From fcb80803cca12d631b31ebe4ecd77bda5004955d Mon Sep 17 00:00:00 2001
From: zhaojiale <zhaojiale@loongson.cn>
Date: Mon, 20 May 2024 10:22:21 +0800
Subject: [PATCH] llvmpipe: add an implementation with llvm orcjit

Signed-off-by: Yukari Chiba <i@0x7f.cc>

---
 meson.build                                   |   4 +-
 meson_options.txt                             |   7 +
 src/gallium/auxiliary/draw/draw_context.c     |   4 +
 src/gallium/auxiliary/draw/draw_llvm.c        |  68 +-
 src/gallium/auxiliary/draw/draw_llvm.h        |  17 +-
 src/gallium/auxiliary/gallivm/lp_bld_coro.c   |   6 +-
 src/gallium/auxiliary/gallivm/lp_bld_coro.h   |   8 +
 src/gallium/auxiliary/gallivm/lp_bld_init.c   |   9 +-
 src/gallium/auxiliary/gallivm/lp_bld_init.h   |  35 +-
 .../auxiliary/gallivm/lp_bld_init_orc.cpp     | 959 ++++++++++++++++++
 src/gallium/auxiliary/meson.build             |   6 +-
 src/gallium/drivers/llvmpipe/lp_context.c     |  18 +
 src/gallium/drivers/llvmpipe/lp_context.h     |   4 +
 src/gallium/drivers/llvmpipe/lp_screen.c      |   4 +
 src/gallium/drivers/llvmpipe/lp_state_cs.c    |  14 +-
 src/gallium/drivers/llvmpipe/lp_state_cs.h    |   1 +
 src/gallium/drivers/llvmpipe/lp_state_fs.c    |  25 +-
 src/gallium/drivers/llvmpipe/lp_state_fs.h    |   2 +
 .../llvmpipe/lp_state_fs_linear_llvm.c        |   2 +
 src/gallium/drivers/llvmpipe/lp_state_setup.c |   4 +-
 src/gallium/drivers/llvmpipe/lp_state_setup.h |   1 +
 src/gallium/drivers/llvmpipe/lp_test_arit.c   |  18 +-
 src/gallium/drivers/llvmpipe/lp_test_blend.c  |  18 +-
 src/gallium/drivers/llvmpipe/lp_test_conv.c   |  18 +-
 src/gallium/drivers/llvmpipe/lp_test_format.c |  46 +-
 .../llvmpipe/lp_test_lookup_multiple.c        | 167 +++
 src/gallium/drivers/llvmpipe/lp_test_printf.c |  18 +-
 .../drivers/llvmpipe/lp_texture_handle.c      |  11 +-
 src/gallium/drivers/llvmpipe/meson.build      |   2 +-
 src/util/detect_arch.h                        |  23 +
 30 files changed, 1479 insertions(+), 40 deletions(-)
 create mode 100644 src/gallium/auxiliary/gallivm/lp_bld_init_orc.cpp
 create mode 100644 src/gallium/drivers/llvmpipe/lp_test_lookup_multiple.c

diff --git a/meson.build b/meson.build
index 5c91048..444c7d2 100644
--- a/meson.build
+++ b/meson.build
@@ -1725,13 +1725,14 @@ if with_clc
   llvm_optional_modules += ['all-targets', 'windowsdriver', 'frontendhlsl']
 endif
 draw_with_llvm = get_option('draw-use-llvm')
+llvm_with_orcjit = get_option('llvm-orcjit')
 if draw_with_llvm
   llvm_modules += 'native'
   # lto is needded with LLVM>=15, but we don't know what LLVM verrsion we are using yet
   llvm_optional_modules += ['lto']
 endif
 
-if with_amd_vk or with_gallium_radeonsi
+if with_amd_vk or with_gallium_radeonsi or llvm_with_orcjit
   _llvm_version = '>= 15.0.0'
 elif with_intel_clc
   _llvm_version = '>= 13.0.0'
@@ -1815,6 +1816,7 @@ else
 endif
 pre_args += '-DLLVM_AVAILABLE=' + (with_llvm ? '1' : '0')
 pre_args += '-DDRAW_LLVM_AVAILABLE=' + (with_llvm and draw_with_llvm ? '1' : '0')
+pre_args += '-DGALLIVM_USE_ORCJIT=' + (with_llvm and llvm_with_orcjit ? '1' : '0')
 
 with_opencl_spirv = (_opencl != 'disabled' and get_option('opencl-spirv')) or with_clc
 if with_opencl_spirv
diff --git a/meson_options.txt b/meson_options.txt
index 92ffd1d..38aae94 100644
--- a/meson_options.txt
+++ b/meson_options.txt
@@ -416,6 +416,13 @@ option(
                 'is included.'
 )
 
+option (
+  'llvm-orcjit',
+  type : 'boolean',
+  value : false,
+  description: 'Build llvmpipe with LLVM ORCJIT support.'
+)
+
 option(
   'valgrind',
   type : 'feature',
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 58e8ace..83ce386 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -77,7 +77,11 @@ draw_create_context(struct pipe_context *pipe, void *context,
 
 #if DRAW_LLVM_AVAILABLE
    if (try_llvm && draw_get_option_use_llvm()) {
+#if GALLIVM_USE_ORCJIT == 1
+      draw->llvm = draw_llvm_create(draw, (LLVMOrcThreadSafeContextRef)context);
+#else
       draw->llvm = draw_llvm_create(draw, (LLVMContextRef)context);
+#endif
    }
 #endif
 
diff --git a/src/gallium/auxiliary/draw/draw_llvm.c b/src/gallium/auxiliary/draw/draw_llvm.c
index 36ff317..7d09d65 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_llvm.c
@@ -381,8 +381,13 @@ get_vertex_header_ptr_type(struct draw_llvm_variant *variant)
 /**
  * Create per-context LLVM info.
  */
+#if GALLIVM_USE_ORCJIT == 1
+struct draw_llvm *
+draw_llvm_create(struct draw_context *draw, LLVMOrcThreadSafeContextRef context)
+#else
 struct draw_llvm *
 draw_llvm_create(struct draw_context *draw, LLVMContextRef context)
+#endif
 {
    struct draw_llvm *llvm;
 
@@ -395,6 +400,16 @@ draw_llvm_create(struct draw_context *draw, LLVMContextRef context)
 
    llvm->draw = draw;
 
+#if GALLIVM_USE_ORCJIT == 1
+   llvm->_ts_context = context;
+   if (!llvm->_ts_context) {
+      llvm->_ts_context = LLVMOrcCreateNewThreadSafeContext();
+      llvm->context_owned = true;
+   }
+   if (!llvm->_ts_context)
+      goto fail;
+   llvm->context = LLVMOrcThreadSafeContextGetContext(llvm->_ts_context);
+#else
    llvm->context = context;
    if (!llvm->context) {
       llvm->context = LLVMContextCreate();
@@ -407,6 +422,7 @@ draw_llvm_create(struct draw_context *draw, LLVMContextRef context)
    }
    if (!llvm->context)
       goto fail;
+#endif
 
    llvm->nr_variants = 0;
    list_inithead(&llvm->vs_variants_list.list);
@@ -434,9 +450,16 @@ fail:
 void
 draw_llvm_destroy(struct draw_llvm *llvm)
 {
+#if GALLIVM_USE_ORCJIT == 1
+   if (llvm->context_owned)
+      LLVMOrcDisposeThreadSafeContext(llvm->_ts_context);
+   llvm->_ts_context = NULL;
+   llvm->context = NULL;
+#else
    if (llvm->context_owned)
       LLVMContextDispose(llvm->context);
    llvm->context = NULL;
+#endif
 
    /* XXX free other draw_llvm data? */
    FREE(llvm);
@@ -510,7 +533,11 @@ draw_llvm_create_variant(struct draw_llvm *llvm,
       if (!cached.data_size)
          needs_caching = true;
    }
+#if GALLIVM_USE_ORCJIT == 1
+   variant->gallivm = gallivm_create(module_name, llvm->_ts_context, &cached);
+#else
    variant->gallivm = gallivm_create(module_name, llvm->context, &cached);
+#endif
 
    create_vs_jit_types(variant);
 
@@ -530,7 +557,8 @@ draw_llvm_create_variant(struct draw_llvm *llvm,
    gallivm_compile_module(variant->gallivm);
 
    variant->jit_func = (draw_jit_vert_func)
-         gallivm_jit_function(variant->gallivm, variant->function);
+         gallivm_jit_function(variant->gallivm, variant->function,
+                              variant->function_name);
 
    if (needs_caching)
       llvm->draw->disk_cache_insert_shader(llvm->draw->disk_cache_cookie,
@@ -1633,6 +1661,8 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 
    variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);
    variant->function = variant_func;
+   variant->function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name, func_name);
 
    LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
    for (i = 0; i < num_arg_types; ++i)
@@ -2249,6 +2279,8 @@ draw_llvm_destroy_variant(struct draw_llvm_variant *variant)
    variant->shader->variants_cached--;
    list_del(&variant->list_item_global.list);
    llvm->nr_variants--;
+   if(variant->function_name)
+      FREE(variant->function_name);
    FREE(variant);
 }
 
@@ -2358,6 +2390,8 @@ draw_gs_llvm_generate(struct draw_llvm *llvm,
    variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);
 
    variant->function = variant_func;
+   variant->function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name, func_name);
 
    LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
 
@@ -2524,7 +2558,11 @@ draw_gs_llvm_create_variant(struct draw_llvm *llvm,
       if (!cached.data_size)
          needs_caching = true;
    }
+#if GALLIVM_USE_ORCJIT == 1
+   variant->gallivm = gallivm_create(module_name, llvm->_ts_context, &cached);
+#else
    variant->gallivm = gallivm_create(module_name, llvm->context, &cached);
+#endif
 
    create_gs_jit_types(variant);
 
@@ -2536,7 +2574,8 @@ draw_gs_llvm_create_variant(struct draw_llvm *llvm,
    gallivm_compile_module(variant->gallivm);
 
    variant->jit_func = (draw_gs_jit_func)
-         gallivm_jit_function(variant->gallivm, variant->function);
+         gallivm_jit_function(variant->gallivm, variant->function,
+                              variant->function_name);
 
    if (needs_caching)
       llvm->draw->disk_cache_insert_shader(llvm->draw->disk_cache_cookie,
@@ -2569,6 +2608,8 @@ draw_gs_llvm_destroy_variant(struct draw_gs_llvm_variant *variant)
    variant->shader->variants_cached--;
    list_del(&variant->list_item_global.list);
    llvm->nr_gs_variants--;
+   if(variant->function_name)
+      FREE(variant->function_name);
    FREE(variant);
 }
 
@@ -2942,6 +2983,8 @@ draw_tcs_llvm_generate(struct draw_llvm *llvm,
    variant_coro = LLVMAddFunction(gallivm->module, func_name_coro, coro_func_type);
 
    variant->function = variant_func;
+   variant->function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name, func_name);
    LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
 
    LLVMSetFunctionCallConv(variant_coro, LLVMCCallConv);
@@ -3177,8 +3220,11 @@ draw_tcs_llvm_create_variant(struct draw_llvm *llvm,
       if (!cached.data_size)
          needs_caching = true;
    }
-
+#if GALLIVM_USE_ORCJIT == 1
+   variant->gallivm = gallivm_create(module_name, llvm->_ts_context, &cached);
+#else
    variant->gallivm = gallivm_create(module_name, llvm->context, &cached);
+#endif
 
    create_tcs_jit_types(variant);
 
