Add support for scalar rcp() intrinsic for GLSL (#918)

4 files changed, 45 insertions, 9 deletions

diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang
index a12e45c36..40bd02597 100644
--- a/source/slang/hlsl.meta.slang
+++ b/source/slang/hlsl.meta.slang
@@ -1133,7 +1133,11 @@ __generic<T : __BuiltinFloatingPointType, let N : int> vector<T,N> radians(vecto
 __generic<T : __BuiltinFloatingPointType, let N : int, let M : int> matrix<T,N,M> radians(matrix<T,N,M> x);
 
 // Approximate reciprocal
-__generic<T : __BuiltinFloatingPointType> T rcp(T x);
+__generic<T : __BuiltinFloatingPointType>
+__target_intrinsic(glsl, "1.0/($0)")
+T rcp(T x);
+
+// TODO: vector and matrix approx. reciprocals needto be deconstructed for GLSL
 __generic<T : __BuiltinFloatingPointType, let N : int> vector<T,N> rcp(vector<T,N> x);
 __generic<T : __BuiltinFloatingPointType, let N : int, let M : int> matrix<T,N,M> rcp(matrix<T,N,M> x);
 
diff --git a/source/slang/hlsl.meta.slang.h b/source/slang/hlsl.meta.slang.h
index a82a733ac..e40e12962 100644
--- a/source/slang/hlsl.meta.slang.h
+++ b/source/slang/hlsl.meta.slang.h
@@ -1209,9 +1209,17 @@ SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int> vector<T,N> ra
 SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int, let M : int> matrix<T,N,M> radians(matrix<T,N,M> x);\n")
 SLANG_RAW("\n")
 SLANG_RAW("// Approximate reciprocal\n")
-SLANG_RAW("__generic<T : __BuiltinFloatingPointType> T rcp(T x);\n")
-SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int> vector<T,N> rcp(vector<T,N> x);\n")
-SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int, let M : int> matrix<T,N,M> rcp(matrix<T,N,M> x);\n")
+SLANG_RAW("__generic<T : __BuiltinFloatingPointType>\n")
+SLANG_RAW("__target_intrinsic(glsl, \"1.0/($0)\")\n")
+SLANG_RAW("T rcp(T x);\n")
+SLANG_RAW("\n")
+SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int>\n")
+SLANG_RAW("__target_intrinsic(glsl, \"1.0/($0)\")\n")
+SLANG_RAW("vector<T,N> rcp(vector<T,N> x);\n")
+SLANG_RAW("\n")
+SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int, let M : int>\n")
+SLANG_RAW("__target_intrinsic(glsl, \"1.0/($0)\")\n")
+SLANG_RAW("matrix<T,N,M> rcp(matrix<T,N,M> x);\n")
 SLANG_RAW("\n")
 SLANG_RAW("// Reflect incident vector across plane with given normal\n")
 SLANG_RAW("__generic<T : __BuiltinFloatingPointType, let N : int>\n")
@@ -1531,7 +1539,7 @@ for (int aa = 0; aa < kBaseBufferAccessLevelCount; ++aa)
 
     sb << "};\n";
 }
-SLANG_RAW("#line 1458 \"hlsl.meta.slang\"")
+SLANG_RAW("#line 1466 \"hlsl.meta.slang\"")
 SLANG_RAW("\n")
 SLANG_RAW("\n")
 SLANG_RAW("\n")
diff --git a/tests/bugs/vk-structured-buffer-load.hlsl b/tests/bugs/vk-structured-buffer-load.hlsl
index daedc5b66..a5f518f93 100644
--- a/tests/bugs/vk-structured-buffer-load.hlsl
+++ b/tests/bugs/vk-structured-buffer-load.hlsl
@@ -1,5 +1,7 @@
 //TEST:CROSS_COMPILE: -profile lib_6_3 -entry HitMain -stage closesthit -target spirv-assembly
 
+#define USE_RCP 0
+
 struct ParameterBlockTest
 {
     SamplerState sam;
@@ -19,7 +21,13 @@ void HitMain(inout RayHitInfoPacked RayData, BuiltInTriangleIntersectionAttribut
     float HitT = RayTCurrent();
     RayData.PackedHitInfoA.x = HitT;
     uint offs = 0;
+    uint use_rcp = USE_RCP;
     float offsfloat = gParamBlock.sbuf.Load(offs);
-    
-    RayData.PackedHitInfoA.y = rsqrt(offsfloat);
+
+    use_rcp |= HitT > 0.0;
+
+    if (use_rcp)
+        RayData.PackedHitInfoA.y = rcp(offsfloat);
+    else
+        RayData.PackedHitInfoA.y = rsqrt(offsfloat);
 }
diff --git a/tests/bugs/vk-structured-buffer-load.hlsl.glsl b/tests/bugs/vk-structured-buffer-load.hlsl.glsl
index 8060cc921..1cf0b00a7 100644
--- a/tests/bugs/vk-structured-buffer-load.hlsl.glsl
+++ b/tests/bugs/vk-structured-buffer-load.hlsl.glsl
@@ -29,11 +29,27 @@ void main()
     float HitT_0 = (gl_HitTNV);
     _S2.PackedHitInfoA_0.x = HitT_0;
 
+    const uint use_rcp_0 = uint(0);
+
     float offsfloat_0 = ((gParamBlock_sbuf_0)._data[(int(uint(0)))]);
 
-    float _S4 = inversesqrt(offsfloat_0);
 
-    _S2.PackedHitInfoA_0.y = _S4;
+
+    if(bool(use_rcp_0 | uint(HitT_0 > 0.00000000000000000000)))
+    {
+
+        float _S4 = (1.0/((offsfloat_0)));
+
+        _S2.PackedHitInfoA_0.y = _S4;
+
+    }
+    else
+    {
+        float _S5 = (inversesqrt((offsfloat_0)));
+
+        _S2.PackedHitInfoA_0.y = _S5;
+
+    }
 
     return;
 }