[DirectX] Remove the "checked" variants of dx.resource.load
#120778
Merged
+84
−83
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
We'd introduced separate versions of `llvm.dx.resource.load` with a
struct return to handle the CheckAccessFullyMapped case without making
the IR for the common case unnecessarily complicated. However, at this
point the common case is really `resource.getpointer`, so the ergonomics
of a simplified version of `load` don't actually gain us as much as the
cost of having multiple opcodes.
Drop the `dx.resource.loadchecked` functions and have `dx.resource.load`
consistently return `{element_type, i1}`.
|
@llvm/pr-subscribers-llvm-ir @llvm/pr-subscribers-backend-directx Author: Justin Bogner (bogner) ChangesWe'd introduced separate versions of Drop the Patch is 27.63 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/120778.diff 9 Files Affected:
diff --git a/llvm/docs/DirectX/DXILResources.rst b/llvm/docs/DirectX/DXILResources.rst
index 3971d3788b8a0d..406d68e3c9fada 100644
--- a/llvm/docs/DirectX/DXILResources.rst
+++ b/llvm/docs/DirectX/DXILResources.rst
@@ -296,8 +296,8 @@ instead. That is, ``llvm.dx.resource.load.typedbuffer`` from a
of 4 floats, and from ``Buffer<double2>`` a vector of two doubles, etc. The
operations are then expanded out to match DXIL's format during lowering.
-In cases where we need ``CheckAccessFullyMapped``, we have a second intrinsic
-that returns an anonymous struct with element-0 being the contained type, and
+In order to support ``CheckAccessFullyMapped``, we need these intrinsics to
+return an anonymous struct with element-0 being the contained type, and
element-1 being the ``i1`` result of a ``CheckAccessFullyMapped`` call. We
don't have a separate call to ``CheckAccessFullyMapped`` at all, since that's
the only operation that can possibly be done on this value. In practice this
@@ -317,8 +317,8 @@ HLSL source, but this actually matches DXC's behaviour in practice.
- Description
* - Return value
-
- - The contained type of the buffer
- - The data loaded from the buffer
+ - A structure of the contained type and the check bit
+ - The data loaded from the buffer and the check bit
* - ``%buffer``
- 0
- ``target(dx.TypedBuffer, ...)``
@@ -332,48 +332,22 @@ Examples:
.. code-block:: llvm
- %ret = call <4 x float>
+ %ret = call {<4 x float>, i1}
@llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_0_0_0t(
target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 %index)
- %ret = call float
+ %ret = call {float, i1}
@llvm.dx.resource.load.typedbuffer.f32.tdx.TypedBuffer_f32_0_0_0t(
target("dx.TypedBuffer", float, 0, 0, 0) %buffer, i32 %index)
- %ret = call <4 x i32>
+ %ret = call {<4 x i32>, i1}
@llvm.dx.resource.load.typedbuffer.v4i32.tdx.TypedBuffer_v4i32_0_0_0t(
target("dx.TypedBuffer", <4 x i32>, 0, 0, 0) %buffer, i32 %index)
- %ret = call <4 x half>
+ %ret = call {<4 x half>, i1}
@llvm.dx.resource.load.typedbuffer.v4f16.tdx.TypedBuffer_v4f16_0_0_0t(
target("dx.TypedBuffer", <4 x half>, 0, 0, 0) %buffer, i32 %index)
- %ret = call <2 x double>
+ %ret = call {<2 x double>, i1}
@llvm.dx.resource.load.typedbuffer.v2f64.tdx.TypedBuffer_v2f64_0_0t(
target("dx.TypedBuffer", <2 x double>, 0, 0, 0) %buffer, i32 %index)
-.. list-table:: ``@llvm.dx.resource.loadchecked.typedbuffer``
- :header-rows: 1
-
- * - Argument
- -
- - Type
- - Description
- * - Return value
- -
- - A structure of the contained type and the check bit
- - The data loaded from the buffer and the check bit
- * - ``%buffer``
- - 0
- - ``target(dx.TypedBuffer, ...)``
- - The buffer to load from
- * - ``%index``
- - 1
- - ``i32``
- - Index into the buffer