@@ -3192,7 +3238,8 @@ draw_tcs_llvm_create_variant(struct draw_llvm *llvm,
    gallivm_compile_module(variant->gallivm);
 
    variant->jit_func = (draw_tcs_jit_func)
-      gallivm_jit_function(variant->gallivm, variant->function);
+         gallivm_jit_function(variant->gallivm, variant->function,
+                              variant->function_name);
 
    if (needs_caching)
       llvm->draw->disk_cache_insert_shader(llvm->draw->disk_cache_cookie,
@@ -3225,6 +3272,8 @@ draw_tcs_llvm_destroy_variant(struct draw_tcs_llvm_variant *variant)
    variant->shader->variants_cached--;
    list_del(&variant->list_item_global.list);
    llvm->nr_tcs_variants--;
+   if(variant->function_name)
+      FREE(variant->function_name);
    FREE(variant);
 }
 
@@ -3507,6 +3556,8 @@ draw_tes_llvm_generate(struct draw_llvm *llvm,
    variant_func = LLVMAddFunction(gallivm->module, func_name, func_type);
 
    variant->function = variant_func;
+   variant->function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name, func_name);
    LLVMSetFunctionCallConv(variant_func, LLVMCCallConv);
 
    for (i = 0; i < ARRAY_SIZE(arg_types); ++i)
@@ -3697,7 +3748,11 @@ draw_tes_llvm_create_variant(struct draw_llvm *llvm,
       if (!cached.data_size)
          needs_caching = true;
    }
+#if GALLIVM_USE_ORCJIT == 1
+   variant->gallivm = gallivm_create(module_name, llvm->_ts_context, &cached);
+#else
    variant->gallivm = gallivm_create(module_name, llvm->context, &cached);
+#endif
 
    create_tes_jit_types(variant);
 
@@ -3714,7 +3769,8 @@ draw_tes_llvm_create_variant(struct draw_llvm *llvm,
    gallivm_compile_module(variant->gallivm);
 
    variant->jit_func = (draw_tes_jit_func)
-      gallivm_jit_function(variant->gallivm, variant->function);
+         gallivm_jit_function(variant->gallivm, variant->function,
+                              variant->function_name);
 
    if (needs_caching)
       llvm->draw->disk_cache_insert_shader(llvm->draw->disk_cache_cookie,
@@ -3747,6 +3803,8 @@ draw_tes_llvm_destroy_variant(struct draw_tes_llvm_variant *variant)
    variant->shader->variants_cached--;
    list_del(&variant->list_item_global.list);
    llvm->nr_tes_variants--;
+   if(variant->function_name)
+      FREE(variant->function_name);
    FREE(variant);
 }
 
diff --git a/src/gallium/auxiliary/draw/draw_llvm.h b/src/gallium/auxiliary/draw/draw_llvm.h
index 65199fe..2d87a95 100644
--- a/src/gallium/auxiliary/draw/draw_llvm.h
+++ b/src/gallium/auxiliary/draw/draw_llvm.h
@@ -42,6 +42,9 @@
 #include "pipe/p_context.h"
 #include "util/list.h"
 
+#if GALLIVM_USE_ORCJIT == 1
+#include <llvm-c/Orc.h>
+#endif
 
 struct draw_llvm;
 struct llvm_vertex_shader;
@@ -398,6 +401,7 @@ struct draw_llvm_variant
    LLVMTypeRef vertex_header_ptr_type;
 
    LLVMValueRef function;
+   char *function_name;
    draw_jit_vert_func jit_func;
 
    struct llvm_vertex_shader *shader;
@@ -431,6 +435,7 @@ struct draw_gs_llvm_variant
    LLVMValueRef io_ptr;
    LLVMValueRef num_prims;
    LLVMValueRef function;
+   char *function_name;
    draw_gs_jit_func jit_func;
 
    struct llvm_geometry_shader *shader;
@@ -457,6 +462,7 @@ struct draw_tcs_llvm_variant
    /* LLVMValueRef io_ptr; */
    LLVMValueRef num_prims;
    LLVMValueRef function;
+   char *function_name;
    draw_tcs_jit_func jit_func;
 
    struct llvm_tess_ctrl_shader *shader;
@@ -487,6 +493,7 @@ struct draw_tes_llvm_variant
    LLVMValueRef io_ptr;
    LLVMValueRef num_prims;
    LLVMValueRef function;
+   char *function_name;
    draw_tes_jit_func jit_func;
 
    struct llvm_tess_eval_shader *shader;
@@ -539,6 +546,9 @@ struct draw_llvm {
    struct draw_context *draw;
 
    LLVMContextRef context;
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef _ts_context;
+#endif
    bool context_owned;
 
    struct draw_vs_jit_context vs_jit_context;
@@ -584,8 +594,13 @@ llvm_tess_eval_shader(struct draw_tess_eval_shader *tes)
    return (struct llvm_tess_eval_shader *)tes;
 }
 
+#if GALLIVM_USE_ORCJIT == 1
 struct draw_llvm *
-draw_llvm_create(struct draw_context *draw, LLVMContextRef llvm_context);
+draw_llvm_create(struct draw_context *draw, LLVMOrcThreadSafeContextRef context);
+#else
+struct draw_llvm *
+draw_llvm_create(struct draw_context *draw, LLVMContextRef context);
+#endif
 
 void
 draw_llvm_destroy(struct draw_llvm *llvm);
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_coro.c b/src/gallium/auxiliary/gallivm/lp_bld_coro.c
index 0214dcf..f5a2b31 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_coro.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_coro.c
@@ -156,12 +156,14 @@ coro_free(char *ptr)
 
 void lp_build_coro_add_malloc_hooks(struct gallivm_state *gallivm)
 {
+#if GALLIVM_USE_ORCJIT == 0
    assert(gallivm->engine);
+#endif
 
    assert(gallivm->coro_malloc_hook);
    assert(gallivm->coro_free_hook);
-   LLVMAddGlobalMapping(gallivm->engine, gallivm->coro_malloc_hook, coro_malloc);
-   LLVMAddGlobalMapping(gallivm->engine, gallivm->coro_free_hook, coro_free);
+   gallivm_add_global_mapping(gallivm, gallivm->coro_malloc_hook, coro_malloc);
+   gallivm_add_global_mapping(gallivm, gallivm->coro_free_hook, coro_free);
 }
 
 void lp_build_coro_declare_malloc_hooks(struct gallivm_state *gallivm)
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_coro.h b/src/gallium/auxiliary/gallivm/lp_bld_coro.h
index 2fbaecc..5421d09 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_coro.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_coro.h
@@ -31,6 +31,10 @@
 #include "gallivm/lp_bld.h"
 #include "gallivm/lp_bld_intr.h"
 
+#ifdef __cplusplus
+extern "C" {
+#endif
+
 struct gallivm_state;
 LLVMValueRef lp_build_coro_id(struct gallivm_state *gallivm);
 
@@ -84,4 +88,8 @@ static inline void lp_build_coro_add_presplit(LLVMValueRef coro)
 #endif
 }
 
+#ifdef __cplusplus
+}
+#endif
+
 #endif
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init.c b/src/gallium/auxiliary/gallivm/lp_bld_init.c
index cd2108f..9750cf7 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_init.c
+++ b/src/gallium/auxiliary/gallivm/lp_bld_init.c
@@ -515,6 +515,11 @@ gallivm_destroy(struct gallivm_state *gallivm)
    FREE(gallivm);
 }
 
+void
+gallivm_add_global_mapping(struct gallivm_state *gallivm, LLVMValueRef sym, void* addr)
+{
+   LLVMAddGlobalMapping(gallivm->engine, sym, addr);
+}
 
 /**
  * Validate a function.
@@ -670,7 +675,7 @@ gallivm_compile_module(struct gallivm_state *gallivm)
    ++gallivm->compiled;
 
    lp_init_printf_hook(gallivm);
-   LLVMAddGlobalMapping(gallivm->engine, gallivm->debug_printf_hook, debug_printf);
+   gallivm_add_global_mapping(gallivm, gallivm->debug_printf_hook, debug_printf);
 
    lp_init_clock_hook(gallivm);
    LLVMAddGlobalMapping(gallivm->engine, gallivm->get_time_hook, os_time_get_nano);
@@ -713,7 +718,7 @@ gallivm_compile_module(struct gallivm_state *gallivm)
 
 func_pointer
 gallivm_jit_function(struct gallivm_state *gallivm,
-                     LLVMValueRef func)
+                     LLVMValueRef func, const char *func_name)
 {
    void *code;
    func_pointer jit_func;
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init.h b/src/gallium/auxiliary/gallivm/lp_bld_init.h
index 418921c..f425ebb 100644
--- a/src/gallium/auxiliary/gallivm/lp_bld_init.h
+++ b/src/gallium/auxiliary/gallivm/lp_bld_init.h
@@ -29,11 +29,14 @@
 #ifndef LP_BLD_INIT_H
 #define LP_BLD_INIT_H
 
-
 #include "util/compiler.h"
 #include "util/u_pointer.h" // for func_pointer
 #include "lp_bld.h"
+#if GALLIVM_USE_ORCJIT == 1
+#include <llvm-c/Orc.h>
+#else
 #include <llvm-c/ExecutionEngine.h>
+#endif
 
 #ifdef __cplusplus
 extern "C" {
@@ -44,18 +47,25 @@ struct gallivm_state
 {
    char *module_name;
    LLVMModuleRef module;
-   LLVMExecutionEngineRef engine;
    LLVMTargetDataRef target;
+#if GALLIVM_USE_ORCJIT == 1
+   /* own this->module */
+   LLVMOrcThreadSafeContextRef _ts_context;
+   /* each module is in its own jitdylib */
+   LLVMOrcJITDylibRef _per_module_jd;
+#else
+   LLVMExecutionEngineRef engine;
 #if GALLIVM_USE_NEW_PASS == 0
    LLVMPassManagerRef passmgr;
 #if GALLIVM_HAVE_CORO == 1
    LLVMPassManagerRef cgpassmgr;
 #endif
 #endif
-   LLVMContextRef context;
-   LLVMBuilderRef builder;
    LLVMMCJITMemoryManagerRef memorymgr;
    struct lp_generated_code *code;
+#endif
+   LLVMContextRef context;
+   LLVMBuilderRef builder;
    struct lp_cached_code *cache;
    unsigned compiled;
    LLVMValueRef coro_malloc_hook;
@@ -77,10 +87,15 @@ lp_build_init_native_width(void);
 bool
 lp_build_init(void);
 
