removed referencing workgroupID in scans

keptsecret · keptsecret · commit a9930a025b4b · 2025-05-15T16:00:34.000+07:00
diff --git a/examples_tests b/examples_tests
@@ -1 +1 @@
-Subproject commit 81238adaecbd8d717bdab0dd73e08e2938a794c6
+Subproject commit 1de31ddfd725009bd650f1fe80f1c4a8c2e6a14a
diff --git a/include/nbl/builtin/hlsl/workgroup2/arithmetic_config.hlsl b/include/nbl/builtin/hlsl/workgroup2/arithmetic_config.hlsl
@@ -77,6 +77,16 @@ struct ArithmeticConfiguration
     NBL_CONSTEXPR_STATIC_INLINE uint16_t ElementCount = conditional_value<LevelCount==1,uint16_t,0,conditional_value<LevelCount==3,uint16_t,SubgroupSize*ItemsPerInvocation_2,0>::value + SubgroupSize*ItemsPerInvocation_1>::value;
 };
 
+template<class T>
+struct is_configuration : bool_constant<false> {};
+
+template<uint16_t W, uint16_t S, uint16_t I>
+struct is_configuration<ArithmeticConfiguration<W,S,I> > : bool_constant<true> {};
+
+template<typename T>
+NBL_CONSTEXPR bool is_configuration_v = is_configuration<T>::value;
+
+
 }
 }
 }
diff --git a/include/nbl/builtin/hlsl/workgroup2/shared_scan.hlsl b/include/nbl/builtin/hlsl/workgroup2/shared_scan.hlsl
@@ -43,9 +43,9 @@ struct reduce<Config, BinOp, 1, device_capabilities>
 
         subgroup2::reduction<params_t> reduction;
         vector_t value;
-        dataAccessor.get(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value);
+        dataAccessor.get(workgroup::SubgroupContiguousIndex(), value);
         value = reduction(value);
-        dataAccessor.set(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value);   // can be safely merged with top line?
+        dataAccessor.set(workgroup::SubgroupContiguousIndex(), value);
     }
 };
 
@@ -63,7 +63,7 @@ struct scan<Config, BinOp, Exclusive, 1, device_capabilities>
         using params_t = subgroup2::ArithmeticParams<config_t, BinOp, Config::ItemsPerInvocation_0, device_capabilities>;
 
         vector_t value;
-        dataAccessor.get(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value);
+        dataAccessor.get(workgroup::SubgroupContiguousIndex(), value);
         if (Exclusive)
         {
             subgroup2::exclusive_scan<params_t> excl_scan;
@@ -74,7 +74,7 @@ struct scan<Config, BinOp, Exclusive, 1, device_capabilities>
             subgroup2::inclusive_scan<params_t> incl_scan;
             value = incl_scan(value);
         }
-        dataAccessor.set(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value);   // can be safely merged with above lines?
+        dataAccessor.set(workgroup::SubgroupContiguousIndex(), value);   // can be safely merged with above lines?
     }
 };
 
@@ -101,7 +101,7 @@ struct reduce<Config, BinOp, 2, device_capabilities>
         [unroll]
         for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)
         {
-            dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
+            dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
             scan_local[idx] = reduction0(scan_local[idx]);
             if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)
             {
@@ -131,7 +131,7 @@ struct reduce<Config, BinOp, 2, device_capabilities>
         {
             scalar_t reduce_val;
             scratchAccessor.get(glsl::gl_SubgroupInvocationID(),reduce_val);
-            dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);
+            dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);
         }
     }
 };
@@ -158,7 +158,7 @@ struct scan<Config, BinOp, Exclusive, 2, device_capabilities>
         [unroll]
         for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)
         {
-            dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
+            dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
             scan_local[idx] = inclusiveScan0(scan_local[idx]);
             if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)
             {
@@ -204,7 +204,7 @@ struct scan<Config, BinOp, Exclusive, 2, device_capabilities>
                 for (uint32_t i = 0; i < Config::ItemsPerInvocation_0; i++)
                     scan_local[idx][i] = binop(left, scan_local[idx][i]);
             }
-            dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
+            dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
         }
     }
 };
@@ -234,7 +234,7 @@ struct reduce<Config, BinOp, 3, device_capabilities>
         [unroll]
         for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)
         {
-            dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
+            dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
             scan_local[idx] = reduction0(scan_local[idx]);
             if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)
             {
@@ -281,7 +281,7 @@ struct reduce<Config, BinOp, 3, device_capabilities>
         {
             scalar_t reduce_val;
             scratchAccessor.get(glsl::gl_SubgroupInvocationID(),reduce_val);
-            dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);
+            dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);
         }
     }
 };
@@ -310,7 +310,7 @@ struct scan<Config, BinOp, Exclusive, 3, device_capabilities>
         [unroll]
         for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)
         {
-            dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
+            dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
             scan_local[idx] = inclusiveScan0(scan_local[idx]);
             if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)
             {
@@ -384,7 +384,7 @@ struct scan<Config, BinOp, Exclusive, 3, device_capabilities>
                 for (uint32_t i = 0; i < Config::ItemsPerInvocation_0; i++)
                     scan_local[idx][i] = binop(left, scan_local[idx][i]);
             }
-            dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
+            dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);
         }
     }
 };