-
-.. code-block:: llvm
-
- %ret = call {<4 x float>, i1}
- @llvm.dx.resource.loadchecked.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_0_0_0t(
- target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 %index)
-
Texture and Typed Buffer Stores
-------------------------------
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index d31d5afe5145a7..dfdfda963b6273 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -31,9 +31,6 @@ def int_dx_resource_getpointer
: DefaultAttrsIntrinsic<[llvm_anyptr_ty], [llvm_any_ty, llvm_i32_ty],
[IntrNoMem]>;
def int_dx_resource_load_typedbuffer
- : DefaultAttrsIntrinsic<[llvm_any_ty], [llvm_any_ty, llvm_i32_ty],
- [IntrReadMem]>;
-def int_dx_resource_loadchecked_typedbuffer
: DefaultAttrsIntrinsic<[llvm_any_ty, llvm_i1_ty],
[llvm_any_ty, llvm_i32_ty], [IntrReadMem]>;
def int_dx_resource_store_typedbuffer
@@ -43,7 +40,7 @@ def int_dx_resource_store_typedbuffer
def int_dx_resource_updatecounter
: DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_any_ty, llvm_i8_ty],
[IntrInaccessibleMemOrArgMemOnly]>;
-
+
// Cast between target extension handle types and dxil-style opaque handles
def int_dx_resource_casthandle : Intrinsic<[llvm_any_ty], [llvm_any_ty]>;
@@ -105,7 +102,7 @@ def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrCon
def int_dx_wave_readlane : DefaultAttrsIntrinsic<[llvm_any_ty], [LLVMMatchType<0>, llvm_i32_ty], [IntrConvergent, IntrNoMem]>;
def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
-def int_dx_splitdouble : DefaultAttrsIntrinsic<[llvm_anyint_ty, LLVMMatchType<0>],
+def int_dx_splitdouble : DefaultAttrsIntrinsic<[llvm_anyint_ty, LLVMMatchType<0>],
[LLVMScalarOrSameVectorWidth<0, llvm_double_ty>], [IntrNoMem]>;
def int_dx_radians : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>], [IntrNoMem]>;
def int_dx_discard : DefaultAttrsIntrinsic<[], [llvm_i1_ty], []>;
diff --git a/llvm/lib/Target/DirectX/DXILOpLowering.cpp b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
index 4e01dd1145a55c..e0ee4d6d6b1305 100644
--- a/llvm/lib/Target/DirectX/DXILOpLowering.cpp
+++ b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
@@ -415,8 +415,16 @@ class OpLowerer {
}
}
- OldResult = cast<Instruction>(
- IRB.CreateExtractValue(Op, 0, OldResult->getName()));
+ if (OldResult->use_empty()) {
+ // Only the check bit was used, so we're done here.
+ OldResult->eraseFromParent();
+ return Error::success();
+ }
+
+ assert(OldResult->hasOneUse() &&
+ isa<ExtractValueInst>(*OldResult->user_begin()) &&
+ "Expected only use to be extract of first element");
+ OldResult = cast<Instruction>(*OldResult->user_begin());
OldTy = ST->getElementType(0);
}
@@ -723,9 +731,6 @@ class OpLowerer {
HasErrors |= lowerGetPointer(F);
break;
case Intrinsic::dx_resource_load_typedbuffer:
- HasErrors |= lowerTypedBufferLoad(F, /*HasCheckBit=*/false);
- break;
- case Intrinsic::dx_resource_loadchecked_typedbuffer:
HasErrors |= lowerTypedBufferLoad(F, /*HasCheckBit=*/true);
break;
case Intrinsic::dx_resource_store_typedbuffer:
diff --git a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
index 1ff8f09f066db5..837624935c5fae 100644
--- a/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
+++ b/llvm/lib/Target/DirectX/DXILResourceAccess.cpp
@@ -30,6 +30,9 @@ static void replaceTypedBufferAccess(IntrinsicInst *II,
"Unexpected typed buffer type");
Type *ContainedType = HandleType->getTypeParameter(0);
+ Type *LoadType =
+ StructType::get(ContainedType, Type::getInt1Ty(II->getContext()));
+
// We need the size of an element in bytes so that we can calculate the offset
// in elements given a total offset in bytes later.
