feat: Add rs_allocate_closure free function. (#1944)

Adds a new free function rs_allocate_closure which allows the renderer services to provide memory for closure storage, for both inbuilt (add/mul) and user defined closures.

The existing osl_* closure handling functions now use this new free function.

The CPU side fallback is to call through to the existing closure pool implementation. The GPU side fallback returns null.

testshade/testrender provide an example implementation showing how to implement a stack based closure pool for the GPU.

---------

Signed-off-by: Curtis Black <[email protected]>

Author: curtisblack

src/include/OSL/rs_free_function.h

@@ -316,6 +316,13 @@ OSL_RSOP OSL_HOSTDEVICE bool
 rs_trace_get(OSL::OpaqueExecContextPtr oec, OSL::ustringhash name,
              OSL::TypeDesc type, void* val, bool derivatives);
 
+/// Allocates memory for a closure color. May return null if no memory could
+/// be allocated. It is the renderers responsibility to clean up these
+/// allocations after a shader is run and the closures have been processed.
+OSL_RSOP OSL_HOSTDEVICE void*
+rs_allocate_closure(OSL::OpaqueExecContextPtr oec, size_t size,
+                    size_t alignment);
+
 /// Report errors, warnings, printf, and fprintf.
 /// Fmtlib style format specifier is used (vs. printf style)
 /// Arguments are represented as EncodedTypes (encodedtypes.h) and

src/include/OSL/strdecls.h

@@ -14,6 +14,10 @@
 #    error Do not include this file unless STRDECL is defined
 #endif
 
+// avoid naming conflict with MSVC macro
+#ifdef RGB
+#    undef RGB
+#endif
 
 STRDECL("", _emptystring_)
 STRDECL("camera", camera)

src/liboslexec/CMakeLists.txt

@@ -194,7 +194,7 @@ if (USE_LLVM_BITCODE)
     EMBED_LLVM_BITCODE_IN_CPP ( "${llvm_ops_srcs}" "_host" "osl_llvm_compiled_ops" lib_src "" "${include_dirs}")
 
     set (rs_dependent_ops_srcs
-         opmatrix.cpp opfmt.cpp optexture.cpp pointcloud.cpp
+         opmatrix.cpp opfmt.cpp optexture.cpp pointcloud.cpp opclosure.cpp
          )
     # Achieve the effect of absorbing osl_llvm_compiled_ops by adding its 
     # sources to rs_dependent_ops_srcs which avoids having to do it at runtime.
@@ -217,6 +217,7 @@ if (USE_LLVM_BITCODE)
             ${CMAKE_SOURCE_DIR}/src/liboslexec/opmatrix.cpp
             ${CMAKE_SOURCE_DIR}/src/liboslexec/optexture.cpp
             ${CMAKE_SOURCE_DIR}/src/liboslexec/pointcloud.cpp
+            ${CMAKE_SOURCE_DIR}/src/liboslexec/opclosure.cpp
             ${CMAKE_SOURCE_DIR}/src/liboslnoise/gabornoise.cpp
             ${CMAKE_SOURCE_DIR}/src/liboslnoise/simplexnoise.cpp
         )

src/liboslexec/builtindecl.h

@@ -109,29 +109,14 @@
     DECL(osl_##name##_dvdvv, "xXXX")
 
 
-
-#ifndef __CUDA_ARCH__
-DECL(osl_add_closure_closure, "CXCC")
-DECL(osl_mul_closure_float, "CXCf")
-DECL(osl_mul_closure_color, "CXCc")
-DECL(osl_allocate_closure_component, "CXii")
-DECL(osl_allocate_weighted_closure_component, "CXiiX")
-DECL(osl_closure_to_string, "sXC")
-DECL(osl_closure_to_ustringhash, "hXC")
-#else
-// TODO: Figure out why trying to match the signatures between host and device
-//       definitions fails with 'LLVM had to make a cast' assertion failure.
-//
-//       In the meantime, use a signature that matches the definitions in rend_lib.cu,
-//       where void* is used instead of ClosureColor* and ShaderGlobals*.
 DECL(osl_add_closure_closure, "XXXX")
 DECL(osl_mul_closure_float, "XXXf")
-DECL(osl_mul_closure_color, "XXXc")
+DECL(osl_mul_closure_color, "XXXX")
 DECL(osl_allocate_closure_component, "XXii")
 DECL(osl_allocate_weighted_closure_component, "XXiiX")
 DECL(osl_closure_to_string, "sXX")
 DECL(osl_closure_to_ustringhash, "hXX")
-#endif
+
 DECL(osl_format, "hh*")
 DECL(osl_gen_ustringhash_pod, "hs")
 DECL(osl_gen_ustring, "sh")

