nir,ac/llvm: add nir_load_smem_buffer_amd

Used by radeonsi to load const buffer.

Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18010>

diff --git a/src/amd/llvm/ac_nir_to_llvm.c b/src/amd/llvm/ac_nir_to_llvm.c
index 23140cb5d93e93a65b6afbc2db2977605f531d80..f67baa13c8fd8c2409211e3f6c7dc2b416ae8cd1 100644 (file)
--- a/src/amd/llvm/ac_nir_to_llvm.c
+++ b/src/amd/llvm/ac_nir_to_llvm.c
@@ -4403,6 +4403,15 @@ static bool visit_intrinsic(struct ac_nir_context *ctx, nir_intrinsic_instr *ins
       LLVMSetMetadata(result, ctx->ac.invariant_load_md_kind, ctx->ac.empty_md);
       break;
    }
+   case nir_intrinsic_load_smem_buffer_amd: {
+      LLVMValueRef descriptor = get_src(ctx, instr->src[0]);
+      LLVMValueRef offset = get_src(ctx, instr->src[1]);
+      unsigned num_components = instr->dest.ssa.num_components;
+
+      result = ac_build_buffer_load(&ctx->ac, descriptor, num_components, NULL, offset, NULL,
+                                    ctx->ac.i32, 0, true, true);
+      break;
+   }
    case nir_intrinsic_ordered_xfb_counter_add_amd: {
       /* must be called in a single lane of a workgroup. */
       LLVMTypeRef gdsptr = LLVMPointerType(ctx->ac.i32, AC_ADDR_SPACE_GDS);

diff --git a/src/compiler/nir/nir_divergence_analysis.c b/src/compiler/nir/nir_divergence_analysis.c
index 20d361c4ba4663e87dd5b0c9dee3f4abc441d0ba..40d6f1666ad4cdc04bda0fedd19af8a86a6885fd 100644 (file)
--- a/src/compiler/nir/nir_divergence_analysis.c
+++ b/src/compiler/nir/nir_divergence_analysis.c
@@ -185,6 +185,7 @@ visit_intrinsic(nir_shader *shader, nir_intrinsic_instr *instr)
    case nir_intrinsic_load_tess_level_outer_default:
    case nir_intrinsic_load_scalar_arg_amd:
    case nir_intrinsic_load_smem_amd:
+   case nir_intrinsic_load_smem_buffer_amd:
    case nir_intrinsic_load_rt_dynamic_callable_stack_base_amd:
    case nir_intrinsic_load_global_const_block_intel:
    case nir_intrinsic_load_reloc_const_intel:

diff --git a/src/compiler/nir/nir_intrinsics.py b/src/compiler/nir/nir_intrinsics.py
index 64e8ae0dfa3782befec3752b128c64bf468b1dbf..937185ce3068965e968571d207b5918597bf646b 100644 (file)
--- a/src/compiler/nir/nir_intrinsics.py
+++ b/src/compiler/nir/nir_intrinsics.py
@@ -1482,6 +1482,11 @@ intrinsic("load_smem_amd", src_comp=[1, 1], dest_comp=0, bit_sizes=[32],
                            indices=[ALIGN_MUL, ALIGN_OFFSET],
                            flags=[CAN_ELIMINATE, CAN_REORDER])
 
+# src[] = { descriptor, offset }
+intrinsic("load_smem_buffer_amd", src_comp=[4, 1], dest_comp=0, bit_sizes=[32],
+                                  indices=[ALIGN_MUL, ALIGN_OFFSET],
+                                  flags=[CAN_ELIMINATE, CAN_REORDER])
+
 # src[] = { offset }.
 intrinsic("load_shared2_amd", [1], dest_comp=2, indices=[OFFSET0, OFFSET1, ST64], flags=[CAN_ELIMINATE])