Type *ScalarType = ContainedType->getScalarType();
@@ -81,13 +84,15 @@ static void replaceTypedBufferAccess(IntrinsicInst *II,
// We're storing a scalar, so we need to load the current value and only
// replace the relevant part.
auto *Load = Builder.CreateIntrinsic(
- ContainedType, Intrinsic::dx_resource_load_typedbuffer,
+ LoadType, Intrinsic::dx_resource_load_typedbuffer,
{II->getOperand(0), II->getOperand(1)});
+ auto *Struct = Builder.CreateExtractValue(Load, {0});
+
// If we have an offset from seeing a GEP earlier, use it.
Value *IndexOp = Current.Index
? Current.Index
: ConstantInt::get(Builder.getInt32Ty(), 0);
- V = Builder.CreateInsertElement(Load, V, IndexOp);
+ V = Builder.CreateInsertElement(Struct, V, IndexOp);
} else {
llvm_unreachable("Store to typed resource has invalid type");
}
@@ -101,8 +106,10 @@ static void replaceTypedBufferAccess(IntrinsicInst *II,
} else if (auto *LI = dyn_cast<LoadInst>(Current.Access)) {
IRBuilder<> Builder(LI);
Value *V = Builder.CreateIntrinsic(
- ContainedType, Intrinsic::dx_resource_load_typedbuffer,
+ LoadType, Intrinsic::dx_resource_load_typedbuffer,
{II->getOperand(0), II->getOperand(1)});
+ V = Builder.CreateExtractValue(V, {0});
+
if (Current.Index)
V = Builder.CreateExtractElement(V, Current.Index);
diff --git a/llvm/test/CodeGen/DirectX/BufferLoad.ll b/llvm/test/CodeGen/DirectX/BufferLoad.ll
index 7f1291bf4a5c8f..86e2217a8e76fa 100644
--- a/llvm/test/CodeGen/DirectX/BufferLoad.ll
+++ b/llvm/test/CodeGen/DirectX/BufferLoad.ll
@@ -17,8 +17,9 @@ define void @loadv4f32() {
; CHECK-NOT: %dx.resource.casthandle
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call <4 x float> @llvm.dx.resource.load.typedbuffer(
+ %load0 = call {<4 x float>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 0)
+ %data0 = extractvalue {<4 x float>, i1} %load0, 0
; The extract order depends on the users, so don't enforce that here.
; CHECK-DAG: [[VAL0_0:%.*]] = extractvalue %dx.types.ResRet.f32 [[DATA0]], 0
@@ -34,8 +35,9 @@ define void @loadv4f32() {
call void @scalar_user(float %data0_2)
; CHECK: [[DATA4:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 4, i32 undef)
- %data4 = call <4 x float> @llvm.dx.resource.load.typedbuffer(
+ %load4 = call {<4 x float>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 4)
+ %data4 = extractvalue {<4 x float>, i1} %load4, 0
; CHECK: extractvalue %dx.types.ResRet.f32 [[DATA4]], 0
; CHECK: extractvalue %dx.types.ResRet.f32 [[DATA4]], 1
@@ -48,8 +50,9 @@ define void @loadv4f32() {
call void @vector_user(<4 x float> %data4)
; CHECK: [[DATA12:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 12, i32 undef)
- %data12 = call <4 x float> @llvm.dx.resource.load.typedbuffer(
+ %load12 = call {<4 x float>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 12)
+ %data12 = extractvalue {<4 x float>, i1} %load12, 0
; CHECK: [[DATA12_3:%.*]] = extractvalue %dx.types.ResRet.f32 [[DATA12]], 3
%data12_3 = extractelement <4 x float> %data12, i32 3
@@ -70,8 +73,9 @@ define void @index_dynamic(i32 %bufindex, i32 %elemindex) {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[LOAD:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 %bufindex, i32 undef)
- %load = call <4 x float> @llvm.dx.resource.load.typedbuffer(
+ %load = call {<4 x float>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 %bufindex)
+ %data = extractvalue {<4 x float>, i1} %load, 0
; CHECK: [[ALLOCA:%.*]] = alloca [4 x float]
; CHECK: [[V0:%.*]] = extractvalue %dx.types.ResRet.f32 [[LOAD]], 0
@@ -89,10 +93,10 @@ define void @index_dynamic(i32 %bufindex, i32 %elemindex) {
;
