[Libclc-dev] [PATCH v3 1/2] AMDGPU: Use clang intrinsics for workitem builtins
Jan Vesely via Libclc-dev
libclc-dev at lists.llvm.org
Thu Jul 14 13:12:41 PDT 2016
v2: split into 2 patches
use clang builtins for other intrinsics as well
v3: Fix warnings
Switch r600 to use implictarg.ptr
Signed-off-by: Jan Vesely <jan.vesely at rutgers.edu>
---
Hi,
sorry for the delay. This is the last iteration of workitem related changes.
It follows the approach taken in llvm.
local_size, global_size, and num_groups still use the old intrinsic.
clover needs to append the required information before they can be switched
to use implictarg intrinsic.
@Tom I know you revied v2, but the changes looked big enough for repost.
thanks.
Jan
amdgcn/lib/SOURCES | 5 +++--
amdgcn/lib/workitem/get_group_id.cl | 11 +++++++++++
amdgcn/lib/workitem/get_group_id.ll | 29 -----------------------------
amdgcn/lib/workitem/get_local_id.cl | 11 +++++++++++
amdgcn/lib/workitem/get_local_id.ll | 31 -------------------------------
amdgcn/lib/workitem/get_work_dim.cl | 9 +++++++++
amdgpu/lib/SOURCES | 7 +++----
amdgpu/lib/workitem/get_work_dim.ll | 8 --------
r600/lib/SOURCES | 5 +++--
r600/lib/workitem/get_group_id.cl | 11 +++++++++++
r600/lib/workitem/get_group_id.ll | 29 -----------------------------
r600/lib/workitem/get_local_id.cl | 11 +++++++++++
r600/lib/workitem/get_local_id.ll | 31 -------------------------------
r600/lib/workitem/get_work_dim.cl | 9 +++++++++
14 files changed, 71 insertions(+), 136 deletions(-)
create mode 100644 amdgcn/lib/workitem/get_group_id.cl
delete mode 100644 amdgcn/lib/workitem/get_group_id.ll
create mode 100644 amdgcn/lib/workitem/get_local_id.cl
delete mode 100644 amdgcn/lib/workitem/get_local_id.ll
create mode 100644 amdgcn/lib/workitem/get_work_dim.cl
delete mode 100644 amdgpu/lib/workitem/get_work_dim.ll
create mode 100644 r600/lib/workitem/get_group_id.cl
delete mode 100644 r600/lib/workitem/get_group_id.ll
create mode 100644 r600/lib/workitem/get_local_id.cl
delete mode 100644 r600/lib/workitem/get_local_id.ll
create mode 100644 r600/lib/workitem/get_work_dim.cl
diff --git a/amdgcn/lib/SOURCES b/amdgcn/lib/SOURCES
index 49c8dd5..4178d70 100644
--- a/amdgcn/lib/SOURCES
+++ b/amdgcn/lib/SOURCES
@@ -1,3 +1,4 @@
synchronization/barrier_impl.ll
-workitem/get_group_id.ll
-workitem/get_local_id.ll
+workitem/get_group_id.cl
+workitem/get_local_id.cl
+workitem/get_work_dim.cl
diff --git a/amdgcn/lib/workitem/get_group_id.cl b/amdgcn/lib/workitem/get_group_id.cl
new file mode 100644
index 0000000..4b4e7a7
--- /dev/null
+++ b/amdgcn/lib/workitem/get_group_id.cl
@@ -0,0 +1,11 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_group_id(uint dim)
+{
+ switch(dim) {
+ case 0: return __builtin_amdgcn_workgroup_id_x();
+ case 1: return __builtin_amdgcn_workgroup_id_y();
+ case 2: return __builtin_amdgcn_workgroup_id_z();
+ default: return 1;
+ }
+}
diff --git a/amdgcn/lib/workitem/get_group_id.ll b/amdgcn/lib/workitem/get_group_id.ll
deleted file mode 100644
index 9d820e0..0000000
--- a/amdgcn/lib/workitem/get_group_id.ll
+++ /dev/null
@@ -1,29 +0,0 @@
-declare i32 @llvm.amdgcn.workgroup.id.x() #0
