ac: replace llvm.SI.tbuffer.store with llvm.amdgcn.buffer.store if ADD_TID=0

ADD_TID doesn't work. Needs more investigation.

v2: remove leftover dead code

Reviewed-by: Dave Airlie <airlied@redhat.com> (v1)

src/amd/common/ac_llvm_build.c

@@ -551,8 +551,64 @@ ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
 			    LLVMValueRef soffset,
 			    unsigned inst_offset,
 			    bool glc,
-			    bool slc)
+			    bool slc,
+			    bool writeonly_memory,
+			    bool has_add_tid)
 {
+	/* TODO: Fix stores with ADD_TID and remove the "has_add_tid" flag. */
+	if (HAVE_LLVM >= 0x0309 && !has_add_tid) {
+		/* Split 3 channel stores, becase LLVM doesn't support 3-channel
+		 * intrinsics. */
+		if (num_channels == 3) {
+			LLVMValueRef v[3], v01;
+
+			for (int i = 0; i < 3; i++) {
+				v[i] = LLVMBuildExtractElement(ctx->builder, vdata,
+						LLVMConstInt(ctx->i32, i, 0), "");
+			}
+			v01 = ac_build_gather_values(ctx, v, 2);
+
+			ac_build_buffer_store_dword(ctx, rsrc, v01, 2, voffset,
+						    soffset, inst_offset, glc, slc,
+						    writeonly_memory, has_add_tid);
+			ac_build_buffer_store_dword(ctx, rsrc, v[2], 1, voffset,
+						    soffset, inst_offset + 8,
+						    glc, slc,
+						    writeonly_memory, has_add_tid);
+			return;
+		}
+
+		unsigned func = CLAMP(num_channels, 1, 3) - 1;
+		static const char *types[] = {"f32", "v2f32", "v4f32"};
+		char name[256];
+		LLVMValueRef offset = soffset;
+
+		if (inst_offset)
+			offset = LLVMBuildAdd(ctx->builder, offset,
+					      LLVMConstInt(ctx->i32, inst_offset, 0), "");
+		if (voffset)
+			offset = LLVMBuildAdd(ctx->builder, offset, voffset, "");
+
+		LLVMValueRef args[] = {
+			bitcast_to_float(ctx, vdata),
+			LLVMBuildBitCast(ctx->builder, rsrc, ctx->v4i32, ""),
+			LLVMConstInt(ctx->i32, 0, 0),
+			offset,
+			LLVMConstInt(ctx->i1, glc, 0),
+			LLVMConstInt(ctx->i1, slc, 0),
+		};
+
+		snprintf(name, sizeof(name), "llvm.amdgcn.buffer.store.%s",
+			 types[func]);
+
+		ac_emit_llvm_intrinsic(ctx, name, ctx->voidt,
+				       args, ARRAY_SIZE(args),
+				       writeonly_memory ?
+					AC_FUNC_ATTR_INACCESSIBLE_MEM_ONLY :
+					AC_FUNC_ATTR_WRITEONLY);
+		return;
+	}
+
 	static unsigned dfmt[] = {
 		V_008F0C_BUF_DATA_FORMAT_32,
 		V_008F0C_BUF_DATA_FORMAT_32_32,

src/amd/common/ac_llvm_build.h

@@ -130,7 +130,9 @@ ac_build_buffer_store_dword(struct ac_llvm_context *ctx,
 			    LLVMValueRef soffset,
 			    unsigned inst_offset,
 		            bool glc,
-		            bool slc);
+		            bool slc,
+			    bool writeonly_memory,
+			    bool has_add_tid);
 LLVMValueRef
 ac_build_buffer_load(struct ac_llvm_context *ctx,
 		     LLVMValueRef rsrc,

src/amd/common/ac_nir_to_llvm.c

@@ -3159,7 +3159,7 @@ visit_emit_vertex(struct nir_to_llvm_context *ctx,
 			ac_build_buffer_store_dword(&ctx->ac, ctx->gsvs_ring,
 						    out_val, 1,
 						    voffset, ctx->gs2vs_offset, 0,
-						    1, 1);
+						    1, 1, true, true);
 		}
 		idx += slot_inc;
 	}
@@ -4675,7 +4675,7 @@ handle_es_outputs_post(struct nir_to_llvm_context *ctx)
 					       out_val, 1,
 					       NULL, ctx->es2gs_offset,
 					       (4 * param_index + j + start) * 4,
-					       1, 1);
+					       1, 1, true, true);
 		}
 	}
 	ctx->shader_info->vs.esgs_itemsize = (max_output_written + 1) * 16;

src/gallium/drivers/radeonsi/si_shader.c

@@ -1049,7 +1049,7 @@ static void store_output_tcs(struct lp_build_tgsi_context *bld_base,
 		if (inst->Dst[0].Register.WriteMask != 0xF && !is_tess_factor) {
 			ac_build_buffer_store_dword(&ctx->ac, buffer, value, 1,
 						    buf_addr, base,
-						    4 * chan_index, 1, 0);
+						    4 * chan_index, 1, 0, true, false);
 		}
 	}
 
