gallivm: add coro malloc hooks earlier and always.

This fixes GALLIVM_DEBUG=asm for compute shaders, changing
the hooks after dumping causes a segfault because the
memory has already been finalised. Just add the hooks always,
and before dumping anything.

Fixes: f511d2a5 ("gallivm: rework coroutine malloc/free callouts.")
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <mesa/mesa!19714>

src/gallium/auxiliary/draw/draw_llvm.c

@@ -3629,12 +3629,10 @@ draw_tcs_llvm_create_variant(struct draw_llvm *llvm,
       draw_tcs_llvm_dump_variant_key(&variant->key);
    }
 
-   lp_build_coro_declare_malloc_hooks(variant->gallivm);
    draw_tcs_llvm_generate(llvm, variant);
 
    gallivm_compile_module(variant->gallivm);
 
-   lp_build_coro_add_malloc_hooks(variant->gallivm);
    variant->jit_func = (draw_tcs_jit_func)
       gallivm_jit_function(variant->gallivm, variant->function);
 

src/gallium/auxiliary/gallivm/lp_bld_init.c

@@ -37,6 +37,7 @@
 #include "lp_bld_debug.h"
 #include "lp_bld_misc.h"
 #include "lp_bld_init.h"
+#include "lp_bld_coro.h"
 #include "lp_bld_printf.h"
 
 #include <llvm/Config/llvm-config.h>
@@ -409,6 +410,7 @@ init_gallivm_state(struct gallivm_state *gallivm, const char *name,
    if (!create_pass_manager(gallivm))
       goto fail;
 
+   lp_build_coro_declare_malloc_hooks(gallivm);
    return TRUE;
 
 fail:
@@ -667,6 +669,8 @@ gallivm_compile_module(struct gallivm_state *gallivm)
    lp_init_clock_hook(gallivm);
    LLVMAddGlobalMapping(gallivm->engine, gallivm->get_time_hook, os_time_get_nano);
 
+   lp_build_coro_add_malloc_hooks(gallivm);
+
    if (gallivm_debug & GALLIVM_DEBUG_ASM) {
       LLVMValueRef llvm_func = LLVMGetFirstFunction(gallivm->module);
 

src/gallium/drivers/llvmpipe/lp_state_cs.c

@@ -148,8 +148,6 @@ generate_compute(struct llvmpipe_context *lp,
       }
    }
 
-   lp_build_coro_declare_malloc_hooks(gallivm);
-
    if (variant->gallivm->cache->data_size)
       return;
 
@@ -830,7 +828,6 @@ generate_variant(struct llvmpipe_context *lp,
 
    gallivm_compile_module(variant->gallivm);
 
-   lp_build_coro_add_malloc_hooks(variant->gallivm);
    variant->nr_instrs += lp_build_count_ir_module(variant->gallivm->module);
 
    variant->jit_function = (lp_jit_cs_func)