-
+#if GALLIVM_USE_ORCJIT == 1
+struct gallivm_state *
+gallivm_create(const char *name, LLVMOrcThreadSafeContextRef context,
+               struct lp_cached_code *cache);
+#else
 struct gallivm_state *
 gallivm_create(const char *name, LLVMContextRef context,
                struct lp_cached_code *cache);
+#endif
 
 void
 gallivm_destroy(struct gallivm_state *gallivm);
@@ -93,11 +108,19 @@ gallivm_verify_function(struct gallivm_state *gallivm,
                         LLVMValueRef func);
 
 void
+gallivm_add_global_mapping(struct gallivm_state *gallivm, LLVMValueRef sym, void* addr);
+
+/**
+ * for ORCJIT, after this function gets called, all access and modification to
+ * module and any structure associated to it should be avoided,
+ * as module has been moved into ORCJIT and may be recycled
+ */
+void
 gallivm_compile_module(struct gallivm_state *gallivm);
 
 func_pointer
 gallivm_jit_function(struct gallivm_state *gallivm,
-                     LLVMValueRef func);
+                     LLVMValueRef func, const char *func_name);
 
 unsigned gallivm_get_perf_flags(void);
 
diff --git a/src/gallium/auxiliary/gallivm/lp_bld_init_orc.cpp b/src/gallium/auxiliary/gallivm/lp_bld_init_orc.cpp
new file mode 100644
index 0000000..797c9e1
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/lp_bld_init_orc.cpp
@@ -0,0 +1,959 @@
+#include "util/detect.h"
+#include "util/u_cpu_detect.h"
+#include "util/u_debug.h"
+#include "util/os_time.h"
+#include "lp_bld.h"
+#include "lp_bld_debug.h"
+#include "lp_bld_init.h"
+#include "lp_bld_coro.h"
+#include "lp_bld_printf.h"
+
+#include <llvm/Config/llvm-config.h>
+#include <llvm-c/Core.h>
+#include <llvm-c/Orc.h>
+#include <llvm-c/LLJIT.h>
+#include <llvm-c/TargetMachine.h>
+#include <llvm-c/Support.h>
+
+#include <llvm-c/Analysis.h>
+#if LLVM_VERSION_MAJOR < 17
+#include <llvm-c/Transforms/Scalar.h>
+#if LLVM_VERSION_MAJOR >= 7
+#include <llvm-c/Transforms/Utils.h>
+#endif
+#endif
+#include <llvm-c/BitWriter.h>
+#if GALLIVM_USE_NEW_PASS == 1
+#include <llvm-c/Transforms/PassBuilder.h>
+#elif GALLIVM_HAVE_CORO == 1
+#if LLVM_VERSION_MAJOR <= 8 && (DETECT_ARCH_AARCH64 == 1 || DETECT_ARCH_ARM == 1 || DETECT_ARCH_S390 == 1 || DETECT_ARCH_MIPS64 == 1)
+#include <llvm-c/Transforms/IPO.h>
+#endif
+#include <llvm-c/Transforms/Coroutines.h>
+#endif
+
+#include <llvm/ADT/StringMap.h>
+#include <llvm/ExecutionEngine/Orc/LLJIT.h>
+#include <llvm/ExecutionEngine/Orc/ObjectLinkingLayer.h>
+#include "llvm/ExecutionEngine/JITLink/JITLink.h"
+#include <llvm/Target/TargetMachine.h>
+#include <llvm/Support/TargetSelect.h>
+#include <llvm/Support/Casting.h>
+#if LLVM_VERSION_MAJOR >= 18
+#include <llvm/TargetParser/Host.h>
+#else
+#include <llvm/Support/Host.h>
+#endif
+#include <llvm/Support/CBindingWrapping.h>
+#if LLVM_USE_INTEL_JITEVENTS
+#include <llvm/ExecutionEngine/JITEventListener.h>
+#endif
+
+/* conflict with ObjectLinkingLayer.h */
+#include "util/u_memory.h"
+
+#if DETECT_ARCH_RISCV64 == 1 || DETECT_ARCH_RISCV32 == 1 || (defined(_WIN32) && LLVM_VERSION_MAJOR >= 15)
+/* use ObjectLinkingLayer (JITLINK backend) */
+#define USE_JITLINK
+#endif
+/* else use old RTDyldObjectLinkingLayer (RuntimeDyld backend) */
+
+unsigned lp_native_vector_width;
+
+unsigned gallivm_perf = 0;
+
+static const struct debug_named_value lp_bld_perf_flags[] = {
+   { "brilinear", GALLIVM_PERF_BRILINEAR, "enable brilinear optimization" },
+   { "rho_approx", GALLIVM_PERF_RHO_APPROX, "enable rho_approx optimization" },
+   { "no_quad_lod", GALLIVM_PERF_NO_QUAD_LOD, "disable quad_lod optimization" },
+   { "no_aos_sampling", GALLIVM_PERF_NO_AOS_SAMPLING, "disable aos sampling optimization" },
+   { "nopt",   GALLIVM_PERF_NO_OPT, "disable optimization passes to speed up shader compilation" },
+   DEBUG_NAMED_VALUE_END
+};
+
+unsigned gallivm_debug = 0;
+
+static const struct debug_named_value lp_bld_debug_flags[] = {
+   { "tgsi",   GALLIVM_DEBUG_TGSI, NULL },
+   { "ir",     GALLIVM_DEBUG_IR, NULL },
+   { "asm",    GALLIVM_DEBUG_ASM, NULL },
+   { "perf",   GALLIVM_DEBUG_PERF, NULL },
+   { "gc",     GALLIVM_DEBUG_GC, NULL },
+/* Don't allow setting DUMP_BC for release builds, since writing the files may be an issue with setuid. */
+#ifdef DEBUG
+   { "dumpbc", GALLIVM_DEBUG_DUMP_BC, NULL },
+#endif
+   DEBUG_NAMED_VALUE_END
+};
+
+DEBUG_GET_ONCE_FLAGS_OPTION(gallivm_debug, "GALLIVM_DEBUG", lp_bld_debug_flags, 0)
+
+struct lp_cached_code {
+   void *data;
+   size_t data_size;
+   bool dont_cache;
+   void *jit_obj_cache;
+};
+
+namespace {
+
+class LPJit;
+
+class LLVMEnsureMultithreaded {
+public:
+   LLVMEnsureMultithreaded()
+   {
+      if (!LLVMIsMultithreaded()) {
+         LLVMStartMultithreaded();
+      }
+   }
+};
+
+LLVMEnsureMultithreaded lLVMEnsureMultithreaded;
+
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(llvm::orc::ThreadSafeContext,
+                                   LLVMOrcThreadSafeContextRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(llvm::orc::IRTransformLayer,
+                                   LLVMOrcIRTransformLayerRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(llvm::orc::JITDylib, LLVMOrcJITDylibRef)
+DEFINE_SIMPLE_CONVERSION_FUNCTIONS(llvm::orc::JITTargetMachineBuilder,
+                                   LLVMOrcJITTargetMachineBuilderRef)
+LLVMTargetMachineRef wrap(const llvm::TargetMachine *P) {
+   return reinterpret_cast<LLVMTargetMachineRef>(const_cast<llvm::TargetMachine*>(P));
+}
+
+llvm::ExitOnError ExitOnErr;
+
+inline const char* get_module_name(LLVMModuleRef mod) {
+   using llvm::Module;
+   return llvm::unwrap(mod)->getModuleIdentifier().c_str();
+}
+
+once_flag init_lpjit_once_flag = ONCE_FLAG_INIT;
+
+/* A JIT singleton built upon LLJIT */
+class LPJit
+{
+public:
+   static LPJit* get_instance() {
+      call_once(&init_lpjit_once_flag, init_lpjit);
+      return jit;
+   }
+
+   gallivm_state *find_gallivm_state(LLVMModuleRef mod) {
+#if DEBUG
+      using llvm::Module;
+      auto I = gallivm_modules.find(llvm::unwrap(mod)->getModuleIdentifier());
+      if (I == gallivm_modules.end()) {
+         debug_printf("No gallivm state found for module: %s", get_module_name(mod));
+         return NULL;
+      }
+      return I->second;
+#endif
+      return NULL;
+   }
+
+   static char *get_unique_name(const char* name) {
+      LPJit* jit = get_instance();
+      size_t size = name == NULL? 16: strlen(name) + 16;
+      char *name_uniq = (char *)MALLOC(size);
+      if (!name_uniq) {
+         return NULL;
+      }
+      do {
+         snprintf(name_uniq, size, "%s_%u", name, jit->jit_dylib_count++);
+      } while(jit->lljit->getExecutionSession().getJITDylibByName(name_uniq));
+      return name_uniq;
+   }
+
+   static LLVMOrcJITDylibRef create_jit_dylib(const char * name) {
+      using llvm::orc::JITDylib;
+      LPJit* jit = get_instance();
+      JITDylib& tmp = ExitOnErr(jit->lljit->createJITDylib(name));
+      return wrap(&tmp);
+   }
+
+   static void register_gallivm_state(gallivm_state *gallivm) {
+#if DEBUG
+      LPJit* jit = get_instance();
+      jit->gallivm_modules[gallivm->module_name] = gallivm;
+#endif
+   }
+
+   static void deregister_gallivm_state(gallivm_state *gallivm) {
+#if DEBUG
+      LPJit* jit = get_instance();
+      (void)jit->gallivm_modules.erase(gallivm->module_name);
+#endif
+   }
+
+   static void add_ir_module_to_jd(
+         LLVMOrcThreadSafeContextRef ts_context,
+         LLVMModuleRef mod,
+         LLVMOrcJITDylibRef jd) {
+      using llvm::Module;
+      using llvm::orc::ThreadSafeModule;
+      using llvm::orc::JITDylib;
+      ThreadSafeModule tsm(
+         std::unique_ptr<Module>(llvm::unwrap(mod)), *::unwrap(ts_context));
+      ExitOnErr(get_instance()->lljit->addIRModule(
+         *::unwrap(jd), std::move(tsm)
+      ));
+   }
+
+   static void add_mapping_to_jd(
+         LLVMValueRef sym,
+         void *addr,
+         LLVMOrcJITDylibRef jd) {
+#if LLVM_VERSION_MAJOR >= 17
+      using llvm::orc::ExecutorAddr;
+      using llvm::orc::ExecutorSymbolDef;
+      using llvm::JITSymbolFlags;
+#else
+      using llvm::JITEvaluatedSymbol;
+#endif
+      using llvm::orc::ExecutionSession;
+      using llvm::orc::JITDylib;
+      using llvm::orc::SymbolMap;
+      JITDylib* JD = ::unwrap(jd);
+      auto& es = LPJit::get_instance()->lljit->getExecutionSession();
+      auto name = es.intern(llvm::unwrap(sym)->getName());
+      SymbolMap map(1);
+#if LLVM_VERSION_MAJOR >= 17
+      map[name] = ExecutorSymbolDef(ExecutorAddr::fromPtr(addr), JITSymbolFlags::Exported);
+#else
+      map[name] = JITEvaluatedSymbol::fromPointer(addr);
+#endif
+      auto munit = llvm::orc::absoluteSymbols(map);
+      llvm::cantFail(JD->define(std::move(munit)));
+   }
+
+   static void *lookup_in_jd(
+         const char *func_name,
+         LLVMOrcJITDylibRef jd) {
+      using llvm::orc::JITDylib;
+      using llvm::JITEvaluatedSymbol;
+      using llvm::orc::ExecutorAddr;
+      JITDylib* JD = ::unwrap(jd);
+      auto func = ExitOnErr(LPJit::get_instance()->lljit->lookup(*JD, func_name));
+#if LLVM_VERSION_MAJOR >= 16
+      return func.toPtr<void *>();
+#else
+      return (void *)(func.getAddress());
+#endif
+   }
+
+   static void remove_jd(LLVMOrcJITDylibRef jd) {
+      using llvm::orc::ExecutionSession;
+      using llvm::orc::JITDylib;
+      auto& es = LPJit::get_instance()->lljit->getExecutionSession();
+      ExitOnErr(es.removeJITDylib(* ::unwrap(jd)));
+   }
+
+   LLVMTargetMachineRef tm;
+
+private:
+   LPJit();
+   ~LPJit() = default;
+   LPJit(const LPJit&) = delete;
+   LPJit& operator=(const LPJit&) = delete;
+
+   static void init_native_targets();
+   llvm::orc::JITTargetMachineBuilder create_jtdb();
+
+   static void init_lpjit() {
+      jit = new LPJit;
+   }
+   static LPJit* jit;
+
+   std::unique_ptr<llvm::orc::LLJIT> lljit;
+   /* avoid name conflict */
+   unsigned jit_dylib_count;
+
+#if DEBUG
+   /* map from module name to gallivm_state */
+   llvm::StringMap<gallivm_state *> gallivm_modules;
+#endif
+};
+
+LPJit* LPJit::jit = NULL;
+
+LLVMErrorRef module_transform(void *Ctx, LLVMModuleRef mod) {
+   int64_t time_begin = 0;
+   if (::gallivm_debug & GALLIVM_DEBUG_PERF)
+      time_begin = os_time_get();
+#if GALLIVM_USE_NEW_PASS == 1
+   char passes[1024];
+   passes[0] = 0;
+
+   /*
+    * there should be some way to combine these two pass runs but I'm not seeing it,
+    * at the time of writing.
+    */
+   strcpy(passes, "default<O0>");
+
+   LLVMPassBuilderOptionsRef opts = LLVMCreatePassBuilderOptions();
+   LLVMRunPasses(mod, passes, LPJit::get_instance()->tm, opts);
+
+   if (!(gallivm_perf & GALLIVM_PERF_NO_OPT))
+      strcpy(passes, "sroa,early-cse,simplifycfg,reassociate,mem2reg,instsimplify,instcombine,");
+   else
+      strcpy(passes, "mem2reg");
+
+   LLVMRunPasses(mod, passes, LPJit::get_instance()->tm, opts);
+   LLVMDisposePassBuilderOptions(opts);
+
+   return LLVMErrorSuccess;
+
+#else /* GALLIVM_USE_NEW_PASS */
+   LLVMPassManagerRef passmgr = LLVMCreateFunctionPassManagerForModule(mod);
+
+#if GALLIVM_HAVE_CORO == 1
+   LLVMPassManagerRef cgpassmgr = LLVMCreatePassManager();
+#endif
+
+#if GALLIVM_HAVE_CORO == 1
+#if LLVM_VERSION_MAJOR <= 8 && (DETECT_ARCH_AARCH64 == 1 || DETECT_ARCH_ARM == 1 || DETECT_ARCH_S390 == 1 || DETECT_ARCH_MIPS64 == 1)
