added emulation of "soa" data types via shared-memory

decl.cpp

@@ -168,11 +168,13 @@ DeclSpecs::GetBaseType(SourcePos pos) const {
     retType = lApplyTypeQualifiers(typeQualifiers, retType, pos);
 
     if (soaWidth > 0) {
+#if 0  /* see stmt.cpp in DeclStmt::EmitCode for work-around of SOAType Declaration */
         if (g->target->getISA() == Target::NVPTX)
         {
             Error(pos, "\"soa\" data types are currently not supported with \"nvptx\" target.");
             return NULL;
         }
+#endif
         const StructType *st = CastType<StructType>(retType);
 
         if (st == NULL) {

stmt.cpp

@@ -264,6 +264,7 @@ DeclStmt::EmitCode(FunctionEmitContext *ctx) const {
             return;
         }
 
+
         if (sym->storageClass == SC_STATIC) {
 
             if (g->target->getISA() == Target::NVPTX && !sym->type->IsConstType())
@@ -328,7 +329,7 @@ DeclStmt::EmitCode(FunctionEmitContext *ctx) const {
             // Tell the FunctionEmitContext about the variable
             ctx->EmitVariableDebugInfo(sym);
         }
-        else if (sym->type->IsUniformType() &&
+        else if ((sym->type->IsUniformType() || sym->type->IsSOAType()) &&
           /* NVPTX:
            * only non-constant uniform data types are stored in shared memory 
            * constant uniform are automatically promoted to varying 
@@ -357,6 +358,8 @@ DeclStmt::EmitCode(FunctionEmitContext *ctx) const {
                  * instead of compile-time constants 
                  */
                 nel *= at->GetElementCount();
+                if (sym->type->IsSOAType())
+                  nel *= sym->type->GetSOAWidth();
                 nat = new ArrayType(at->GetElementType(), nel);
                 variable = false;
               }

tests/soa-16.ispc

@@ -15,6 +15,16 @@ static void p(uniform float *uniform ptr) {
 }
 
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+#ifdef __NVPTX__  /* soa is converted to shared memory story for now, use smaller amount to check the test */
+    soa<4> Point pts[10];
+    for (uniform int i = 0; i < 40; ++i) {
+        pts[i].x = b*i;
+        pts[i].y[0] = 2*b*i;
+        pts[i].y[1] = 2*b*i+1;
+        pts[i].y[2] = 2*b*i+2;
+        pts[i].z = 3*b*i;
+    }
+#else
     soa<4> Point pts[30];
     for (uniform int i = 0; i < 120; ++i) {
         pts[i].x = b*i;
@@ -23,6 +33,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
         pts[i].y[2] = 2*b*i+2;
         pts[i].z = 3*b*i;
     }
+#endif
 
     float a = aFOO[programIndex]; 
     a *= -1;

tests/soa-17.ispc

@@ -16,6 +16,16 @@ static void p(uniform float *uniform ptr) {
 }
 
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
+#ifdef __NVPTX__  /* soa is converted to shared memory story for now, use smaller amount to check the test */
+    soa<4> Point pts[15];
+    for (uniform int i = 0; i < 60; ++i) {
+        pts[i].x = b*i;
+        pts[i].y[0] = 2*b*i;
+        pts[i].y[1] = 2*b*i+1;
+        pts[i].y[2] = 2*b*i+2;
+        pts[i].z = 3*b*i;
+    }
+#else
     soa<4> Point pts[40];
     for (uniform int i = 0; i < 160; ++i) {
         pts[i].x = b*i;
@@ -24,11 +34,14 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
         pts[i].y[2] = 2*b*i+2;
         pts[i].z = 3*b*i;
     }
+#endif
 
     float a = aFOO[programIndex]; 
     a *= -1;
     Point vp = { a, { 2*a, 3*a, 4*a }, {5*a} };
+#ifndef __NVPTX__
     assert(2+programIndex < 160);
+#endif
     pts[2+programIndex] = vp;
 
     RET[programIndex] = pts[programIndex].y[2];

tests/soa-3.ispc

@@ -6,6 +6,17 @@ export uniform int width() { return programCount; }
 export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
     float a = aFOO[programIndex]; 
 
+#ifdef __NVPTX__  /* soa is converted to shared memory story for now, use smaller amount to check the test */
+    soa<8> Point pts[4];
+//CO    uniform Point pts[80];
+    foreach (i = 0 ... 40) {
+        pts[i].x = b*i;
+        pts[i].y[0] = 2*b*i;
+        pts[i].y[1] = 2*b*i+1;
+        pts[i].y[2] = 2*b*i+2;
+        pts[i].z = 3*b*i;
+    }
+#else
     soa<8> Point pts[10];
 //CO    uniform Point pts[80];
     foreach (i = 0 ... 80) {
@@ -15,6 +26,7 @@ export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
         pts[i].y[2] = 2*b*i+2;
         pts[i].z = 3*b*i;
     }
+#endif
 
     assert(programCount < 80);
     RET[programIndex] = pts[programIndex].y[2];