeline.c
> +++ b/src/amd/vulkan/radv_pipeline.c
> @@ -246,6 +246,7 @@ radv_shader_compile_to_nir(struct radv_device *device,
> */
> NIR_PASS_V(nir, nir_lower_constant_initializers, ~0);
> NIR_PASS_V(nir, nir_lower_system_values);
> + NIR_PASS_V(nir, nir_lower_clip_cull_distance_arrays);
> }
>
> /* Vulkan uses the separate-shader linking model */
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: OpenPGP digital signature
URL: <https://lists.freedesktop.org/archives/mesa-dev/attachments/20170330/e43ee670/attachment.sig>
More information about the mesa-dev
mailing list