+   LLVMAddArgumentPromotionPass(cgpassmgr);
+   LLVMAddFunctionAttrsPass(cgpassmgr);
+#endif
+   LLVMAddCoroEarlyPass(cgpassmgr);
+   LLVMAddCoroSplitPass(cgpassmgr);
+   LLVMAddCoroElidePass(cgpassmgr);
+#endif
+
+   if ((gallivm_perf & GALLIVM_PERF_NO_OPT) == 0) {
+      /*
+       * TODO: Evaluate passes some more - keeping in mind
+       * both quality of generated code and compile times.
+       */
+      /*
+       * NOTE: if you change this, don't forget to change the output
+       * with GALLIVM_DEBUG_DUMP_BC in gallivm_compile_module.
+       */
+      LLVMAddScalarReplAggregatesPass(passmgr);
+      LLVMAddEarlyCSEPass(passmgr);
+      LLVMAddCFGSimplificationPass(passmgr);
+      /*
+       * FIXME: LICM is potentially quite useful. However, for some
+       * rather crazy shaders the compile time can reach _hours_ per shader,
+       * due to licm implying lcssa (since llvm 3.5), which can take forever.
+       * Even for sane shaders, the cost of licm is rather high (and not just
+       * due to lcssa, licm itself too), though mostly only in cases when it
+       * can actually move things, so having to disable it is a pity.
+       * LLVMAddLICMPass(passmgr);
+       */
+      LLVMAddReassociatePass(passmgr);
+      LLVMAddPromoteMemoryToRegisterPass(passmgr);
+#if LLVM_VERSION_MAJOR <= 11
+      LLVMAddConstantPropagationPass(passmgr);
+#else
+      LLVMAddInstructionSimplifyPass(passmgr);
+#endif
+      LLVMAddInstructionCombiningPass(passmgr);
+      LLVMAddGVNPass(passmgr);
+   }
+   else {
+      /* We need at least this pass to prevent the backends to fail in
+       * unexpected ways.
+       */
+      LLVMAddPromoteMemoryToRegisterPass(passmgr);
+   }
+#if GALLIVM_HAVE_CORO == 1
+   LLVMAddCoroCleanupPass(passmgr);
+
+   LLVMRunPassManager(cgpassmgr, mod);
+#endif
+   /* Run optimization passes */
+   LLVMInitializeFunctionPassManager(passmgr);
+   LLVMValueRef func;
+   func = LLVMGetFirstFunction(mod);
+   while (func) {
+      if (0) {
+         debug_printf("optimizing func %s...\n", LLVMGetValueName(func));
+      }
+
+   /* Disable frame pointer omission on debug/profile builds */
+   /* XXX: And workaround http://llvm.org/PR21435 */
+#if defined(DEBUG) || defined(PROFILE) || (DETECT_ARCH_X86 == 1 || DETECT_ARCH_X86_64 == 1)
+      LLVMAddTargetDependentFunctionAttr(func, "no-frame-pointer-elim", "true");
+      LLVMAddTargetDependentFunctionAttr(func, "no-frame-pointer-elim-non-leaf", "true");
+#endif
+
+      LLVMRunFunctionPassManager(passmgr, func);
+      func = LLVMGetNextFunction(func);
+   }
+   LLVMFinalizeFunctionPassManager(passmgr);
+   if (gallivm_debug & GALLIVM_DEBUG_PERF) {
+      int64_t time_end = os_time_get();
+      int time_msec = (int)((time_end - time_begin) / 1000);
+      
+      const char *module_name = get_module_name(mod);
+      debug_printf("optimizing module %s took %d msec\n",
+                   module_name, time_msec);
+   }
+
+#if GALLIVM_HAVE_CORO == 1
+   LLVMDisposePassManager(cgpassmgr);
+#endif
+   LLVMDisposePassManager(passmgr);
+   return LLVMErrorSuccess;
+#endif /* GALLIVM_USE_NEW_PASS */
+}
+
+LLVMErrorRef module_transform_wrapper(
+      void *Ctx, LLVMOrcThreadSafeModuleRef *ModInOut,
+      LLVMOrcMaterializationResponsibilityRef MR) {
+   return LLVMOrcThreadSafeModuleWithModuleDo(*ModInOut, *module_transform, Ctx);
+}
+
+LPJit::LPJit() :jit_dylib_count(0) {
+   using namespace llvm::orc;
+
+   ::gallivm_debug = debug_get_option_gallivm_debug();
+
+
+   gallivm_perf = debug_get_flags_option("GALLIVM_PERF", lp_bld_perf_flags, 0 );
+
+   init_native_targets();
+   JITTargetMachineBuilder JTMB = create_jtdb();
+   tm = wrap(ExitOnErr(JTMB.createTargetMachine()).release());
+
+   /* Create an LLJIT instance with an ObjectLinkingLayer (JITLINK)
+    * or RuntimeDyld as the base layer.
+    * intel & perf listeners are not supported by ObjectLinkingLayer yet
+    */
+   lljit = ExitOnErr(
+      LLJITBuilder()
+         .setJITTargetMachineBuilder(std::move(JTMB))
+#ifdef USE_JITLINK
+         .setObjectLinkingLayerCreator(
+            [&](ExecutionSession &ES, const llvm::Triple &TT) {
+               return std::make_unique<ObjectLinkingLayer>(
+                  ES, ExitOnErr(llvm::jitlink::InProcessMemoryManager::Create()));
+            })
+#else
+#if LLVM_USE_INTEL_JITEVENTS
+         .RegisterJITEventListener(
+               llvm::JITEventListener::createIntelJITEventListener())
+#endif
+#endif
+         .create());
+
+   LLVMOrcIRTransformLayerRef TL = wrap(&lljit->getIRTransformLayer());
+   LLVMOrcIRTransformLayerSetTransform(TL, *module_transform_wrapper, NULL);
+}
+
+void LPJit::init_native_targets() {
+   // If we have a native target, initialize it to ensure it is linked in and
+   // usable by the JIT.
+   llvm::InitializeNativeTarget();
+
+   llvm::InitializeNativeTargetAsmPrinter();
+
+   llvm::InitializeNativeTargetDisassembler();
+#if DEBUG
+   {
+      char *env_llc_options = getenv("GALLIVM_LLC_OPTIONS");
+      if (env_llc_options) {
+         char *option;
+         char *options[64] = {(char *) "llc"};      // Warning without cast
+         int   n;
+         for (n = 0, option = strtok(env_llc_options, " "); option; n++, option = strtok(NULL, " ")) {
+            options[n + 1] = option;
+         }
+         if (::gallivm_debug & (GALLIVM_DEBUG_IR | GALLIVM_DEBUG_ASM | GALLIVM_DEBUG_DUMP_BC)) {
+            debug_printf("llc additional options (%d):\n", n);
+            for (int i = 1; i <= n; i++)
+               debug_printf("\t%s\n", options[i]);
+            debug_printf("\n");
+         }
+         LLVMParseCommandLineOptions(n + 1, options, NULL);
+      }
+   }
+#endif
+
+   if (util_get_cpu_caps()->has_avx2 || util_get_cpu_caps()->has_avx) {
+      ::lp_native_vector_width = 256;
+   } else {
+      /* Leave it at 128, even when no SIMD extensions are available.
+       * Really needs to be a multiple of 128 so can fit 4 floats.
+       */
+      ::lp_native_vector_width = 128;
+   }
+
+   ::lp_native_vector_width = debug_get_num_option("LP_NATIVE_VECTOR_WIDTH",
+                                                 lp_native_vector_width);
+
+#if DETECT_ARCH_PPC_64 == 1
+   /* Set the NJ bit in VSCR to 0 so denormalized values are handled as
+    * specified by IEEE standard (PowerISA 2.06 - Section 6.3). This guarantees
+    * that some rounding and half-float to float handling does not round
+    * incorrectly to 0.
+    * XXX: should eventually follow same logic on all platforms.
+    * Right now denorms get explicitly disabled (but elsewhere) for x86,
+    * whereas ppc64 explicitly enables them...
+    */
+   if (util_get_cpu_caps()->has_altivec) {
+      unsigned short mask[] = { 0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
+                                0xFFFF, 0xFFFF, 0xFFFE, 0xFFFF };
+      __asm (
+        "mfvscr %%v1\n"
+        "vand   %0,%%v1,%0\n"
+        "mtvscr %0"
+        :
+        : "r" (*mask)
+      );
+   }
+#endif
+}
+
+llvm::orc::JITTargetMachineBuilder LPJit::create_jtdb() {
+   using namespace llvm;
+   using orc::JITTargetMachineBuilder;
+
+#if defined(_WIN32) && LLVM_VERSION_MAJOR < 15
+   /*
+    * JITLink works on Windows, but only through ELF object format.
+    *
+    * XXX: We could use `LLVM_HOST_TRIPLE "-elf"` but LLVM_HOST_TRIPLE has
+    * different strings for MinGW/MSVC, so better play it safe and be
+    * explicit.
+    */
+#  ifdef _WIN64
+   JITTargetMachineBuilder JTMB((Triple("x86_64-pc-win32-elf")));
+#  else
+   JITTargetMachineBuilder JTMB((Triple("i686-pc-win32-elf")));
+#  endif
+#else
+   /*
+    * llvm::sys::getProcessTriple() is bogus. It returns the host LLVM was 
+    * compiled on. Be careful when doing cross compilation
+    */
+   JITTargetMachineBuilder JTMB((Triple(sys::getProcessTriple())));
+#endif
+
+   TargetOptions options;
+   /**
+    * LLVM 3.1+ haven't more "extern unsigned llvm::StackAlignmentOverride" and
+    * friends for configuring code generation options, like stack alignment.
+    */
+#if DETECT_ARCH_X86 == 1 && LLVM_VERSION_MAJOR < 13
+   options.StackAlignmentOverride = 4;
+#endif
+
+#if DETECT_ARCH_RISCV64 == 1
+#if defined(__riscv_float_abi_soft)
+   options.MCOptions.ABIName = "lp64";
+#elif defined(__riscv_float_abi_single)
+   options.MCOptions.ABIName = "lp64f";
+#elif defined(__riscv_float_abi_double)
+   options.MCOptions.ABIName = "lp64d";
+#else
+#error "GALLIVM: unknown target riscv float abi"
+#endif
+#endif
+
+#if DETECT_ARCH_RISCV32 == 1
+#if defined(__riscv_float_abi_soft)
+   options.MCOptions.ABIName = "ilp32";
+#elif defined(__riscv_float_abi_single)
+   options.MCOptions.ABIName = "ilp32f";
+#elif defined(__riscv_float_abi_double)
+   options.MCOptions.ABIName = "ilp32d";
+#else
+#error "GALLIVM: unknown target riscv float abi"
+#endif
+#endif
+
+   JTMB.setOptions(options);
+
+   std::vector<std::string> MAttrs;
+
+#if LLVM_VERSION_MAJOR >= 4 && (DETECT_ARCH_X86 == 1 || DETECT_ARCH_X86_64 == 1 || DETECT_ARCH_ARM == 1)
+   /* llvm-3.3+ implements sys::getHostCPUFeatures for Arm
+    * and llvm-3.7+ for x86, which allows us to enable/disable
+    * code generation based on the results of cpuid on these
+    * architectures.
+    */
+   StringMap<bool> features;
+   sys::getHostCPUFeatures(features);
+
+   for (StringMapIterator<bool> f = features.begin();
+        f != features.end();
+        ++f) {
+      MAttrs.push_back(((*f).second ? "+" : "-") + (*f).first().str());
+   }
+#elif DETECT_ARCH_X86 == 1 || DETECT_ARCH_X86_64 == 1
+   /*
+    * We need to unset attributes because sometimes LLVM mistakenly assumes
+    * certain features are present given the processor name.
+    *
+    * https://bugs.freedesktop.org/show_bug.cgi?id=92214
+    * http://llvm.org/PR25021
+    * http://llvm.org/PR19429
+    * http://llvm.org/PR16721
+    */
+   MAttrs.push_back(util_get_cpu_caps()->has_sse    ? "+sse"    : "-sse"   );
+   MAttrs.push_back(util_get_cpu_caps()->has_sse2   ? "+sse2"   : "-sse2"  );
+   MAttrs.push_back(util_get_cpu_caps()->has_sse3   ? "+sse3"   : "-sse3"  );
+   MAttrs.push_back(util_get_cpu_caps()->has_ssse3  ? "+ssse3"  : "-ssse3" );
+   MAttrs.push_back(util_get_cpu_caps()->has_sse4_1 ? "+sse4.1" : "-sse4.1");
+   MAttrs.push_back(util_get_cpu_caps()->has_sse4_2 ? "+sse4.2" : "-sse4.2");
+   /*
+    * AVX feature is not automatically detected from CPUID by the X86 target
+    * yet, because the old (yet default) JIT engine is not capable of
+    * emitting the opcodes. On newer llvm versions it is and at least some
+    * versions (tested with 3.3) will emit avx opcodes without this anyway.
+    */
+   MAttrs.push_back(util_get_cpu_caps()->has_avx  ? "+avx"  : "-avx");
+   MAttrs.push_back(util_get_cpu_caps()->has_f16c ? "+f16c" : "-f16c");
+   MAttrs.push_back(util_get_cpu_caps()->has_fma  ? "+fma"  : "-fma");
+   MAttrs.push_back(util_get_cpu_caps()->has_avx2 ? "+avx2" : "-avx2");
+   /* disable avx512 and all subvariants */
+   MAttrs.push_back("-avx512cd");
+   MAttrs.push_back("-avx512er");
+   MAttrs.push_back("-avx512f");
+   MAttrs.push_back("-avx512pf");
+   MAttrs.push_back("-avx512bw");
+   MAttrs.push_back("-avx512dq");
+   MAttrs.push_back("-avx512vl");
+#endif
+#if DETECT_ARCH_ARM == 1
+   if (!util_get_cpu_caps()->has_neon) {
+      MAttrs.push_back("-neon");
