freedreno: re-work fd_batch_reference() locking
[mesa.git] / src / amd / vulkan / radv_nir_to_llvm.c
index 5168c9d554aa8e661e8c515a7455de2448dc8837..c7d772fa652bad4ba5d32dfe3a6f3914aba0e2de 100644 (file)
@@ -27,6 +27,7 @@
 
 #include "radv_private.h"
 #include "radv_shader.h"
+#include "radv_shader_helper.h"
 #include "nir/nir.h"
 
 #include <llvm-c/Core.h>
@@ -81,7 +82,6 @@ struct radv_shader_context {
        LLVMValueRef hs_ring_tess_offchip;
        LLVMValueRef hs_ring_tess_factor;
 
-       LLVMValueRef sample_pos_offset;
        LLVMValueRef persp_sample, persp_center, persp_centroid;
        LLVMValueRef linear_sample, linear_center, linear_centroid;
 
@@ -579,11 +579,14 @@ static void
 set_loc_desc(struct radv_shader_context *ctx, int idx,  uint8_t *sgpr_idx,
             uint32_t indirect_offset)
 {
-       struct radv_userdata_info *ud_info =
-               &ctx->shader_info->user_sgprs_locs.descriptor_sets[idx];
+       struct radv_userdata_locations *locs =
+               &ctx->shader_info->user_sgprs_locs;
+       struct radv_userdata_info *ud_info = &locs->descriptor_sets[idx];
        assert(ud_info);
 
        set_loc(ud_info, sgpr_idx, HAVE_32BIT_POINTERS ? 1 : 2, indirect_offset);
+       if (indirect_offset == 0)
+               locs->descriptor_sets_enabled |= 1 << idx;
 }
 
 struct user_sgpr_info {
@@ -1095,10 +1098,6 @@ static void create_function(struct radv_shader_context *ctx,
                                           previous_stage, &user_sgpr_info,
                                           &args, &desc_sets);
 
-               if (ctx->shader_info->info.ps.needs_sample_positions)
-                       add_arg(&args, ARG_SGPR, ctx->ac.i32,
-                               &ctx->sample_pos_offset);
-
                add_arg(&args, ARG_SGPR, ctx->ac.i32, &ctx->abi.prim_mask);
                add_arg(&args, ARG_VGPR, ctx->ac.v2i32, &ctx->persp_sample);
                add_arg(&args, ARG_VGPR, ctx->ac.v2i32, &ctx->persp_center);
@@ -1194,10 +1193,6 @@ static void create_function(struct radv_shader_context *ctx,
                        set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx, 1);
                break;
        case MESA_SHADER_FRAGMENT:
-               if (ctx->shader_info->info.ps.needs_sample_positions) {
-                       set_loc_shader(ctx, AC_UD_PS_SAMPLE_POS_OFFSET,
-                                      &user_sgpr_idx, 1);
-               }
                break;
        default:
                unreachable("Shader stage not implemented");
@@ -1627,6 +1622,30 @@ static LLVMValueRef lookup_interp_param(struct ac_shader_abi *abi,
        return NULL;
 }
 
+static uint32_t
+radv_get_sample_pos_offset(uint32_t num_samples)
+{
+       uint32_t sample_pos_offset = 0;
+
+       switch (num_samples) {
+       case 2:
+               sample_pos_offset = 1;
+               break;
+       case 4:
+               sample_pos_offset = 3;
+               break;
+       case 8:
+               sample_pos_offset = 7;
+               break;
+       case 16:
+               sample_pos_offset = 15;
+               break;
+       default:
+               break;
+       }
+       return sample_pos_offset;
+}
+
 static LLVMValueRef load_sample_position(struct ac_shader_abi *abi,
                                         LLVMValueRef sample_id)
 {
@@ -1638,7 +1657,12 @@ static LLVMValueRef load_sample_position(struct ac_shader_abi *abi,
        ptr = LLVMBuildBitCast(ctx->ac.builder, ptr,
                               ac_array_in_const_addr_space(ctx->ac.v2f32), "");
 