Original file line number	Diff line number	Diff line change
`@@ -77,6 +77,16 @@ struct ArithmeticConfiguration`
`77`	`77`	`NBL_CONSTEXPR_STATIC_INLINE uint16_t ElementCount = conditional_value<LevelCount==1,uint16_t,0,conditional_value<LevelCount==3,uint16_t,SubgroupSizeItemsPerInvocation_2,0>::value + SubgroupSizeItemsPerInvocation_1>::value;`
`78`	`78`	`};`
`79`	`79`
	`80`	`+template<class T>`
	`81`	`+struct is_configuration : bool_constant<false> {};`
	`82`	`+`
	`83`	`+template<uint16_t W, uint16_t S, uint16_t I>`
	`84`	`+struct is_configuration<ArithmeticConfiguration<W,S,I> > : bool_constant<true> {};`
	`85`	`+`
	`86`	`+template<typename T>`
	`87`	`+NBL_CONSTEXPR bool is_configuration_v = is_configuration<T>::value;`
	`88`	`+`
	`89`	`+`
`80`	`90`	`}`
`81`	`91`	`}`
`82`	`92`	`}`
Original file line number	Diff line number	Diff line change
`@@ -43,9 +43,9 @@ struct reduce<Config, BinOp, 1, device_capabilities>`
`43`	`43`
`44`	`44`	`subgroup2::reduction<params_t> reduction;`
`45`	`45`	`vector_t value;`
`46`		`- dataAccessor.get(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value);`
	`46`	`+ dataAccessor.get(workgroup::SubgroupContiguousIndex(), value);`
`47`	`47`	`value = reduction(value);`
`48`		`- dataAccessor.set(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value); // can be safely merged with top line?`
	`48`	`+ dataAccessor.set(workgroup::SubgroupContiguousIndex(), value);`
`49`	`49`	`}`
`50`	`50`	`};`
`51`	`51`
`@@ -63,7 +63,7 @@ struct scan<Config, BinOp, Exclusive, 1, device_capabilities>`
`63`	`63`	`using params_t = subgroup2::ArithmeticParams<config_t, BinOp, Config::ItemsPerInvocation_0, device_capabilities>;`
`64`	`64`
`65`	`65`	`vector_t value;`
`66`		`- dataAccessor.get(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value);`
	`66`	`+ dataAccessor.get(workgroup::SubgroupContiguousIndex(), value);`
`67`	`67`	`if (Exclusive)`
`68`	`68`	`{`
`69`	`69`	`subgroup2::exclusive_scan<params_t> excl_scan;`
`@@ -74,7 +74,7 @@ struct scan<Config, BinOp, Exclusive, 1, device_capabilities>`
`74`	`74`	`subgroup2::inclusive_scan<params_t> incl_scan;`
`75`	`75`	`value = incl_scan(value);`
`76`	`76`	`}`
`77`		`- dataAccessor.set(glsl::gl_WorkGroupID().x * Config::SubgroupSize + workgroup::SubgroupContiguousIndex(), value); // can be safely merged with above lines?`
	`77`	`+ dataAccessor.set(workgroup::SubgroupContiguousIndex(), value); // can be safely merged with above lines?`
`78`	`78`	`}`
`79`	`79`	`};`
`80`	`80`
`@@ -101,7 +101,7 @@ struct reduce<Config, BinOp, 2, device_capabilities>`
`101`	`101`	`[unroll]`
`102`	`102`	`for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)`
`103`	`103`	`{`
`104`		`- dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
	`104`	`+ dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
`105`	`105`	`scan_local[idx] = reduction0(scan_local[idx]);`
`106`	`106`	`if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)`
`107`	`107`	`{`
`@@ -131,7 +131,7 @@ struct reduce<Config, BinOp, 2, device_capabilities>`
`131`	`131`	`{`
`132`	`132`	`scalar_t reduce_val;`
`133`	`133`	`scratchAccessor.get(glsl::gl_SubgroupInvocationID(),reduce_val);`