+      MAttrs.push_back("-crypto");
+      MAttrs.push_back("-vfp2");
+   }
+#endif
+
+#if DETECT_ARCH_PPC == 1
+   MAttrs.push_back(util_get_cpu_caps()->has_altivec ? "+altivec" : "-altivec");
+#if (LLVM_VERSION_MAJOR < 4)
+   /*
+    * Make sure VSX instructions are disabled
+    * See LLVM bugs:
+    * https://llvm.org/bugs/show_bug.cgi?id=25503#c7 (fixed in 3.8.1)
+    * https://llvm.org/bugs/show_bug.cgi?id=26775 (fixed in 3.8.1)
+    * https://llvm.org/bugs/show_bug.cgi?id=33531 (fixed in 4.0)
+    * https://llvm.org/bugs/show_bug.cgi?id=34647 (llc performance on certain unusual shader IR; intro'd in 4.0, pending as of 5.0)
+    */
+   if (util_get_cpu_caps()->has_altivec) {
+      MAttrs.push_back("-vsx");
+   }
+#else
+   /*
+    * Bug 25503 is fixed, by the same fix that fixed
+    * bug 26775, in versions of LLVM later than 3.8 (starting with 3.8.1).
+    * BZ 33531 actually comprises more than one bug, all of
+    * which are fixed in LLVM 4.0.
+    *
+    * With LLVM 4.0 or higher:
+    * Make sure VSX instructions are ENABLED (if supported), unless
+    * VSX instructions are explicitly enabled/disabled via GALLIVM_VSX=1 or 0.
+    */
+   if (util_get_cpu_caps()->has_altivec) {
+      MAttrs.push_back(util_get_cpu_caps()->has_vsx ? "+vsx" : "-vsx");
+   }
+#endif
+#endif
+
+#if DETECT_ARCH_MIPS64 == 1
+   MAttrs.push_back(util_get_cpu_caps()->has_msa ? "+msa" : "-msa");
+   /* MSA requires a 64-bit FPU register file */
+   MAttrs.push_back("+fp64");
+#endif
+
+#if DETECT_ARCH_RISCV64 == 1
+   /* Before riscv is more matured and util_get_cpu_caps() is implemented,
+    * assume this for now since most of linux capable riscv machine are
+    * riscv64gc
+    */
+   MAttrs = {"+m","+c","+a","+d","+f"};
+#endif
+
+   JTMB.addFeatures(MAttrs);
+
+   if (::gallivm_debug & (GALLIVM_DEBUG_IR | GALLIVM_DEBUG_ASM | GALLIVM_DEBUG_DUMP_BC)) {
+      int n = MAttrs.size();
+      if (n > 0) {
+         debug_printf("llc -mattr option(s): ");
+         for (int i = 0; i < n; i++)
+            debug_printf("%s%s", MAttrs[i].c_str(), (i < n - 1) ? "," : "");
+         debug_printf("\n");
+      }
+   }
+
+   std::string MCPU = llvm::sys::getHostCPUName().str();
+   /*
+    * Note that the MAttrs set above will be sort of ignored (since we should
+    * not set any which would not be set by specifying the cpu anyway).
+    * It ought to be safe though since getHostCPUName() should include bits
+    * not only from the cpu but environment as well (for instance if it's safe
+    * to use avx instructions which need OS support). According to
+    * http://llvm.org/bugs/show_bug.cgi?id=19429 however if I understand this
+    * right it may be necessary to specify older cpu (or disable mattrs) though
+    * when not using MCJIT so no instructions are generated which the old JIT
+    * can't handle. Not entirely sure if we really need to do anything yet.
+    * 
+    * Not sure if the above is also the case for ORCJIT, but we need set CPU
+    * manually since we don't use JITTargetMachineBuilder::detectHost()
+    */
+
+#if DETECT_ARCH_PPC_64 == 1
+   /*
+    * Large programs, e.g. gnome-shell and firefox, may tax the addressability
+    * of the Medium code model once dynamically generated JIT-compiled shader
+    * programs are linked in and relocated.  Yet the default code model as of
+    * LLVM 8 is Medium or even Small.
+    * The cost of changing from Medium to Large is negligible:
+    * - an additional 8-byte pointer stored immediately before the shader entrypoint;
+    * - change an add-immediate (addis) instruction to a load (ld).
+    */
+   JTMB.setCodeModel(CodeModel::Large);
+
+#if UTIL_ARCH_LITTLE_ENDIAN
+   /*
+    * Versions of LLVM prior to 4.0 lacked a table entry for "POWER8NVL",
+    * resulting in (big-endian) "generic" being returned on
+    * little-endian Power8NVL systems.  The result was that code that
+    * attempted to load the least significant 32 bits of a 64-bit quantity
+    * from memory loaded the wrong half.  This resulted in failures in some
+    * Piglit tests, e.g.
+    * .../arb_gpu_shader_fp64/execution/conversion/frag-conversion-explicit-double-uint
+    */
+   if (MCPU == "generic")
+      MCPU = "pwr8";
+#endif
+#endif
+
+#if DETECT_ARCH_MIPS64 == 1
+   /*
+    * ls3a4000 CPU and ls2k1000 SoC is a mips64r5 compatible with MSA SIMD
+    * instruction set implemented, while ls3a3000 is mips64r2 compatible
+    * only. getHostCPUName() return "generic" on all loongson
+    * mips CPU currently. So we override the MCPU to mips64r5 if MSA is
+    * implemented, feedback to mips64r2 for all other ordinary mips64 cpu.
+    */
+   if (MCPU == "generic")
+      MCPU = util_get_cpu_caps()->has_msa ? "mips64r5" : "mips64r2";
+#endif
+
+#if DETECT_ARCH_RISCV64 == 1
+   /**
+    * should be fixed with https://reviews.llvm.org/D121149 in llvm 15,
+    * set it anyway for llvm 14
+    */
+   if (MCPU == "generic")
+      MCPU = "generic-rv64";
+
+   JTMB.setCodeModel(CodeModel::Medium);
+   JTMB.setRelocationModel(Reloc::PIC_);
+#endif
+
+#if DETECT_ARCH_RISCV32 == 1
+   /**
+    * should be fixed with https://reviews.llvm.org/D121149 in llvm 15,
+    * set it anyway for llvm 14
+    */
+   if (MCPU == "generic")
+      MCPU = "generic-rv32";
+
+   JTMB.setCodeModel(CodeModel::Medium);
+   JTMB.setRelocationModel(Reloc::PIC_);
+#endif
+
+   JTMB.setCPU(MCPU);
+   if (gallivm_debug & (GALLIVM_DEBUG_IR | GALLIVM_DEBUG_ASM | GALLIVM_DEBUG_DUMP_BC)) {
+      debug_printf("llc -mcpu option: %s\n", MCPU.c_str());
+   }
+
+   return JTMB;
+}
+
+
+} /* Anonymous namespace */
+
+unsigned
+lp_build_init_native_width(void)
+{
+   // Default to 256 until we're confident llvmpipe with 512 is as correct and not slower than 256
+   lp_native_vector_width = MIN2(util_get_cpu_caps()->max_vector_bits, 256);
+   assert(lp_native_vector_width);
+
+   lp_native_vector_width = debug_get_num_option("LP_NATIVE_VECTOR_WIDTH", lp_native_vector_width);
+   assert(lp_native_vector_width);
+
+   return lp_native_vector_width;
+}
+
+bool
+lp_build_init(void)
+{
+   (void)LPJit::get_instance();
+   return true;
+}
+
+bool
+init_gallivm_state(struct gallivm_state *gallivm, const char *name,
+                   LLVMOrcThreadSafeContextRef context, struct lp_cached_code *cache)
+{
+   assert(!gallivm->context);
+   assert(!gallivm->_ts_context);
+   assert(!gallivm->module);
+
+   if (!lp_build_init())
+      return false;
+
+   // cache is not implemented
+   gallivm->cache = cache;
+   if (gallivm->cache)
+      gallivm->cache->data_size = 0;
+
+   gallivm->_ts_context = context;
+   gallivm->context = LLVMOrcThreadSafeContextGetContext(context);
+
+   gallivm->module_name = LPJit::get_unique_name(name);
+   gallivm->module = LLVMModuleCreateWithNameInContext(gallivm->module_name,
+                                                       gallivm->context);
+#if DETECT_ARCH_X86 == 1
+   lp_set_module_stack_alignment_override(gallivm->module, 4);
+#endif
+   gallivm->builder = LLVMCreateBuilderInContext(gallivm->context);
+   gallivm->_per_module_jd = LPJit::create_jit_dylib(gallivm->module_name);
+
+   gallivm->target = LLVMCreateTargetDataLayout(LPJit::get_instance()->tm);
+   
+   lp_build_coro_declare_malloc_hooks(gallivm);
+   return true;
+}
+
+struct gallivm_state *
+gallivm_create(const char *name, LLVMOrcThreadSafeContextRef context,
+               struct lp_cached_code *cache){
+   struct gallivm_state *gallivm;
+
+   gallivm = CALLOC_STRUCT(gallivm_state);
+   if (gallivm) {
+      if (!init_gallivm_state(gallivm, name, context, cache)) {
+         FREE(gallivm);
+         gallivm = NULL;
+      }
+   }
+
+   assert(gallivm != NULL);
+   return gallivm;
+}
+
+void
+gallivm_destroy(struct gallivm_state *gallivm)
+{
+   LPJit::remove_jd(gallivm->_per_module_jd);
+   gallivm->_per_module_jd = nullptr;
+}
+
+void
+gallivm_free_ir(struct gallivm_state *gallivm)
+{
+   if (gallivm->module)
+      LLVMDisposeModule(gallivm->module);
+   FREE(gallivm->module_name);
+
+   if (gallivm->target) {
+      LLVMDisposeTargetData(gallivm->target);
+   }
+
+   if (gallivm->builder)
+      LLVMDisposeBuilder(gallivm->builder);
+
+   gallivm->target = NULL;
+   gallivm->module=NULL;
+   gallivm->module_name=NULL;
+   gallivm->builder=NULL;
+   gallivm->context=NULL;
+   gallivm->_ts_context=NULL;
+   gallivm->cache=NULL;
+   LPJit::deregister_gallivm_state(gallivm);
+}
+
+void
+gallivm_verify_function(struct gallivm_state *gallivm,
+                        LLVMValueRef func)
+{
+   /* Verify the LLVM IR.  If invalid, dump and abort */
+#ifdef DEBUG
+   if (LLVMVerifyFunction(func, LLVMPrintMessageAction)) {
+      lp_debug_dump_value(func);
+      assert(0);
+      return;
+   }
+#endif
+
+   if (gallivm_debug & GALLIVM_DEBUG_IR) {
+      /* Print the LLVM IR to stderr */
+      lp_debug_dump_value(func);
+      debug_printf("\n");
+   }
+return;
+}
+
+void
+gallivm_add_global_mapping(struct gallivm_state *gallivm, LLVMValueRef sym, void* addr)
+{
+   LPJit::add_mapping_to_jd(sym, addr, gallivm->_per_module_jd);
+}
+
+void lp_init_clock_hook(struct gallivm_state *gallivm)
+{
+   if (gallivm->get_time_hook)
+      return;
+
+   LLVMTypeRef get_time_type = LLVMFunctionType(LLVMInt64TypeInContext(gallivm->context), NULL, 0, 1);
+   gallivm->get_time_hook = LLVMAddFunction(gallivm->module, "get_time_hook", get_time_type);
+}
+
+void
+gallivm_compile_module(struct gallivm_state *gallivm)
+{
+   lp_init_printf_hook(gallivm);
+   gallivm_add_global_mapping(gallivm, gallivm->debug_printf_hook, 
+         (void *)debug_printf);
+
+   lp_init_clock_hook(gallivm);
+   gallivm_add_global_mapping(gallivm, gallivm->get_time_hook,
+         (void *)os_time_get_nano);
+
+   lp_build_coro_add_malloc_hooks(gallivm);
+
+   LPJit::add_ir_module_to_jd(gallivm->_ts_context, gallivm->module,
+      gallivm->_per_module_jd);
+   /* ownership of module is now transferred into orc jit,
+    * disallow modifying it
+    */
+   LPJit::register_gallivm_state(gallivm);
+   gallivm->module=nullptr;
+
+   /* defer compilation till first lookup by gallivm_jit_function */
+}
+
+func_pointer
+gallivm_jit_function(struct gallivm_state *gallivm,
+                     LLVMValueRef func, const char *func_name)
+{
+   return pointer_to_func(
+      LPJit::lookup_in_jd(func_name, gallivm->_per_module_jd));
+}
+
+unsigned
+gallivm_get_perf_flags(void){
+   return gallivm_perf;
+}
+
+void
+lp_set_module_stack_alignment_override(LLVMModuleRef MRef, unsigned align)
+{
+#if LLVM_VERSION_MAJOR >= 13
+   llvm::Module *M = llvm::unwrap(MRef);
+   M->setOverrideStackAlignment(align);
+#endif
+}
diff --git a/src/gallium/auxiliary/meson.build b/src/gallium/auxiliary/meson.build
index 96b0272..f77ee22 100644
--- a/src/gallium/auxiliary/meson.build
+++ b/src/gallium/auxiliary/meson.build
@@ -347,7 +347,6 @@ if draw_with_llvm
     'gallivm/lp_bld_gather.c',
     'gallivm/lp_bld_gather.h',
     'gallivm/lp_bld.h',
-    'gallivm/lp_bld_init.c',
     'gallivm/lp_bld_init.h',
     'gallivm/lp_bld_intr.c',
     'gallivm/lp_bld_intr.h',
@@ -399,6 +398,11 @@ if draw_with_llvm
     'nir/nir_to_tgsi_info.c',
     'nir/nir_to_tgsi_info.h',
   )
+  if llvm_with_orcjit
+    files_libgallium += files('gallivm/lp_bld_init_orc.cpp',)
+  else
+    files_libgallium += files('gallivm/lp_bld_init.c',)
+  endif
 endif
 