src/liboslexec/llvm_gen.cpp

@@ -3863,19 +3863,15 @@ LLVMGEN(llvm_gen_closure)
                                      id_int, size_int);
     llvm::Value* comp_void_ptr = return_ptr;
 
-    // For the weighted closures, we need a surrounding "if" so that it's safe
-    // for osl_allocate_weighted_closure_component to return NULL (unless we
-    // know for sure that it's constant weighted and that the weight is
-    // not zero).
-    llvm::BasicBlock* next_block = NULL;
-    if (weighted && !(weight->is_constant() && !rop.is_zero(*weight))) {
-        llvm::BasicBlock* notnull_block = rop.ll.new_basic_block(
-            "non_null_closure");
-        next_block        = rop.ll.new_basic_block("");
-        llvm::Value* cond = rop.ll.op_ne(return_ptr, rop.ll.void_ptr_null());
-        rop.ll.op_branch(cond, notnull_block, next_block);
-        // new insert point is nonnull_block
-    }
+    // We need a surrounding "if" so that it's safe for closure allocation to
+    // return NULL, either because it has zero weight, or renderer services ran
+    // out of memory in the closure pool.
+    llvm::BasicBlock* notnull_block = rop.ll.new_basic_block(
+        "non_null_closure");
+    llvm::BasicBlock* next_block = rop.ll.new_basic_block("");
+    llvm::Value* cond = rop.ll.op_ne(return_ptr, rop.ll.void_ptr_null());
+    rop.ll.op_branch(cond, notnull_block, next_block);
+    // new insert point is nonnull_block
 
     llvm::Value* comp_ptr
         = rop.ll.ptr_cast(comp_void_ptr, rop.llvm_type_closure_component_ptr());

src/liboslexec/opclosure.cpp

@@ -7,72 +7,120 @@
 
 #include "oslexec_pvt.h"
 #include <OSL/genclosure.h>
+#include <OSL/rs_free_function.h>
 
 
 OSL_NAMESPACE_BEGIN
 namespace pvt {
 
 
 
-OSL_SHADEOP const ClosureColor*
-osl_add_closure_closure(ShaderGlobals* sg, const ClosureColor* a,
-                        const ClosureColor* b)
+OSL_SHADEOP OSL_HOSTDEVICE const void*
+osl_add_closure_closure(OpaqueExecContextPtr oec, const void* a_,
+                        const void* b_)
 {
+    const ClosureColor* a = (const ClosureColor*)a_;
+    const ClosureColor* b = (const ClosureColor*)b_;
     if (a == NULL)
         return b;
     if (b == NULL)
         return a;
-    return sg->context->closure_add_allot(a, b);
+    ClosureAdd* add = (ClosureAdd*)rs_allocate_closure(oec, sizeof(ClosureAdd),
+                                                       alignof(ClosureAdd));
+    if (add) {
+        add->id       = ClosureColor::ADD;
+        add->closureA = a;
+        add->closureB = b;
+    }
+    return add;
 }
 
 
-OSL_SHADEOP const ClosureColor*
-osl_mul_closure_color(ShaderGlobals* sg, ClosureColor* a, const Color3* w)
+OSL_SHADEOP OSL_HOSTDEVICE const void*
+osl_mul_closure_color(OpaqueExecContextPtr oec, const void* a_, const void* w_)
 {
+    const ClosureColor* a = (const ClosureColor*)a_;
+    const Color3* w       = (const Color3*)w_;
     if (a == NULL)
         return NULL;
     if (w->x == 0.0f && w->y == 0.0f && w->z == 0.0f)
         return NULL;
     if (w->x == 1.0f && w->y == 1.0f && w->z == 1.0f)
         return a;
-    return sg->context->closure_mul_allot(*w, a);
+    ClosureMul* mul = (ClosureMul*)rs_allocate_closure(oec, sizeof(ClosureMul),
+                                                       alignof(ClosureMul));
+    if (mul) {
+        mul->id      = ClosureColor::MUL;
+        mul->weight  = *w;
+        mul->closure = a;
+    }
+    return mul;
 }
 