`134`		`- dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);`
	`134`	`+ dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);`
`135`	`135`	`}`
`136`	`136`	`}`
`137`	`137`	`};`
`@@ -158,7 +158,7 @@ struct scan<Config, BinOp, Exclusive, 2, device_capabilities>`
`158`	`158`	`[unroll]`
`159`	`159`	`for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)`
`160`	`160`	`{`
`161`		`- dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
	`161`	`+ dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
`162`	`162`	`scan_local[idx] = inclusiveScan0(scan_local[idx]);`
`163`	`163`	`if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)`
`164`	`164`	`{`
`@@ -204,7 +204,7 @@ struct scan<Config, BinOp, Exclusive, 2, device_capabilities>`
`204`	`204`	`for (uint32_t i = 0; i < Config::ItemsPerInvocation_0; i++)`
`205`	`205`	`scan_local[idx][i] = binop(left, scan_local[idx][i]);`
`206`	`206`	`}`
`207`		`- dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
	`207`	`+ dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
`208`	`208`	`}`
`209`	`209`	`}`
`210`	`210`	`};`
`@@ -234,7 +234,7 @@ struct reduce<Config, BinOp, 3, device_capabilities>`
`234`	`234`	`[unroll]`
`235`	`235`	`for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)`
`236`	`236`	`{`
`237`		`- dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
	`237`	`+ dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
`238`	`238`	`scan_local[idx] = reduction0(scan_local[idx]);`
`239`	`239`	`if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)`
`240`	`240`	`{`
`@@ -281,7 +281,7 @@ struct reduce<Config, BinOp, 3, device_capabilities>`
`281`	`281`	`{`
`282`	`282`	`scalar_t reduce_val;`
`283`	`283`	`scratchAccessor.get(glsl::gl_SubgroupInvocationID(),reduce_val);`
`284`		`- dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);`
	`284`	`+ dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, reduce_val);`
`285`	`285`	`}`
`286`	`286`	`}`
`287`	`287`	`};`
`@@ -310,7 +310,7 @@ struct scan<Config, BinOp, Exclusive, 3, device_capabilities>`
`310`	`310`	`[unroll]`
`311`	`311`	`for (uint32_t idx = 0, virtualInvocationIndex = invocationIndex; idx < Config::VirtualWorkgroupSize / Config::WorkgroupSize; idx++)`
`312`	`312`	`{`
`313`		`- dataAccessor.get(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
	`313`	`+ dataAccessor.get(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
`314`	`314`	`scan_local[idx] = inclusiveScan0(scan_local[idx]);`
`315`	`315`	`if (glsl::gl_SubgroupInvocationID()==Config::SubgroupSize-1)`
`316`	`316`	`{`
`@@ -384,7 +384,7 @@ struct scan<Config, BinOp, Exclusive, 3, device_capabilities>`
`384`	`384`	`for (uint32_t i = 0; i < Config::ItemsPerInvocation_0; i++)`
`385`	`385`	`scan_local[idx][i] = binop(left, scan_local[idx][i]);`
`386`	`386`	`}`
`387`		`- dataAccessor.set(glsl::gl_WorkGroupID().x * Config::VirtualWorkgroupSize + idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
	`387`	`+ dataAccessor.set(idx * Config::WorkgroupSize + virtualInvocationIndex, scan_local[idx]);`
`388`	`388`	`}`
`389`	`389`	`}`
`390`	`390`	`};`