Skip to content

Commit a057506

Browse files
JOE1994JOE1994
authored
[docs][NVPTX] Use opaque pointers in IR examples (NFC) (#81941)
Replace outdated typed-pointer notation in LLVM IR examples. NFC.
1 parent d0c2c0a commit a057506

File tree

1 file changed

+33
-39
lines changed

1 file changed

+33
-39
lines changed

‎llvm/docs/NVPTXUsage.rst

Lines changed: 33 additions & 39 deletions
Original file line numberDiff line numberDiff line change
@@ -53,15 +53,15 @@ function ``@my_kernel`` is callable from host code, but ``@my_fmad`` is not.
5353
ret float %add
5454
}
5555
56-
define void @my_kernel(float* %ptr) {
57-
%val = load float, float* %ptr
56+
define void @my_kernel(ptr %ptr) {
57+
%val = load float, ptr %ptr
5858
%ret = call float @my_fmad(float %val, float %val, float %val)
59-
store float %ret, float* %ptr
59+
store float %ret, ptr %ptr
6060
ret void
6161
}
6262
6363
!nvvm.annotations = !{!1}
64-
!1 = !{void (float*)* @my_kernel, !"kernel", i32 1}
64+
!1 = !{ptr @my_kernel, !"kernel", i32 1}
6565
6666
When compiled, the PTX kernel functions are callable by host-side code.
6767

@@ -140,10 +140,10 @@ These are overloaded intrinsics. You can use these on any pointer types.
140140

141141
.. code-block:: llvm
142142
143-
declare i8* @llvm.nvvm.ptr.global.to.gen.p0i8.p1i8(i8 addrspace(1)*)
144-
declare i8* @llvm.nvvm.ptr.shared.to.gen.p0i8.p3i8(i8 addrspace(3)*)
145-
declare i8* @llvm.nvvm.ptr.constant.to.gen.p0i8.p4i8(i8 addrspace(4)*)
146-
declare i8* @llvm.nvvm.ptr.local.to.gen.p0i8.p5i8(i8 addrspace(5)*)
143+
declare ptr @llvm.nvvm.ptr.global.to.gen.p0.p1(ptr addrspace(1))
144+
declare ptr @llvm.nvvm.ptr.shared.to.gen.p0.p3(ptr addrspace(3))
145+
declare ptr @llvm.nvvm.ptr.constant.to.gen.p0.p4(ptr addrspace(4))
146+
declare ptr @llvm.nvvm.ptr.local.to.gen.p0.p5(ptr addrspace(5))
147147
148148
Overview:
149149
"""""""""
@@ -168,10 +168,10 @@ These are overloaded intrinsics. You can use these on any pointer types.
168168