; CHECK: [[PTR:%.*]] = getelementptr inbounds [4 x float], ptr [[ALLOCA]], i32 0, i32 %elemindex
; CHECK: [[X:%.*]] = load float, ptr [[PTR]]
- %data = extractelement <4 x float> %load, i32 %elemindex
+ %x = extractelement <4 x float> %data, i32 %elemindex
; CHECK: call void @scalar_user(float [[X]])
- call void @scalar_user(float %data)
+ call void @scalar_user(float %x)
ret void
}
@@ -105,8 +109,9 @@ define void @loadf32() {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call float @llvm.dx.resource.load.typedbuffer(
+ %load0 = call {float, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", float, 0, 0, 0) %buffer, i32 0)
+ %data0 = extractvalue {float, i1} %load0, 0
; CHECK: [[VAL0:%.*]] = extractvalue %dx.types.ResRet.f32 [[DATA0]], 0
; CHECK: call void @scalar_user(float [[VAL0]])
@@ -123,7 +128,7 @@ define void @loadv2f32() {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call <2 x float> @llvm.dx.resource.load.typedbuffer(
+ %data0 = call {<2 x float>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <2 x float>, 0, 0, 0) %buffer, i32 0)
ret void
@@ -137,7 +142,7 @@ define void @loadv4f32_checkbit() {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.f32 @dx.op.bufferLoad.f32(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call {<4 x float>, i1} @llvm.dx.resource.loadchecked.typedbuffer.f32(
+ %data0 = call {<4 x float>, i1} @llvm.dx.resource.load.typedbuffer.f32(
target("dx.TypedBuffer", <4 x float>, 0, 0, 0) %buffer, i32 0)
; CHECK: [[STATUS:%.*]] = extractvalue %dx.types.ResRet.f32 [[DATA0]], 4
@@ -158,7 +163,7 @@ define void @loadv4i32() {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.i32 @dx.op.bufferLoad.i32(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call <4 x i32> @llvm.dx.resource.load.typedbuffer(
+ %data0 = call {<4 x i32>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x i32>, 0, 0, 0) %buffer, i32 0)
ret void
@@ -172,7 +177,7 @@ define void @loadv4f16() {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.f16 @dx.op.bufferLoad.f16(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call <4 x half> @llvm.dx.resource.load.typedbuffer(
+ %data0 = call {<4 x half>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x half>, 0, 0, 0) %buffer, i32 0)
ret void
@@ -186,7 +191,7 @@ define void @loadv4i16() {
i32 0, i32 0, i32 1, i32 0, i1 false)
; CHECK: [[DATA0:%.*]] = call %dx.types.ResRet.i16 @dx.op.bufferLoad.i16(i32 68, %dx.types.Handle [[HANDLE]], i32 0, i32 undef)
- %data0 = call <4 x i16> @llvm.dx.resource.load.typedbuffer(
+ %data0 = call {<4 x i16>, i1} @llvm.dx.resource.load.typedbuffer(
target("dx.TypedBuffer", <4 x i16>, 0, 0, 0) %buffer, i32 0)
ret void
diff --git a/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll b/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll
index 9b7e7fd04f6052..8769e6ec66d8e0 100644
--- a/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll
+++ b/llvm/test/CodeGen/DirectX/ResourceAccess/load_typedbuffer.ll
@@ -15,17 +15,19 @@ define void @load_float4(i32 %index, i32 %elemindex) {
%ptr = call ptr @llvm.dx.resource.getpointer(
target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
- ; CHECK: %[[VALUE:.*]] = call <4 x float> @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VALUE:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
%vec_data = load <4 x float>, ptr %ptr
call void @use_float4(<4 x float> %vec_data)
- ; CHECK: %[[VALUE:.*]] = call <4 x float> @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VALUE:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
; CHECK: extractelement <4 x float> %[[VALUE]], i32 1
%y_ptr = getelementptr inbounds <4 x float>, ptr %ptr, i32 0, i32 1