 files_libgalliumvl = files(
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index 8e5e8ce..1032468 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -49,6 +49,10 @@
 #include "lp_screen.h"
 #include "lp_fence.h"
 
+#if ! (GALLIVM_USE_ORCJIT == 1)
+#define USE_GLOBAL_LLVM_CONTEXT
+#endif
+
 static void
 llvmpipe_destroy(struct pipe_context *pipe)
 {
@@ -108,7 +112,11 @@ llvmpipe_destroy(struct pipe_context *pipe)
    llvmpipe_sampler_matrix_destroy(llvmpipe);
 
 #ifndef USE_GLOBAL_LLVM_CONTEXT
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(llvmpipe->context);
+#else
    LLVMContextDispose(llvmpipe->context);
+#endif
 #endif
    llvmpipe->context = NULL;
 
@@ -258,15 +266,25 @@ llvmpipe_create_context(struct pipe_screen *screen, void *priv,
 
 #ifdef USE_GLOBAL_LLVM_CONTEXT
    llvmpipe->context = LLVMGetGlobalContext();
+#else
+#if GALLIVM_USE_ORCJIT == 1
+   llvmpipe->context = LLVMOrcCreateNewThreadSafeContext();
 #else
    llvmpipe->context = LLVMContextCreate();
+#endif
 #endif
 
    if (!llvmpipe->context)
       goto fail;
 
+#if GALLIVM_USE_ORCJIT == 1
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(llvmpipe->context), false);
+#endif
+#else
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(llvmpipe->context, false);
+#endif
 #endif
 
    /*
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 7dd6a3f..d30aa89 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -190,7 +190,11 @@ struct llvmpipe_context {
    unsigned render_cond_offset;
 
    /** The LLVMContext to use for LLVM related work */
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
 
    int max_global_buffers;
    struct pipe_resource **global_buffers;
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index ad3d664..514217d 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -950,7 +950,11 @@ static void
 lp_disk_cache_create(struct llvmpipe_screen *screen)
 {
    struct mesa_sha1 ctx;
+#if GALLIVM_USE_ORCJIT == 1
+   unsigned gallivm_perf = 0;
+#else
    unsigned gallivm_perf = gallivm_get_perf_flags();
+#endif
    unsigned char sha1[20];
    char cache_id[20 * 2 + 1];
    _mesa_sha1_init(&ctx);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_cs.c b/src/gallium/drivers/llvmpipe/lp_state_cs.c
index 46cc5ff..cb3845e 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_cs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_cs.c
@@ -397,6 +397,8 @@ generate_compute(struct llvmpipe_context *lp,
    lp_build_coro_add_presplit(coro);
 
    variant->function = function;
+   variant->function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name, func_name);
 
    for (i = 0; i < CS_ARG_MAX - !is_mesh; ++i) {
       if (LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind) {
@@ -1006,6 +1008,8 @@ llvmpipe_remove_cs_shader_variant(struct llvmpipe_context *lp,
    lp->nr_cs_variants--;
    lp->nr_cs_instrs -= variant->nr_instrs;
 
+   if(variant->function_name)
+      FREE(variant->function_name);
    FREE(variant);
 }
 
@@ -1264,12 +1268,20 @@ generate_variant(struct llvmpipe_context *lp,
 
    generate_compute(lp, shader, variant);
 
+#if GALLIVM_USE_ORCJIT == 1
+/* module has been moved into ORCJIT after gallivm_compile_module */
+   variant->nr_instrs += lp_build_count_ir_module(variant->gallivm->module);
+
+   gallivm_compile_module(variant->gallivm);
+#else
    gallivm_compile_module(variant->gallivm);
 
    variant->nr_instrs += lp_build_count_ir_module(variant->gallivm->module);
+#endif
 
    variant->jit_function = (lp_jit_cs_func)
-      gallivm_jit_function(variant->gallivm, variant->function);
+      gallivm_jit_function(variant->gallivm, variant->function,
+                           variant->function_name);
 
    if (needs_caching) {
       lp_disk_cache_insert_shader(screen, &cached, ir_sha1_cache_key);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_cs.h b/src/gallium/drivers/llvmpipe/lp_state_cs.h
index 11d2e2c..0f69d05 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_cs.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_cs.h
@@ -91,6 +91,7 @@ struct lp_compute_shader_variant
    LLVMTypeRef jit_vertex_header_ptr_type;
    LLVMTypeRef jit_prim_type;
    LLVMValueRef function;
+   char* function_name;
    lp_jit_cs_func jit_function;
 
    /* Total number of LLVM instructions generated */
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 21312aa..b723392 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -3207,6 +3207,8 @@ generate_fragment(struct llvmpipe_context *lp,
    LLVMSetFunctionCallConv(function, LLVMCCallConv);
 
    variant->function[partial_mask] = function;
+   variant->function_name[partial_mask] = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name[partial_mask], func_name);
 
    /* XXX: need to propagate noalias down into color param now we are
     * passing a pointer-to-pointer?
@@ -3306,6 +3308,7 @@ generate_fragment(struct llvmpipe_context *lp,
       LLVMValueRef glob_sample_pos =
          LLVMAddGlobal(gallivm->module,
                        LLVMArrayType(flt_type, key->coverage_samples * 2), "");
+      LLVMSetLinkage(glob_sample_pos, LLVMInternalLinkage);
       LLVMValueRef sample_pos_array;
 
       if (key->multisample && key->coverage_samples == 4) {
@@ -3913,20 +3916,29 @@ generate_variant(struct llvmpipe_context *lp,
     * Compile everything
     */
 
