ac: add v3i32 to the common code and make use of it

Reviewed-by: Marek Olšák <marek.olsak@amd.com
Acked-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

diff --git a/src/amd/common/ac_llvm_build.c b/src/amd/common/ac_llvm_build.c
index 1a0d44bcdd30764024068533c43cc934c069af1c..1519262b3ddd601143a495e7f0eb9c7921d1dea7 100644 (file)
--- a/src/amd/common/ac_llvm_build.c
+++ b/src/amd/common/ac_llvm_build.c
@@ -67,6 +67,7 @@ ac_llvm_context_init(struct ac_llvm_context *ctx, LLVMContextRef context,
        ctx->f32 = LLVMFloatTypeInContext(ctx->context);
        ctx->f64 = LLVMDoubleTypeInContext(ctx->context);
        ctx->v2i32 = LLVMVectorType(ctx->i32, 2);
+       ctx->v3i32 = LLVMVectorType(ctx->i32, 3);
        ctx->v4i32 = LLVMVectorType(ctx->i32, 4);
        ctx->v4f32 = LLVMVectorType(ctx->f32, 4);
        ctx->v8i32 = LLVMVectorType(ctx->i32, 8);

diff --git a/src/amd/common/ac_llvm_build.h b/src/amd/common/ac_llvm_build.h
index 7fc336c3f980083a20020cbecb78b71335e79a2d..f662cc4dd89fab729eb61a3dc78679adfaf74ce2 100644 (file)
--- a/src/amd/common/ac_llvm_build.h
+++ b/src/amd/common/ac_llvm_build.h
@@ -53,6 +53,7 @@ struct ac_llvm_context {
        LLVMTypeRef f32;
        LLVMTypeRef f64;
        LLVMTypeRef v2i32;
+       LLVMTypeRef v3i32;
        LLVMTypeRef v4i32;
        LLVMTypeRef v4f32;
        LLVMTypeRef v8i32;

diff --git a/src/amd/common/ac_nir_to_llvm.c b/src/amd/common/ac_nir_to_llvm.c
index 0bcb0b0525c5963d85194e034a37737af6ab5e0d..6a638e3f16613a56e124c07af593f8f2d522aefb 100644 (file)
--- a/src/amd/common/ac_nir_to_llvm.c
+++ b/src/amd/common/ac_nir_to_llvm.c
@@ -134,7 +134,6 @@ struct nir_to_llvm_context {
        LLVMValueRef persp_sample, persp_center, persp_centroid;
        LLVMValueRef linear_sample, linear_center, linear_centroid;
 
-       LLVMTypeRef v3i32;
        LLVMTypeRef v4i32;
        LLVMTypeRef v8i32;
        LLVMTypeRef f64;
@@ -734,9 +733,9 @@ static void create_function(struct nir_to_llvm_context *ctx,
                radv_define_common_user_sgprs_phase1(ctx, stage, has_previous_stage, previous_stage, &user_sgpr_info, &args, &desc_sets);
                if (ctx->shader_info->info.cs.grid_components_used)
                        add_user_sgpr_argument(&args, LLVMVectorType(ctx->ac.i32, ctx->shader_info->info.cs.grid_components_used), &ctx->num_work_groups); /* grid size */
-               add_sgpr_argument(&args, LLVMVectorType(ctx->ac.i32, 3), &ctx->workgroup_ids);
+               add_sgpr_argument(&args, ctx->ac.v3i32, &ctx->workgroup_ids);
                add_sgpr_argument(&args, ctx->ac.i32, &ctx->tg_size);
-               add_vgpr_argument(&args, LLVMVectorType(ctx->ac.i32, 3), &ctx->local_invocation_ids);
+               add_vgpr_argument(&args, ctx->ac.v3i32, &ctx->local_invocation_ids);
                break;
        case MESA_SHADER_VERTEX:
                radv_define_common_user_sgprs_phase1(ctx, stage, has_previous_stage, previous_stage, &user_sgpr_info, &args, &desc_sets);
@@ -879,7 +878,7 @@ static void create_function(struct nir_to_llvm_context *ctx,
                add_vgpr_argument(&args, ctx->ac.v2i32, &ctx->persp_sample); /* persp sample */
                add_vgpr_argument(&args, ctx->ac.v2i32, &ctx->persp_center); /* persp center */
                add_vgpr_argument(&args, ctx->ac.v2i32, &ctx->persp_centroid); /* persp centroid */
-               add_vgpr_argument(&args, ctx->v3i32, NULL); /* persp pull model */
+               add_vgpr_argument(&args, ctx->ac.v3i32, NULL); /* persp pull model */
                add_vgpr_argument(&args, ctx->ac.v2i32, &ctx->linear_sample); /* linear sample */
                add_vgpr_argument(&args, ctx->ac.v2i32, &ctx->linear_center); /* linear center */
                add_vgpr_argument(&args, ctx->ac.v2i32, &ctx->linear_centroid); /* linear centroid */
@@ -991,7 +990,6 @@ static void create_function(struct nir_to_llvm_context *ctx,
 
 static void setup_types(struct nir_to_llvm_context *ctx)
 {
-       ctx->v3i32 = LLVMVectorType(ctx->ac.i32, 3);
        ctx->v4i32 = LLVMVectorType(ctx->ac.i32, 4);
        ctx->v8i32 = LLVMVectorType(ctx->ac.i32, 8);
        ctx->f32 = LLVMFloatTypeInContext(ctx->context);