-       sample_id = LLVMBuildAdd(ctx->ac.builder, sample_id, ctx->sample_pos_offset, "");
+       uint32_t sample_pos_offset =
+               radv_get_sample_pos_offset(ctx->options->key.fs.num_samples);
+
+       sample_id =
+               LLVMBuildAdd(ctx->ac.builder, sample_id,
+                            LLVMConstInt(ctx->ac.i32, sample_pos_offset, false), "");
        result = ac_build_load_invariant(&ctx->ac, ptr, sample_id);
 
        return result;
@@ -1648,9 +1672,14 @@ static LLVMValueRef load_sample_position(struct ac_shader_abi *abi,
 static LLVMValueRef load_sample_mask_in(struct ac_shader_abi *abi)
 {
        struct radv_shader_context *ctx = radv_shader_context_from_abi(abi);
-       uint8_t log2_ps_iter_samples = ctx->shader_info->info.ps.force_persample ?
-               ctx->options->key.fs.log2_num_samples :
-               ctx->options->key.fs.log2_ps_iter_samples;
+       uint8_t log2_ps_iter_samples;
+
+       if (ctx->shader_info->info.ps.force_persample) {
+               log2_ps_iter_samples =
+                       util_logbase2(ctx->options->key.fs.num_samples);
+       } else {
+               log2_ps_iter_samples = ctx->options->key.fs.log2_ps_iter_samples;
+       }
 
        /* The bit pattern matches that used by fixed function fragment
         * processing. */
@@ -2968,35 +2997,11 @@ handle_shader_outputs_post(struct ac_shader_abi *abi, unsigned max_outputs,
 }
 
 static void ac_llvm_finalize_module(struct radv_shader_context *ctx,
+                                   LLVMPassManagerRef passmgr,
                                    const struct radv_nir_compiler_options *options)
 {
-       LLVMPassManagerRef passmgr;
-       /* Create the pass manager */
-       passmgr = LLVMCreateFunctionPassManagerForModule(
-                                                       ctx->ac.module);
-
-       if (options->check_ir)
-               LLVMAddVerifierPass(passmgr);
-
-       /* This pass should eliminate all the load and store instructions */
-       LLVMAddPromoteMemoryToRegisterPass(passmgr);
-
-       /* Add some optimization passes */
-       LLVMAddScalarReplAggregatesPass(passmgr);
-       LLVMAddLICMPass(passmgr);
-       LLVMAddAggressiveDCEPass(passmgr);
-       LLVMAddCFGSimplificationPass(passmgr);
-       /* This is recommended by the instruction combining pass. */
-       LLVMAddEarlyCSEMemSSAPass(passmgr);
-       LLVMAddInstructionCombiningPass(passmgr);
-
-       /* Run the pass */
-       LLVMInitializeFunctionPassManager(passmgr);
-       LLVMRunFunctionPassManager(passmgr, ctx->main_function);
-       LLVMFinalizeFunctionPassManager(passmgr);
-
+       LLVMRunPassManager(passmgr, ctx->ac.module);
        LLVMDisposeBuilder(ctx->ac.builder);
-       LLVMDisposePassManager(passmgr);
 
        ac_llvm_context_dispose(&ctx->ac);
 }
@@ -3107,7 +3112,6 @@ static void ac_nir_fixup_ls_hs_input_vgprs(struct radv_shader_context *ctx)
        LLVMValueRef hs_empty = LLVMBuildICmp(ctx->ac.builder, LLVMIntEQ, count,
                                              ctx->ac.i32_0, "");
        ctx->abi.instance_id = LLVMBuildSelect(ctx->ac.builder, hs_empty, ctx->rel_auto_id, ctx->abi.instance_id, "");
-       ctx->vs_prim_id = LLVMBuildSelect(ctx->ac.builder, hs_empty, ctx->abi.vertex_id, ctx->vs_prim_id, "");
        ctx->rel_auto_id = LLVMBuildSelect(ctx->ac.builder, hs_empty, ctx->abi.tcs_rel_ids, ctx->rel_auto_id, "");
        ctx->abi.vertex_id = LLVMBuildSelect(ctx->ac.builder, hs_empty, ctx->abi.tcs_patch_id, ctx->abi.vertex_id, "");
 }
@@ -3127,7 +3131,7 @@ static void prepare_gs_input_vgprs(struct radv_shader_context *ctx)
 
