xe: jit: move dsl into gemmstone

Author: rjoursler

src/gpu/intel/block_structure.hpp

@@ -24,8 +24,8 @@
 #include "common/memory_desc_wrapper.hpp"
 #include "common/serialization.hpp"
 #include "common/utils.hpp"
+#include "gemmstone/dsl/tensor.hpp"
 #include "gpu/intel/compute/kernel_arg_list.hpp"
-#include "gpu/intel/jit/dsl/tensor.hpp"
 #include "gpu/intel/utils.hpp"
 
 namespace dnnl {

src/gpu/intel/conv/jit/config.hpp

@@ -25,10 +25,9 @@
 #include "gpu/intel/conv/jit/key.hpp"
 #include "gpu/intel/conv/jit/problem.hpp"
 #include "gpu/intel/jit/ir/config.hpp"
-#include "gpu/intel/jit/ir/fma.hpp"
 #include "gpu/intel/jit/ir/hw.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/ir/tensor_config.hpp"
-#include "gpu/intel/jit/ir/walk_order.hpp"
 #include "gpu/intel/jit/utils/utils.hpp"
 
 namespace dnnl {
@@ -441,7 +440,7 @@ class walk_order_param_t : public value_param_t<walk_order_t> {
 
     void set_from_str(const std::string &s) override {
         if (s.empty()) return;
-        value_ = walk_order_t(s);
+        value_ = make_walk_order(s);
     }
 
     std::string str() const override {

src/gpu/intel/conv/jit/ir_builder.cpp

@@ -27,11 +27,9 @@
 #include "gpu/intel/conv/jit/pipeline.hpp"
 #include "gpu/intel/conv/jit/plan.hpp"
 #include "gpu/intel/jit/ir/epilogue.hpp"
-#include "gpu/intel/jit/ir/fma.hpp"
 #include "gpu/intel/jit/ir/gemm_schedule.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/ir/post_ops.hpp"
-#include "gpu/intel/jit/ir/reduce.hpp"
-#include "gpu/intel/jit/ir/reorder.hpp"
 #include "gpu/intel/jit/ir/send_builder.hpp"
 #include "gpu/intel/jit/ir/slm_reduce_builder.hpp"
 #include "gpu/intel/jit/ir/tensor.hpp"

src/gpu/intel/conv/jit/pipeline.cpp

@@ -16,7 +16,7 @@
 
 #include "gpu/intel/conv/jit/pipeline.hpp"
 
-#include "gpu/intel/jit/ir/send.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/pass/alloc.hpp"
 
 namespace dnnl {

src/gpu/intel/conv/jit/plan.cpp

@@ -18,8 +18,8 @@
 
 #include "gpu/intel/conv/jit/config.hpp"
 #include "gpu/intel/jit/ir/gemm_schedule.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/ir/reduce.hpp"
-#include "gpu/intel/jit/ir/reorder.hpp"
 #include "gpu/intel/jit/ir/send_builder.hpp"
 #include "gpu/intel/jit/ir/send_plan.hpp"
 #include "gpu/intel/jit/ir/tensor.hpp"

src/gpu/intel/conv/jit/plan.hpp

@@ -22,8 +22,8 @@
 #include "gpu/intel/conv/jit/plan_utils.hpp"
 #include "gpu/intel/conv/jit/zp_plan.hpp"
 #include "gpu/intel/jit/grf_usage.hpp"
-#include "gpu/intel/jit/ir/fma.hpp"
 #include "gpu/intel/jit/ir/gemm_schedule.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/ir/send_plan.hpp"
 #include "gpu/intel/jit/ir/tensor.hpp"
 #include "gpu/intel/jit/utils/utils.hpp"

src/gpu/intel/conv/jit/problem.cpp

@@ -16,8 +16,8 @@
 
 #include "gpu/intel/conv/jit/problem.hpp"
 #include "gpu/intel/jit/ir/block_2d_utils.hpp"
-#include "gpu/intel/jit/ir/fma.hpp"
 #include "gpu/intel/jit/ir/hw.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/ir/tensor_config.hpp"
 
 namespace dnnl {

src/gpu/intel/conv/jit/v2/builder.cpp

@@ -17,6 +17,7 @@
 #include "gpu/intel/conv/jit/v2/builder.hpp"
 