%y_data = load float, ptr %y_ptr
call void @use_float(float %y_data)
- ; CHECK: %[[VALUE:.*]] = call <4 x float> @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VALUE:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
; CHECK: extractelement <4 x float> %[[VALUE]], i32 %elemindex
%dynamic = getelementptr inbounds <4 x float>, ptr %ptr, i32 0, i32 %elemindex
%dyndata = load float, ptr %dynamic
diff --git a/llvm/test/CodeGen/DirectX/ResourceAccess/store_typedbuffer.ll b/llvm/test/CodeGen/DirectX/ResourceAccess/store_typedbuffer.ll
index 17606408cadfff..0b7882ac722e87 100644
--- a/llvm/test/CodeGen/DirectX/ResourceAccess/store_typedbuffer.ll
+++ b/llvm/test/CodeGen/DirectX/ResourceAccess/store_typedbuffer.ll
@@ -18,21 +18,24 @@ define void @store_float4(<4 x float> %data, i32 %index, i32 %elemindex) {
; Store just the .x component
%scalar = extractelement <4 x float> %data, i32 0
- ; CHECK: %[[LOAD:.*]] = call <4 x float> @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
- ; CHECK: %[[INSERT:.*]] = insertelement <4 x float> %[[LOAD]], float %scalar, i32 0
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VEC:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
+ ; CHECK: %[[INSERT:.*]] = insertelement <4 x float> %[[VEC]], float %scalar, i32 0
; CHECK: call void @llvm.dx.resource.store.typedbuffer.tdx.TypedBuffer_v4f32_1_0_0t.v4f32(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index, <4 x float> %[[INSERT]])
store float %scalar, ptr %ptr
; Store just the .y component
- ; CHECK: %[[LOAD:.*]] = call <4 x float> @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
- ; CHECK: %[[INSERT:.*]] = insertelement <4 x float> %[[LOAD]], float %scalar, i32 1
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VEC:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
+ ; CHECK: %[[INSERT:.*]] = insertelement <4 x float> %[[VEC]], float %scalar, i32 1
; CHECK: call void @llvm.dx.resource.store.typedbuffer.tdx.TypedBuffer_v4f32_1_0_0t.v4f32(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index, <4 x float> %[[INSERT]])
%y_ptr = getelementptr inbounds i8, ptr %ptr, i32 4
store float %scalar, ptr %y_ptr
; Store to one of the elements dynamically
- ; CHECK: %[[LOAD:.*]] = call <4 x float> @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
- ; CHECK: %[[INSERT:.*]] = insertelement <4 x float> %[[LOAD]], float %scalar, i32 %elemindex
+ ; CHECK: %[[LOAD:.*]] = call { <4 x float>, i1 } @llvm.dx.resource.load.typedbuffer.v4f32.tdx.TypedBuffer_v4f32_1_0_0t(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index)
+ ; CHECK: %[[VEC:.*]] = extractvalue { <4 x float>, i1 } %[[LOAD]], 0
+ ; CHECK: %[[INSERT:.*]] = insertelement <4 x float> %[[VEC]], float %scalar, i32 %elemindex
; CHECK: call void @llvm.dx.resource.store.typedbuffer.tdx.TypedBuffer_v4f32_1_0_0t.v4f32(target("dx.TypedBuffer", <4 x float>, 1, 0, 0) %buffer, i32 %index, <4 x float> %[[INSERT]])
%dynamic = getelementptr inbounds <4 x float>, ptr %ptr, i32 0, i32 %elemindex
store float %scal...
[truncated]
|
llvm-beanz
approved these changes
Dec 20, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We'd introduced separate versions of
llvm.dx.resource.loadwith a struct return to handle the CheckAccessFullyMapped case without making the IR for the common case unnecessarily complicated. However, at this point the common case is reallyresource.getpointer, so the ergonomics of a simplified version ofloaddon't actually gain us as much as the cost of having multiple opcodes.Drop the
dx.resource.loadcheckedfunctions and havedx.resource.loadconsistently return{element_type, i1}.