 
 static
-LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
+LLVMModuleRef ac_translate_nir_to_llvm(struct ac_llvm_compiler *ac_llvm,
                                        struct nir_shader *const *shaders,
                                        int shader_count,
                                        struct radv_shader_variant_info *shader_info,
@@ -3137,18 +3141,10 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
        unsigned i;
        ctx.options = options;
        ctx.shader_info = shader_info;
-       ctx.context = LLVMContextCreate();
 
-       ac_llvm_context_init(&ctx.ac, ctx.context, options->chip_class,
-                            options->family);
-       ctx.ac.module = LLVMModuleCreateWithNameInContext("shader", ctx.context);
-       LLVMSetTarget(ctx.ac.module, options->supports_spill ? "amdgcn-mesa-mesa3d" : "amdgcn--");
-
-       LLVMTargetDataRef data_layout = LLVMCreateTargetDataLayout(tm);
-       char *data_layout_str = LLVMCopyStringRepOfTargetData(data_layout);
-       LLVMSetDataLayout(ctx.ac.module, data_layout_str);
-       LLVMDisposeTargetData(data_layout);
-       LLVMDisposeMessage(data_layout_str);
+       ac_llvm_context_init(&ctx.ac, options->chip_class, options->family);
+       ctx.context = ctx.ac.context;
+       ctx.ac.module = ac_create_module(ac_llvm->tm, ctx.context);
 
        enum ac_float_mode float_mode =
                options->unsafe_math ? AC_FLOAT_MODE_UNSAFE_FP_MATH :
@@ -3303,7 +3299,7 @@ LLVMModuleRef ac_translate_nir_to_llvm(LLVMTargetMachineRef tm,
        if (options->dump_preoptir)
                ac_dump_module(ctx.ac.module);
 
-       ac_llvm_finalize_module(&ctx, options);
+       ac_llvm_finalize_module(&ctx, ac_llvm->passmgr, options);
 
        if (shader_count == 1)
                ac_nir_eliminate_const_vs_outputs(&ctx);
@@ -3333,15 +3329,10 @@ static void ac_diagnostic_handler(LLVMDiagnosticInfoRef di, void *context)
 
 static unsigned ac_llvm_compile(LLVMModuleRef M,
                                 struct ac_shader_binary *binary,
-                                LLVMTargetMachineRef tm)
+                                struct ac_llvm_compiler *ac_llvm)
 {
        unsigned retval = 0;
-       char *err;
        LLVMContextRef llvm_ctx;
-       LLVMMemoryBufferRef out_buffer;
-       unsigned buffer_size;
-       const char *buffer_data;
-       LLVMBool mem_err;
 
        /* Setup Diagnostic Handler*/
        llvm_ctx = LLVMGetModuleContext(M);
@@ -3350,31 +3341,12 @@ static unsigned ac_llvm_compile(LLVMModuleRef M,
                                        &retval);
 
        /* Compile IR*/
-       mem_err = LLVMTargetMachineEmitToMemoryBuffer(tm, M, LLVMObjectFile,
-                                                     &err, &out_buffer);
-
-       /* Process Errors/Warnings */
-       if (mem_err) {
-               fprintf(stderr, "%s: %s", __FUNCTION__, err);
-               free(err);
+       if (!radv_compile_to_binary(ac_llvm, M, binary))
                retval = 1;
-               goto out;
-       }
-
-       /* Extract Shader Code*/
-       buffer_size = LLVMGetBufferSize(out_buffer);
-       buffer_data = LLVMGetBufferStart(out_buffer);
-
-       ac_elf_read(buffer_data, buffer_size, binary);
-
-       /* Clean up */
-       LLVMDisposeMemoryBuffer(out_buffer);
-
-out:
        return retval;
 }
 