 
-OSL_SHADEOP const ClosureColor*
-osl_mul_closure_float(ShaderGlobals* sg, ClosureColor* a, float w)
+OSL_SHADEOP OSL_HOSTDEVICE const void*
+osl_mul_closure_float(OpaqueExecContextPtr oec, const void* a_, float w)
 {
+    const ClosureColor* a = (const ClosureColor*)a_;
     if (a == NULL)
         return NULL;
     if (w == 0.0f)
         return NULL;
     if (w == 1.0f)
         return a;
-    return sg->context->closure_mul_allot(w, a);
+    ClosureMul* mul = (ClosureMul*)rs_allocate_closure(oec, sizeof(ClosureMul),
+                                                       alignof(ClosureMul));
+    if (mul) {
+        mul->id      = ClosureColor::MUL;
+        mul->weight  = Color3(w);
+        mul->closure = a;
+    }
+    return mul;
 }
 
 
-OSL_SHADEOP ClosureComponent*
-osl_allocate_closure_component(ShaderGlobals* sg, int id, int size)
+OSL_SHADEOP OSL_HOSTDEVICE void*
+osl_allocate_closure_component(OpaqueExecContextPtr oec, int id, int size)
 {
-    return sg->context->closure_component_allot(id, size, Color3(1.0f));
+    // Allocate the component and the mul back to back
+    const size_t needed = sizeof(ClosureComponent) + size;
+    ClosureComponent* comp
+        = (ClosureComponent*)rs_allocate_closure(oec, needed,
+                                                 alignof(ClosureComponent));
+    if (comp) {
+        comp->id = id;
+        comp->w  = Color3(1.0f);
+    }
+    return comp;
 }
 
 
 
-OSL_SHADEOP ClosureColor*
-osl_allocate_weighted_closure_component(ShaderGlobals* sg, int id, int size,
-                                        const Color3* w)
+OSL_SHADEOP OSL_HOSTDEVICE void*
+osl_allocate_weighted_closure_component(OpaqueExecContextPtr oec, int id,
+                                        int size, const void* w_)
 {
+    const Color3* w = (const Color3*)w_;
     if (w->x == 0.0f && w->y == 0.0f && w->z == 0.0f)
         return NULL;
-    return sg->context->closure_component_allot(id, size, *w);
+    // Allocate the component and the mul back to back
+    const size_t needed = sizeof(ClosureComponent) + size;
+    ClosureComponent* comp
+        = (ClosureComponent*)rs_allocate_closure(oec, needed,
+                                                 alignof(ClosureComponent));
+    if (comp) {
+        comp->id = id;
+        comp->w  = *w;
+    }
+    return comp;
 }
 
 // Deprecated, remove when conversion from ustring to ustringhash is finished
 OSL_SHADEOP const char*
-osl_closure_to_string(ShaderGlobals* sg, ClosureColor* c)
+osl_closure_to_string(OpaqueExecContextPtr oec, const void* c_)
 {
+    ShaderGlobals* sg     = (ShaderGlobals*)oec;
+    const ClosureColor* c = (const ClosureColor*)c_;
     // Special case for printing closures
     std::ostringstream stream;
     stream.imbue(std::locale::classic());  // force C locale
@@ -82,8 +130,10 @@ osl_closure_to_string(ShaderGlobals* sg, ClosureColor* c)
 }
 