+#if GALLIVM_USE_ORCJIT == 1
+/* module has been moved into ORCJIT after gallivm_compile_module */
+   variant->nr_instrs += lp_build_count_ir_module(variant->gallivm->module);
+
+   gallivm_compile_module(variant->gallivm);
+#else
    gallivm_compile_module(variant->gallivm);
 
    variant->nr_instrs += lp_build_count_ir_module(variant->gallivm->module);
+#endif
 
    if (variant->function[RAST_EDGE_TEST]) {
       variant->jit_function[RAST_EDGE_TEST] = (lp_jit_frag_func)
             gallivm_jit_function(variant->gallivm,
-                                 variant->function[RAST_EDGE_TEST]);
+                                 variant->function[RAST_EDGE_TEST],
+                                 variant->function_name[RAST_EDGE_TEST]);
    }
 
    if (variant->function[RAST_WHOLE]) {
       variant->jit_function[RAST_WHOLE] = (lp_jit_frag_func)
          gallivm_jit_function(variant->gallivm,
-                              variant->function[RAST_WHOLE]);
+                              variant->function[RAST_WHOLE],
+                              variant->function_name[RAST_WHOLE]);
    } else if (!variant->jit_function[RAST_WHOLE]) {
       variant->jit_function[RAST_WHOLE] = (lp_jit_frag_func)
          variant->jit_function[RAST_EDGE_TEST];
@@ -3935,7 +3947,8 @@ generate_variant(struct llvmpipe_context *lp,
    if (linear_pipeline) {
       if (variant->linear_function) {
          variant->jit_linear_llvm = (lp_jit_linear_llvm_func)
-            gallivm_jit_function(variant->gallivm, variant->linear_function);
+            gallivm_jit_function(variant->gallivm, variant->linear_function,
+                                 variant->linear_function_name);
       }
 
       /*
@@ -4111,6 +4124,12 @@ llvmpipe_destroy_shader_variant(struct llvmpipe_context *lp,
 {
    gallivm_destroy(variant->gallivm);
    lp_fs_reference(lp, &variant->shader, NULL);
+   if (variant->function_name[RAST_EDGE_TEST])
+      FREE(variant->function_name[RAST_EDGE_TEST]);
+   if (variant->function_name[RAST_WHOLE])
+      FREE(variant->function_name[RAST_WHOLE]);
+   if (variant->linear_function_name)
+      FREE(variant->linear_function_name);
    FREE(variant);
 }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.h b/src/gallium/drivers/llvmpipe/lp_state_fs.h
index 195cdd5..c2fa210 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.h
@@ -168,6 +168,7 @@ struct lp_fragment_shader_variant
    LLVMTypeRef jit_linear_textures_type;
 
    LLVMValueRef function[2]; // [RAST_WHOLE], [RAST_EDGE_TEST]
+   char* function_name[2];
 
    lp_jit_frag_func jit_function[2]; // [RAST_WHOLE], [RAST_EDGE_TEST]
 
@@ -177,6 +178,7 @@ struct lp_fragment_shader_variant
    /* Functions within the linear path:
     */
    LLVMValueRef linear_function;
+   char* linear_function_name;
    lp_jit_linear_llvm_func jit_linear_llvm;
 
    /* Bitmask to say what cbufs are unswizzled */
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs_linear_llvm.c b/src/gallium/drivers/llvmpipe/lp_state_fs_linear_llvm.c
index c64f5cf..70db126 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs_linear_llvm.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs_linear_llvm.c
@@ -289,6 +289,8 @@ llvmpipe_fs_variant_linear_llvm(struct llvmpipe_context *lp,
    LLVMSetFunctionCallConv(function, LLVMCCallConv);
 
    variant->linear_function = function;
+   variant->linear_function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->linear_function_name, func_name);
 
    /* XXX: need to propagate noalias down into color param now we are
     * passing a pointer-to-pointer?
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.c b/src/gallium/drivers/llvmpipe/lp_state_setup.c
index 2f4bbdf..a68e960 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.c
@@ -688,6 +688,8 @@ generate_setup_variant(struct lp_setup_variant_key *key,
                        arg_types, ARRAY_SIZE(arg_types), 0);
 
    variant->function = LLVMAddFunction(gallivm->module, func_name, func_type);
+   variant->function_name = MALLOC(strlen(func_name)+1);
+   strcpy(variant->function_name, func_name);
    if (!variant->function)
       goto fail;
 
@@ -732,7 +734,7 @@ generate_setup_variant(struct lp_setup_variant_key *key,
    gallivm_compile_module(gallivm);
 
    variant->jit_function = (lp_jit_setup_triangle)
-      gallivm_jit_function(gallivm, variant->function);
+      gallivm_jit_function(gallivm, variant->function, variant->function_name);
    if (!variant->jit_function)
       goto fail;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.h b/src/gallium/drivers/llvmpipe/lp_state_setup.h
index ef20893..5882ce1 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.h
@@ -66,6 +66,7 @@ struct lp_setup_variant {
     * assembly.
     */
    LLVMValueRef function;
+   char *function_name;
 
    /* The actual generated setup function:
     */
diff --git a/src/gallium/drivers/llvmpipe/lp_test_arit.c b/src/gallium/drivers/llvmpipe/lp_test_arit.c
index fd8489f..aeef6c5 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_arit.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_arit.c
@@ -417,7 +417,11 @@ test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test, unsigned
 {
    char test_name[128];
    snprintf(test_name, sizeof test_name, "%s.v%u", test->name, length);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
    struct gallivm_state *gallivm;
    LLVMValueRef test_func;
    unary_func_t test_func_jit;
@@ -433,9 +437,16 @@ test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test, unsigned
       in[i] = 1.0;
    }
 
+#if GALLIVM_USE_ORCJIT == 1
+   context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
    context = LLVMContextCreate();
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(context, false);
+#endif
 #endif
    gallivm = gallivm_create("test_module", context, NULL);
 
@@ -443,7 +454,8 @@ test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test, unsigned
 
    gallivm_compile_module(gallivm);
 
-   test_func_jit = (unary_func_t) gallivm_jit_function(gallivm, test_func);
+   test_func_jit = (unary_func_t) gallivm_jit_function(gallivm, test_func, 
+                                                      test_name);
 
    gallivm_free_ir(gallivm);
 
@@ -512,7 +524,11 @@ test_unary(unsigned verbose, FILE *fp, const struct unary_test_t *test, unsigned
    }
 
    gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
    LLVMContextDispose(context);
+#endif
 
    align_free(in);
    align_free(out);
diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c
index 7851bf2..c527cbf 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c
@@ -437,7 +437,11 @@ test_one(unsigned verbose,
          const struct pipe_blend_state *blend,
          struct lp_type type)
 {
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
    struct gallivm_state *gallivm;
    LLVMValueRef func = NULL;
    blend_test_ptr_t blend_test_ptr;
@@ -451,9 +455,16 @@ test_one(unsigned verbose,
    if (verbose >= 1)
       dump_blend_type(stdout, blend, type);
 
+#if GALLIVM_USE_ORCJIT == 1
+   context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
    context = LLVMContextCreate();
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(context, false);
+#endif
 #endif
    gallivm = gallivm_create("test_module", context, NULL);
 
@@ -461,7 +472,8 @@ test_one(unsigned verbose,
 
    gallivm_compile_module(gallivm);
 
-   blend_test_ptr = (blend_test_ptr_t)gallivm_jit_function(gallivm, func);
+   blend_test_ptr = (blend_test_ptr_t)gallivm_jit_function(gallivm, func,
+                                                         "test");
 
    gallivm_free_ir(gallivm);
 
@@ -584,7 +596,11 @@ test_one(unsigned verbose,
       write_tsv_row(fp, blend, type, cycles_avg, success);
 
    gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
    LLVMContextDispose(context);
+#endif
 
    return success;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c
index 9fb6448..4709996 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_conv.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c
@@ -157,7 +157,11 @@ test_one(unsigned verbose,
          struct lp_type src_type,
          struct lp_type dst_type)
 {
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
    struct gallivm_state *gallivm;
    LLVMValueRef func = NULL;
    conv_test_ptr_t conv_test_ptr;
@@ -222,9 +226,16 @@ test_one(unsigned verbose,
       eps *= 2;
    }
 
+#if GALLIVM_USE_ORCJIT == 1
+   context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
    context = LLVMContextCreate();
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(context, false);
+#endif
 #endif
    gallivm = gallivm_create("test_module", context, NULL);
 
@@ -232,7 +243,8 @@ test_one(unsigned verbose,
 
    gallivm_compile_module(gallivm);
 
-   conv_test_ptr = (conv_test_ptr_t)gallivm_jit_function(gallivm, func);
+   conv_test_ptr = (conv_test_ptr_t)gallivm_jit_function(gallivm, func, 
+                                                         "test");
 
    gallivm_free_ir(gallivm);
 
@@ -337,7 +349,11 @@ test_one(unsigned verbose,
       write_tsv_row(fp, src_type, dst_type, cycles_avg, success);
 
    gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
    LLVMContextDispose(context);
+#endif
 
    return success;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 6d3a60b..c2c9289 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -79,9 +79,9 @@ static LLVMValueRef
 add_fetch_rgba_test(struct gallivm_state *gallivm, unsigned verbose,
                     const struct util_format_description *desc,
                     struct lp_type type,
-                    unsigned use_cache)
+                    unsigned use_cache,
+		              char *name)
 {
-   char name[256];
    LLVMContextRef context = gallivm->context;
    LLVMModuleRef module = gallivm->module;
    LLVMBuilderRef builder = gallivm->builder;
@@ -96,7 +96,7 @@ add_fetch_rgba_test(struct gallivm_state *gallivm, unsigned verbose,
    LLVMValueRef rgba;
    LLVMValueRef cache = NULL;
 
-   snprintf(name, sizeof name, "fetch_%s_%s", desc->short_name,
+   snprintf(name, 64 * sizeof(char), "fetch_%s_%s", desc->short_name,
             type.floating ? "float" : "unorm8");
 
    args[0] = LLVMPointerType(lp_build_vec_type(gallivm, type), 0);
@@ -139,9 +139,14 @@ test_format_float(unsigned verbose, FILE *fp,
                   const struct util_format_description *desc,
                   unsigned use_cache)
 {
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
    struct gallivm_state *gallivm;
    LLVMValueRef fetch = NULL;
+   char fetch_name[64];
    fetch_ptr_t fetch_ptr;
    alignas(16) uint8_t packed[UTIL_FORMAT_MAX_PACKED_BYTES];
    alignas(16) float unpacked[4];
@@ -149,18 +154,25 @@ test_format_float(unsigned verbose, FILE *fp,
    bool success = true;
    unsigned i, j, k, l;
 
+#if GALLIVM_USE_ORCJIT == 1
+   context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
    context = LLVMContextCreate();
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(context, false);
+#endif
 #endif
    gallivm = gallivm_create("test_module_float", context, NULL);
 
    fetch = add_fetch_rgba_test(gallivm, verbose, desc,
-                               lp_float32_vec4_type(), use_cache);
+                               lp_float32_vec4_type(), use_cache, fetch_name);
 
    gallivm_compile_module(gallivm);
 
-   fetch_ptr = (fetch_ptr_t) gallivm_jit_function(gallivm, fetch);
+   fetch_ptr = (fetch_ptr_t) gallivm_jit_function(gallivm, fetch, fetch_name);
 
    gallivm_free_ir(gallivm);
 