169169
.. code-block:: llvm
170170
171-
declare i8 addrspace(1)* @llvm.nvvm.ptr.gen.to.global.p1i8.p0i8(i8*)
172-
declare i8 addrspace(3)* @llvm.nvvm.ptr.gen.to.shared.p3i8.p0i8(i8*)
173-
declare i8 addrspace(4)* @llvm.nvvm.ptr.gen.to.constant.p4i8.p0i8(i8*)
174-
declare i8 addrspace(5)* @llvm.nvvm.ptr.gen.to.local.p5i8.p0i8(i8*)
171+
declare ptr addrspace(1) @llvm.nvvm.ptr.gen.to.global.p1.p0(ptr)
172+
declare ptr addrspace(3) @llvm.nvvm.ptr.gen.to.shared.p3.p0(ptr)
173+
declare ptr addrspace(4) @llvm.nvvm.ptr.gen.to.constant.p4.p0(ptr)
174+
declare ptr addrspace(5) @llvm.nvvm.ptr.gen.to.local.p5.p0(ptr)
175175
176176
Overview:
177177
"""""""""
@@ -436,35 +436,33 @@ The Kernel
436436
; Intrinsic to read X component of thread ID
437437
declare i32 @llvm.nvvm.read.ptx.sreg.tid.x() readnone nounwind
438438
439-
define void @kernel(float addrspace(1)* %A,
440-
float addrspace(1)* %B,
441-
float addrspace(1)* %C) {
439+
define void @kernel(ptr addrspace(1) %A,
440+
ptr addrspace(1) %B,
441+
ptr addrspace(1) %C) {
442442
entry:
443443
; What is my ID?
444444
%id = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x() readnone nounwind
445445
446446
; Compute pointers into A, B, and C
447-
%ptrA = getelementptr float, float addrspace(1)* %A, i32 %id
448-
%ptrB = getelementptr float, float addrspace(1)* %B, i32 %id
449-
%ptrC = getelementptr float, float addrspace(1)* %C, i32 %id
447+
%ptrA = getelementptr float, ptr addrspace(1) %A, i32 %id
448+
%ptrB = getelementptr float, ptr addrspace(1) %B, i32 %id
449+
%ptrC = getelementptr float, ptr addrspace(1) %C, i32 %id
450450
451451
; Read A, B
452-
%valA = load float, float addrspace(1)* %ptrA, align 4
453-
%valB = load float, float addrspace(1)* %ptrB, align 4
452+
%valA = load float, ptr addrspace(1) %ptrA, align 4
453+
%valB = load float, ptr addrspace(1) %ptrB, align 4
454454
455455
; Compute C = A + B
456456
%valC = fadd float %valA, %valB
457457
458458
; Store back to C
459-
store float %valC, float addrspace(1)* %ptrC, align 4
459+
store float %valC, ptr addrspace(1) %ptrC, align 4
460460
461461
ret void
462462
}
463463
464464
!nvvm.annotations = !{!0}
465-
!0 = !{void (float addrspace(1)*,
466-
float addrspace(1)*,
467-
float addrspace(1)*)* @kernel, !"kernel", i32 1}
465+
!0 = !{ptr @kernel, !"kernel", i32 1}
468466
469467
470468
We can use the LLVM ``llc`` tool to directly run the NVPTX code generator:
@@ -613,9 +611,7 @@ For the previous example, we have:
613611
.. code-block:: llvm
614612
615613
!nvvm.annotations = !{!0}
616-
!0 = !{void (float addrspace(1)*,
617-
float addrspace(1)*,
618-
float addrspace(1)*)* @kernel, !"kernel", i32 1}
614+
!0 = !{ptr @kernel, !"kernel", i32 1}
619615
620616
Here, we have a single metadata declaration in ``nvvm.annotations``. This
621617
metadata annotates our ``@kernel`` function with the ``kernel`` attribute.
@@ -820,35 +816,33 @@ Libdevice provides an ``__nv_powf`` function that we will use.
820816
; libdevice function
821817
declare float @__nv_powf(float, float)
822818
823-
define void @kernel(float addrspace(1)* %A,
824-
float addrspace(1)* %B,
825-
float addrspace(1)* %C) {
819+
define void @kernel(ptr addrspace(1) %A,
820+
ptr addrspace(1) %B,
821+
ptr addrspace(1) %C) {
826822
entry:
827823
; What is my ID?
828824
%id = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x() readnone nounwind
829825
830826
; Compute pointers into A, B, and C
831-
%ptrA = getelementptr float, float addrspace(1)* %A, i32 %id
832-
%ptrB = getelementptr float, float addrspace(1)* %B, i32 %id
833-
%ptrC = getelementptr float, float addrspace(1)* %C, i32 %id
827+
%ptrA = getelementptr float, ptr addrspace(1) %A, i32 %id
828+
%ptrB = getelementptr float, ptr addrspace(1) %B, i32 %id
829+
%ptrC = getelementptr float, ptr addrspace(1) %C, i32 %id
834830
835831
; Read A, B
836-
%valA = load float, float addrspace(1)* %ptrA, align 4
837-
%valB = load float, float addrspace(1)* %ptrB, align 4
832+
%valA = load float, ptr addrspace(1) %ptrA, align 4
833+
%valB = load float, ptr addrspace(1) %ptrB, align 4
838834
839835
; Compute C = pow(A, B)
840836
%valC = call float @__nv_powf(float %valA, float %valB)
841837
842838
; Store back to C
843-
store float %valC, float addrspace(1)* %ptrC, align 4
839+
store float %valC, ptr addrspace(1) %ptrC, align 4
844840
845841
ret void
846842
}
847843
848844
!nvvm.annotations = !{!0}
849-
!0 = !{void (float addrspace(1)*,
850-
float addrspace(1)*,
851-
float addrspace(1)*)* @kernel, !"kernel", i32 1}
845+
!0 = !{ptr @kernel, !"kernel", i32 1}
852846
853847
854848
To compile this kernel, we perform the following steps:

0 commit comments

Comments
 (0)