-static void ac_compile_llvm_module(LLVMTargetMachineRef tm,
+static void ac_compile_llvm_module(struct ac_llvm_compiler *ac_llvm,
                                   LLVMModuleRef llvm_module,
                                   struct ac_shader_binary *binary,
                                   struct ac_shader_config *config,
@@ -3393,7 +3365,7 @@ static void ac_compile_llvm_module(LLVMTargetMachineRef tm,
                LLVMDisposeMessage(llvm_ir);
        }
 
-       int v = ac_llvm_compile(llvm_module, binary, tm);
+       int v = ac_llvm_compile(llvm_module, binary, ac_llvm);
        if (v) {
                fprintf(stderr, "compile failed\n");
        }
@@ -3503,7 +3475,7 @@ ac_fill_shader_info(struct radv_shader_variant_info *shader_info, struct nir_sha
 }
 
 void
-radv_compile_nir_shader(LLVMTargetMachineRef tm,
+radv_compile_nir_shader(struct ac_llvm_compiler *ac_llvm,
                        struct ac_shader_binary *binary,
                        struct ac_shader_config *config,
                        struct radv_shader_variant_info *shader_info,
@@ -3514,10 +3486,10 @@ radv_compile_nir_shader(LLVMTargetMachineRef tm,
 
        LLVMModuleRef llvm_module;
 
-       llvm_module = ac_translate_nir_to_llvm(tm, nir, nir_count, shader_info,
+       llvm_module = ac_translate_nir_to_llvm(ac_llvm, nir, nir_count, shader_info,
                                               options);
 
-       ac_compile_llvm_module(tm, llvm_module, binary, config, shader_info,
+       ac_compile_llvm_module(ac_llvm, llvm_module, binary, config, shader_info,
                               nir[0]->info.stage, options);
 
        for (int i = 0; i < nir_count; ++i)
@@ -3575,7 +3547,7 @@ ac_gs_copy_shader_emit(struct radv_shader_context *ctx)
 }
 
 void
-radv_compile_gs_copy_shader(LLVMTargetMachineRef tm,
+radv_compile_gs_copy_shader(struct ac_llvm_compiler *ac_llvm,
                            struct nir_shader *geom_shader,
                            struct ac_shader_binary *binary,
                            struct ac_shader_config *config,
@@ -3583,16 +3555,14 @@ radv_compile_gs_copy_shader(LLVMTargetMachineRef tm,
                            const struct radv_nir_compiler_options *options)
 {
        struct radv_shader_context ctx = {0};
-       ctx.context = LLVMContextCreate();
        ctx.options = options;
        ctx.shader_info = shader_info;
 
-       ac_llvm_context_init(&ctx.ac, ctx.context, options->chip_class,
-                            options->family);
-       ctx.ac.module = LLVMModuleCreateWithNameInContext("shader", ctx.context);
+       ac_llvm_context_init(&ctx.ac, options->chip_class, options->family);
+       ctx.context = ctx.ac.context;
+       ctx.ac.module = ac_create_module(ac_llvm->tm, ctx.context);
 
        ctx.is_gs_copy_shader = true;
-       LLVMSetTarget(ctx.ac.module, "amdgcn--");
 
        enum ac_float_mode float_mode =
                options->unsafe_math ? AC_FLOAT_MODE_UNSAFE_FP_MATH :
@@ -3621,8 +3591,8 @@ radv_compile_gs_copy_shader(LLVMTargetMachineRef tm,
 
        LLVMBuildRetVoid(ctx.ac.builder);
 
-       ac_llvm_finalize_module(&ctx, options);
+       ac_llvm_finalize_module(&ctx, ac_llvm->passmgr, options);
 
-       ac_compile_llvm_module(tm, ctx.ac.module, binary, config, shader_info,
+       ac_compile_llvm_module(ac_llvm, ctx.ac.module, binary, config, shader_info,
                               MESA_SHADER_VERTEX, options);
 }