 #include "gemmstone/../../dsl/ir/ir.hpp"
+#include "gemmstone/../../dsl/ir/pass/dpas.hpp"
 #include "gemmstone/../../dsl/ir/pass/trace.hpp"
 #include "gpu/intel/conv/jit/v2/bridge.hpp"
 #include "gpu/intel/conv/jit/v2/plan.hpp"
@@ -25,7 +26,6 @@
 #include "gpu/intel/jit/ir/post_ops.hpp"
 #include "gpu/intel/jit/ir/v2/builder.hpp"
 #include "gpu/intel/jit/ir/v2/send.hpp"
-#include "gpu/intel/jit/pass/dpas.hpp"
 #include "gpu/intel/jit/pass/pass.hpp"
 #include "gpu/intel/logging.hpp"
 
@@ -986,7 +986,7 @@ class builder_t : public v2::ir_builder_t {
     void emit_thread_index_let() {
         for (int i = 0; i < 3; i++) {
             auto value = var_t::make(
-                    dsl::type_t::u16(), jit::ir_builder_t::local_id(i));
+                    dsl::type_t::u16(), jit::ir::local_id_name(i));
             if (i == 0) value /= plan_.desc.simd;
             auto thr_idx = plan_.thr_grid.index_var(i);
             let(thr_idx, cast(value, thr_idx.type()));

src/gpu/intel/conv/jit/v2/kernel_desc.cpp

@@ -24,7 +24,6 @@
 #include "gpu/intel/conv/jit/v2/kernel.hpp"
 #include "gpu/intel/conv/jit/v2/problem.hpp"
 #include "gpu/intel/conv/jit/v2/tensor_utils.hpp"
-#include "gpu/intel/jit/codegen/kernel.hpp"
 #include "gpu/intel/jit/ir/config.hpp"
 #include "gpu/intel/jit/ir/kernel_info.hpp"
 #include "gpu/intel/jit/ir/tensor_config.hpp"
@@ -263,7 +262,7 @@ bool kernel_desc_t::is_supported(
             << "HW mismatch, desc: " << jit::to_string(hw_desc.hw)
             << ", problem: " << jit::to_string(prb->hw().ngen_hw());
     gpu_check(fma != fma_kind_t::undef)
-            << "Invalid fma: " << jit::to_string(fma);
+            << "Invalid fma: " << ir::to_string(fma);
     gpu_check(simd != 0) << "Invalid simd: " << simd;
     gpu_check(regs != 0) << "Invalid regs: " << regs;
     gpu_check(is_tg_size_ok(*this, hw))
@@ -1117,7 +1116,7 @@ void kernel_desc_t::show_help() {
 }
 
 grid_t create_thread_group_grid(const kernel_desc_t &desc) {
-    grid_t grid(jit::ir_builder_t::tg_idx);
+    grid_t grid(jit::ir::tg_idx_name);
     auto set = [&](const pvar_t &dim, int idx) {
         grid.add_mapping(dim, desc.use_stream_k ? 0 : idx);
     };
@@ -1152,7 +1151,7 @@ grid_t create_thread_group_grid(const kernel_desc_t &desc) {
 }
 
 grid_t create_thread_grid(const kernel_desc_t &desc) {
-    grid_t grid(jit::ir_builder_t::thr_idx);
+    grid_t grid(jit::ir::thr_idx_name);
     switch (desc.prop) {
         case prop_kind::forward:
             grid.add_mapping(pvars::oc, 0);

src/gpu/intel/conv/jit/v2/kernel_desc.hpp

@@ -20,10 +20,10 @@
 #include "gpu/intel/compute/kernel.hpp"
 #include "gpu/intel/conv/jit/v2/primitive_plan.hpp"
 #include "gpu/intel/conv/jit/v2/problem.hpp"
-#include "gpu/intel/jit/ir/fma.hpp"
 #include "gpu/intel/jit/ir/hw.hpp"
 #include "gpu/intel/jit/ir/kernel_desc.hpp"
 #include "gpu/intel/jit/ir/kernel_info.hpp"
+#include "gpu/intel/jit/ir/legacy.hpp"
 #include "gpu/intel/jit/ir/v2/reqs.hpp"
 #include "gpu/intel/jit/ir/v2/send.hpp"
 #include "gpu/intel/jit/ir/v2/tensor.hpp"
@@ -396,7 +396,7 @@ class grid_t {
     static const int N = 3;
 
     grid_t() = default;
-    grid_t(std::string (*genname)(int)) {
+    grid_t(const std::string &(*genname)(int)) {
         for (int i = 0; i < N; i++)
             entries_[i].idx_var = var_t::make(dsl::type_t::s32(), genname(i));
     }