 OSL_SHADEOP ustringhash_pod
-osl_closure_to_ustringhash(ShaderGlobals* sg, ClosureColor* c)
+osl_closure_to_ustringhash(OpaqueExecContextPtr oec, const void* c_)
 {
+    ShaderGlobals* sg     = (ShaderGlobals*)oec;
+    const ClosureColor* c = (const ClosureColor*)c_;
     // Special case for printing closures
     std::ostringstream stream;
     stream.imbue(std::locale::classic());  // force C locale

src/liboslexec/oslexec_pvt.h

@@ -2242,49 +2242,11 @@ class OSLEXECPUBLIC ShadingContext {
     }
 #endif
 
-    ClosureComponent* closure_component_allot(int id, size_t prim_size,
-                                              const Color3& w)
+    void* allocate_closure(size_t size, size_t alignment)
     {
-        // Allocate the component and the mul back to back
-        size_t needed          = sizeof(ClosureComponent) + prim_size;
-        ClosureComponent* comp = (ClosureComponent*)m_closure_pool.alloc(
-            needed, alignof(ClosureComponent));
-        comp->id = id;
-        comp->w  = w;
-        return comp;
+        return m_closure_pool.alloc(size, alignment);
     }
 
-    ClosureMul* closure_mul_allot(const Color3& w, const ClosureColor* c)
-    {
-        ClosureMul* mul = (ClosureMul*)m_closure_pool.alloc(sizeof(ClosureMul),
-                                                            alignof(ClosureMul));
-        mul->id         = ClosureColor::MUL;
-        mul->weight     = w;
-        mul->closure    = c;
-        return mul;
-    }
-
-    ClosureMul* closure_mul_allot(float w, const ClosureColor* c)
-    {
-        ClosureMul* mul = (ClosureMul*)m_closure_pool.alloc(sizeof(ClosureMul),
-                                                            alignof(ClosureMul));
-        mul->id         = ClosureColor::MUL;
-        mul->weight.setValue(w, w, w);
-        mul->closure = c;
-        return mul;
-    }
-
-    ClosureAdd* closure_add_allot(const ClosureColor* a, const ClosureColor* b)
-    {
-        ClosureAdd* add = (ClosureAdd*)m_closure_pool.alloc(sizeof(ClosureAdd),
-                                                            alignof(ClosureAdd));
-        add->id         = ClosureColor::ADD;
-        add->closureA   = a;
-        add->closureB   = b;
-        return add;
-    }
-
-
     /// Find the named symbol in the (already-executed!) stack of shaders of
     /// the given use. If a layer is given, search just that layer. If no
     /// layer is specified, priority is given to later laters over earlier

src/liboslexec/rs_fallback.cpp

@@ -8,6 +8,7 @@
 
 #include <OSL/journal.h>
 
+#include "oslexec_pvt.h"
 
 // Fallback is to reroute calls back through the virtual function
 // based RendererServices from ShaderGlobals.
@@ -316,6 +317,18 @@ rs_trace_get(OSL::OpaqueExecContextPtr exec_ctx, OSL::ustringhash name,
 #endif
 }
 
+OSL_RSOP OSL_HOSTDEVICE void*
+rs_allocate_closure(OSL::OpaqueExecContextPtr exec_ctx, size_t size,
+                    size_t alignment)
+{
+#ifndef __CUDA_ARCH__
+    auto sg = get_sg(exec_ctx);
+    return sg->context->allocate_closure(size, alignment);
+#else
+    return nullptr;
+#endif
+}
+
 OSL_RSOP OSL_HOSTDEVICE void
 rs_errorfmt(OSL::OpaqueExecContextPtr exec_ctx,
             OSL::ustringhash fmt_specification, int32_t count,

src/liboslexec/shadeimage.cpp

@@ -116,9 +116,6 @@ shade_image(ShadingSystem& shadingsys, ShaderGroup& group,
             // That also implies that our normal points to (0,0,1)
             sg.N  = Vec3(0, 0, 1);
             sg.Ng = Vec3(0, 0, 1);
-            // In our SimpleRenderer, the "renderstate" itself just a pointer to
-            // the ShaderGlobals.
-            // sg.renderstate = &sg;
         }
 
         // Loop over all pixels in the image (in x and y)...

src/osltoy/osltoyrenderer.cpp

@@ -115,9 +115,6 @@ OSLToyRenderer::OSLToyRenderer()
     // That also implies that our normal points to (0,0,1)
     sg.N  = Vec3(0, 0, 1);
     sg.Ng = Vec3(0, 0, 1);
-    // In our SimpleRenderer, the "renderstate" itself just a pointer to
-    // the ShaderGlobals.
-    // sg.renderstate = &sg;
 }