@@ -1057,7 +1057,7 @@ static void store_output_tcs(struct lp_build_tgsi_context *bld_base,
 		LLVMValueRef value = lp_build_gather_values(bld_base->base.gallivm,
 		                                            values, 4);
 		ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, buf_addr,
-					    base, 0, 1, 0);
+					    base, 0, 1, 0, true, false);
 	}
 }
 
@@ -2087,7 +2087,7 @@ static void emit_streamout_output(struct si_shader_context *ctx,
 				    vdata, num_comps,
 				    so_write_offsets[buf_idx],
 				    LLVMConstInt(ctx->i32, 0, 0),
-				    stream_out->dst_offset * 4, 1, 1);
+				    stream_out->dst_offset * 4, 1, 1, true, false);
 }
 
 /**
@@ -2412,7 +2412,7 @@ static void si_copy_tcs_inputs(struct lp_build_tgsi_context *bld_base)
 		                              lds_ptr);
 
 		ac_build_buffer_store_dword(&ctx->ac, buffer, value, 4, buffer_addr,
-					    buffer_offset, 0, 1, 0);
+					    buffer_offset, 0, 1, 0, true, false);
 	}
 }
 
@@ -2527,18 +2527,18 @@ static void si_write_tess_factors(struct lp_build_tgsi_context *bld_base,
 	ac_build_buffer_store_dword(&ctx->ac, buffer,
 				    lp_build_const_int32(gallivm, 0x80000000),
 				    1, lp_build_const_int32(gallivm, 0), tf_base,
-				    0, 1, 0);
+				    0, 1, 0, true, false);
 
 	lp_build_endif(&inner_if_ctx);
 
 	/* Store the tessellation factors. */
 	ac_build_buffer_store_dword(&ctx->ac, buffer, vec0,
 				    MIN2(stride, 4), byteoffset, tf_base,
-				    4, 1, 0);
+				    4, 1, 0, true, false);
 	if (vec1)
 		ac_build_buffer_store_dword(&ctx->ac, buffer, vec1,
 					    stride - 4, byteoffset, tf_base,
-					    20, 1, 0);
+					    20, 1, 0, true, false);
 
 	/* Store the tess factors into the offchip buffer if TES reads them. */
 	if (shader->key.part.tcs.epilog.tes_reads_tess_factors) {
@@ -2560,7 +2560,7 @@ static void si_write_tess_factors(struct lp_build_tgsi_context *bld_base,
 
 		ac_build_buffer_store_dword(&ctx->ac, buf, outer_vec,
 					    outer_comps, tf_outer_offset,
-					    base, 0, 1, 0);
+					    base, 0, 1, 0, true, false);
 		if (inner_comps) {
 			param_inner = si_shader_io_get_unique_index(
 					      TGSI_SEMANTIC_TESSINNER, 0);
@@ -2571,7 +2571,7 @@ static void si_write_tess_factors(struct lp_build_tgsi_context *bld_base,
 				    lp_build_gather_values(gallivm, inner, inner_comps);
 			ac_build_buffer_store_dword(&ctx->ac, buf, inner_vec,
 						    inner_comps, tf_inner_offset,
-						    base, 0, 1, 0);
+						    base, 0, 1, 0, true, false);
 		}
 	}
 
@@ -2695,7 +2695,7 @@ static void si_llvm_emit_es_epilogue(struct lp_build_tgsi_context *bld_base)
 						    ctx->esgs_ring,
 						    out_val, 1, NULL, soffset,
 						    (4 * param_index + chan) * 4,
-						    1, 1);
+						    1, 1, true, true);
 		}
 	}
 }
@@ -5063,7 +5063,7 @@ static void si_llvm_emit_vertex(
 						    ctx->gsvs_ring[stream],
 						    out_val, 1,
 						    voffset, soffset, 0,
-						    1, 1);
+						    1, 1, true, true);
 		}
 	}