@@ -228,7 +240,11 @@ test_format_float(unsigned verbose, FILE *fp,
    }
 
    gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
    LLVMContextDispose(context);
+#endif
 
    if (fp)
       write_tsv_row(fp, desc, success);
@@ -243,9 +259,14 @@ test_format_unorm8(unsigned verbose, FILE *fp,
                    const struct util_format_description *desc,
                    unsigned use_cache)
 {
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
    struct gallivm_state *gallivm;
    LLVMValueRef fetch = NULL;
+   char fetch_name[64];
    fetch_ptr_t fetch_ptr;
    alignas(16) uint8_t packed[UTIL_FORMAT_MAX_PACKED_BYTES];
    uint8_t unpacked[4];
@@ -253,18 +274,25 @@ test_format_unorm8(unsigned verbose, FILE *fp,
    bool success = true;
    unsigned i, j, k, l;
 
+#if GALLIVM_USE_ORCJIT == 1
+   context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
    context = LLVMContextCreate();
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(context, false);
+#endif
 #endif
    gallivm = gallivm_create("test_module_unorm8", context, NULL);
 
    fetch = add_fetch_rgba_test(gallivm, verbose, desc,
-                               lp_unorm8_vec4_type(), use_cache);
+                               lp_unorm8_vec4_type(), use_cache, fetch_name);
 
    gallivm_compile_module(gallivm);
 
-   fetch_ptr = (fetch_ptr_t) gallivm_jit_function(gallivm, fetch);
+   fetch_ptr = (fetch_ptr_t) gallivm_jit_function(gallivm, fetch, fetch_name);
 
    gallivm_free_ir(gallivm);
 
@@ -331,7 +359,11 @@ test_format_unorm8(unsigned verbose, FILE *fp,
    }
 
    gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
    LLVMContextDispose(context);
+#endif
 
    if (fp)
       write_tsv_row(fp, desc, success);
diff --git a/src/gallium/drivers/llvmpipe/lp_test_lookup_multiple.c b/src/gallium/drivers/llvmpipe/lp_test_lookup_multiple.c
new file mode 100644
index 0000000..9ee7b96
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_test_lookup_multiple.c
@@ -0,0 +1,167 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include <stdlib.h>
+#include <stdio.h>
+
+#include "util/u_pointer.h"
+#include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_init.h"
+#include "gallivm/lp_bld_assert.h"
+#include "gallivm/lp_bld_printf.h"
+
+#include "lp_test.h"
+
+
+struct printf_test_case {
+   int foo;
+};
+
+void
+write_tsv_header(FILE *fp)
+{
+   fprintf(fp,
+           "result\t"
+           "format\n");
+
+   fflush(fp);
+}
+
+
+
+typedef void (*test_printf_t)(int i);
+
+
+static LLVMValueRef
+add_printf_test(struct gallivm_state *gallivm, int n, char *func_name)
+{
+   LLVMModuleRef module = gallivm->module;
+   LLVMTypeRef args[1] = { LLVMIntTypeInContext(gallivm->context, 32) };
+   snprintf(func_name, 64 * sizeof(char), "test_lookup_multiple_%d", n);
+   LLVMValueRef func = LLVMAddFunction(module, func_name, LLVMFunctionType(LLVMVoidTypeInContext(gallivm->context), args, 1, 0));
+   LLVMBuilderRef builder = gallivm->builder;
+   LLVMBasicBlockRef block = LLVMAppendBasicBlockInContext(gallivm->context, func, "entry");
+
+   LLVMSetFunctionCallConv(func, LLVMCCallConv);
+
+   LLVMPositionBuilderAtEnd(builder, block);
+   lp_build_printf(gallivm, "hello, world from ");
+   lp_build_printf(gallivm, func_name);
+   lp_build_printf(gallivm, "print 5 6: %d %d\n", LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 5, 0),
+				LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 6, 0));
+
+   /* Also test lp_build_assert().  This should not fail. */
+   lp_build_assert(gallivm, LLVMConstInt(LLVMInt32TypeInContext(gallivm->context), 1, 0), "assert(1)");
+
+   LLVMBuildRetVoid(builder);
+
+   gallivm_verify_function(gallivm, func);
+
+   return func;
+}
+
+
+static bool
+test_lookup_multiple(unsigned verbose, FILE *fp,
+            const struct printf_test_case *testcase)
+{
+   struct gallivm_state *gallivm;
+   const int N = 10;
+   LLVMValueRef *func = 
+      (LLVMValueRef *) malloc(N * sizeof(LLVMValueRef));
+   char func_name[N][64];
+   test_printf_t *test_lookup_multiple_func = 
+      (test_printf_t *)malloc(N * sizeof(test_printf_t));
+   bool success = true;
+   int i;
+
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
+   LLVMContextRef context = LLVMContextCreate();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(context, false);
+#endif
+#endif
+   gallivm = gallivm_create("test_module", context, NULL);
+
+   for(i = 0; i < N; i++){
+      func[i] = add_printf_test(gallivm, i, func_name[i]);
+   }
+
+   gallivm_compile_module(gallivm);
+
+   for(i = 0; i < N; i++){
+      test_lookup_multiple_func[i] = (test_printf_t) gallivm_jit_function(gallivm, func[i], func_name[i]);
+   }
+
+   gallivm_free_ir(gallivm);
+
+   for(i = 0; i < N; i++){
+      test_lookup_multiple_func[i](0);
+   }
+   FREE(func);
+   FREE(test_lookup_multiple_func);
+   gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
+   LLVMContextDispose(context);
+#endif
+   return success;
+}
+
+
+bool
+test_all(unsigned verbose, FILE *fp)
+{
+   bool success = true;
+
+   test_lookup_multiple(verbose, fp, NULL);
+
+   return success;
+}
+
+
+bool
+test_some(unsigned verbose, FILE *fp,
+          unsigned long n)
+{
+   return test_all(verbose, fp);
+}
+
+
+bool
+test_single(unsigned verbose, FILE *fp)
+{
+   printf("no test_single()");
+   return true;
+}
diff --git a/src/gallium/drivers/llvmpipe/lp_test_printf.c b/src/gallium/drivers/llvmpipe/lp_test_printf.c
index 7e53aa5..f679228 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_printf.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_printf.c
@@ -89,15 +89,26 @@ static bool
 test_printf(unsigned verbose, FILE *fp,
             const struct printf_test_case *testcase)
 {
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcThreadSafeContextRef context;
+#else
    LLVMContextRef context;
+#endif
    struct gallivm_state *gallivm;
    LLVMValueRef test;
    test_printf_t test_printf_func;
    bool success = true;
 
+#if GALLIVM_USE_ORCJIT == 1
+   context = LLVMOrcCreateNewThreadSafeContext();
+#if LLVM_VERSION_MAJOR == 15
+   LLVMContextSetOpaquePointers(LLVMOrcThreadSafeContextGetContext(context), false);
+#endif
+#else
    context = LLVMContextCreate();
 #if LLVM_VERSION_MAJOR == 15
    LLVMContextSetOpaquePointers(context, false);
+#endif
 #endif
    gallivm = gallivm_create("test_module", context, NULL);
 
@@ -105,14 +116,19 @@ test_printf(unsigned verbose, FILE *fp,
 
    gallivm_compile_module(gallivm);
 
-   test_printf_func = (test_printf_t) gallivm_jit_function(gallivm, test);
+   test_printf_func = (test_printf_t) gallivm_jit_function(gallivm, test, 
+                                                            "test_printf");
 
    gallivm_free_ir(gallivm);
 
    test_printf_func(0);
 
    gallivm_destroy(gallivm);
+#if GALLIVM_USE_ORCJIT == 1
+   LLVMOrcDisposeThreadSafeContext(context);
+#else
    LLVMContextDispose(context);
+#endif
 
    return success;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_texture_handle.c b/src/gallium/drivers/llvmpipe/lp_texture_handle.c
index 72152ba..b805ec6 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture_handle.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture_handle.c
@@ -201,14 +201,15 @@ llvmpipe_sampler_matrix_destroy(struct llvmpipe_context *ctx)
 }
 
 static void *
-compile_function(struct llvmpipe_context *ctx, struct gallivm_state *gallivm, LLVMValueRef function,
+compile_function(struct llvmpipe_context *ctx, struct gallivm_state *gallivm, LLVMValueRef function, const char *func_name,
                  bool needs_caching,
                  uint8_t cache_key[SHA1_DIGEST_LENGTH])
 {
    gallivm_verify_function(gallivm, function);
    gallivm_compile_module(gallivm);
 
-   void *function_ptr = func_to_pointer(gallivm_jit_function(gallivm, function));
+   void *function_ptr = func_to_pointer(gallivm_jit_function(gallivm, function, 
+                                                            func_name));
 
    if (needs_caching)
       lp_disk_cache_insert_shader(llvmpipe_screen(ctx->pipe.screen), gallivm->cache, cache_key);
@@ -336,7 +337,7 @@ compile_image_function(struct llvmpipe_context *ctx, struct lp_static_texture_st
 
    free(image_soa);
 
-   return compile_function(ctx, gallivm, function, needs_caching, cache_key);
+   return compile_function(ctx, gallivm, function, "image", needs_caching, cache_key);
 }
 
 static void *
@@ -484,7 +485,7 @@ compile_sample_function(struct llvmpipe_context *ctx, struct lp_static_texture_s
 
    free(sampler_soa);
 
-   return compile_function(ctx, gallivm, function, needs_caching, cache_key);
+   return compile_function(ctx, gallivm, function, "sample", needs_caching, cache_key);
 }
 
 static void *
@@ -566,7 +567,7 @@ compile_size_function(struct llvmpipe_context *ctx, struct lp_static_texture_sta
 
    free(sampler_soa);
 
-   return compile_function(ctx, gallivm, function, needs_caching, cache_key);
+   return compile_function(ctx, gallivm, function, "size", needs_caching, cache_key);
 }
 
 static void
diff --git a/src/gallium/drivers/llvmpipe/meson.build b/src/gallium/drivers/llvmpipe/meson.build
index 38ff889..d405c67 100644
--- a/src/gallium/drivers/llvmpipe/meson.build
+++ b/src/gallium/drivers/llvmpipe/meson.build
@@ -132,7 +132,7 @@ driver_swrast = declare_dependency(
 
 if with_tests and with_gallium_softpipe and draw_with_llvm
   foreach t : ['lp_test_format', 'lp_test_arit', 'lp_test_blend',
-               'lp_test_conv', 'lp_test_printf']
+               'lp_test_conv', 'lp_test_printf', 'lp_test_lookup_multiple']
     test(
       t,
       executable(
diff --git a/src/util/detect_arch.h b/src/util/detect_arch.h
index b3f7f90..36f638a 100644
--- a/src/util/detect_arch.h
+++ b/src/util/detect_arch.h
@@ -101,6 +101,17 @@
 #define DETECT_ARCH_HPPA 1
 #endif
 
+#if defined(__riscv)
+#define DETECT_ARCH_RISCV 1
+#if __riscv_xlen == 64
+#define DETECT_ARCH_RISCV64 1
+#elif __riscv_xlen == 32
+#define DETECT_ARCH_RISCV32 1
+#else
+#error "detect_arch: unknown target riscv xlen"
+#endif
+#endif
+
 #ifndef DETECT_ARCH_X86
 #define DETECT_ARCH_X86 0
 #endif
@@ -145,4 +156,16 @@
 #define DETECT_ARCH_HPPA 0
 #endif
 
+#ifndef DETECT_ARCH_RISCV
+#define DETECT_ARCH_RISCV 0
+#endif
+
+#ifndef DETECT_ARCH_RISCV32
+#define DETECT_ARCH_RISCV32 0
+#endif
+
+#ifndef DETECT_ARCH_RISCV64
+#define DETECT_ARCH_RISCV64 0
+#endif
+
 #endif /* UTIL_DETECT_ARCH_H_ */
-- 
2.43.0