-declare i32 @llvm.amdgcn.workgroup.id.y() #0
-declare i32 @llvm.amdgcn.workgroup.id.z() #0
-
-define i32 @get_group_id(i32 %dim) #1 {
- switch i32 %dim, label %default [
- i32 0, label %x_dim
- i32 1, label %y_dim
- i32 2, label %z_dim
- ]
-
-x_dim:
- %x = tail call i32 @llvm.amdgcn.workgroup.id.x()
- ret i32 %x
-
-y_dim:
- %y = tail call i32 @llvm.amdgcn.workgroup.id.y()
- ret i32 %y
-
-z_dim:
- %z = tail call i32 @llvm.amdgcn.workgroup.id.z()
- ret i32 %z
-
-default:
- ret i32 0
-}
-
-attributes #0 = { nounwind readnone }
-attributes #1 = { alwaysinline norecurse nounwind readnone }
diff --git a/amdgcn/lib/workitem/get_local_id.cl b/amdgcn/lib/workitem/get_local_id.cl
new file mode 100644
index 0000000..257c30f
--- /dev/null
+++ b/amdgcn/lib/workitem/get_local_id.cl
@@ -0,0 +1,11 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_local_id(uint dim)
+{
+ switch(dim) {
+ case 0: return __builtin_amdgcn_workitem_id_x();
+ case 1: return __builtin_amdgcn_workitem_id_y();
+ case 2: return __builtin_amdgcn_workitem_id_z();
+ default: return 1;
+ }
+}
diff --git a/amdgcn/lib/workitem/get_local_id.ll b/amdgcn/lib/workitem/get_local_id.ll
deleted file mode 100644
index c54291c..0000000
--- a/amdgcn/lib/workitem/get_local_id.ll
+++ /dev/null
@@ -1,31 +0,0 @@
-declare i32 @llvm.amdgcn.workitem.id.x() #0
-declare i32 @llvm.amdgcn.workitem.id.y() #0
-declare i32 @llvm.amdgcn.workitem.id.z() #0
-
-define i32 @get_local_id(i32 %dim) #1 {
- switch i32 %dim, label %default [
- i32 0, label %x_dim
- i32 1, label %y_dim
- i32 2, label %z_dim
- ]
-
-x_dim:
- %x = tail call i32 @llvm.amdgcn.workitem.id.x(), !range !0
- ret i32 %x
-
-y_dim:
- %y = tail call i32 @llvm.amdgcn.workitem.id.y(), !range !0
- ret i32 %y
-
-z_dim:
- %z = tail call i32 @llvm.amdgcn.workitem.id.z(), !range !0
- ret i32 %z
-
-default:
- ret i32 0
-}
-
-attributes #0 = { nounwind readnone }
-attributes #1 = { alwaysinline norecurse nounwind readnone }
-
-!0 = !{ i32 0, i32 2048 }
diff --git a/amdgcn/lib/workitem/get_work_dim.cl b/amdgcn/lib/workitem/get_work_dim.cl
new file mode 100644
index 0000000..dd2c64f
--- /dev/null
+++ b/amdgcn/lib/workitem/get_work_dim.cl
@@ -0,0 +1,9 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_work_dim()
+{
+ __attribute__((address_space(2))) uint * ptr =
+ (__attribute__((address_space(2))) uint *)
+ __builtin_amdgcn_implicitarg_ptr();
+ return ptr[0];
+}
diff --git a/amdgpu/lib/SOURCES b/amdgpu/lib/SOURCES
index 0f99fe1..1496474 100644
--- a/amdgpu/lib/SOURCES
+++ b/amdgpu/lib/SOURCES
@@ -2,10 +2,6 @@ atomic/atomic.cl
math/ldexp.cl
math/nextafter.cl
math/sqrt.cl
-workitem/get_num_groups.ll
-workitem/get_local_size.ll
-workitem/get_global_size.ll
-workitem/get_work_dim.ll
synchronization/barrier.cl
image/get_image_width.cl
image/get_image_height.cl
@@ -21,3 +17,6 @@ image/write_imagef.cl
image/write_imagei.cl
image/write_imageui.cl
image/write_image_impl.ll
+workitem/get_num_groups.ll
+workitem/get_local_size.ll
+workitem/get_global_size.ll
diff --git a/amdgpu/lib/workitem/get_work_dim.ll b/amdgpu/lib/workitem/get_work_dim.ll
deleted file mode 100644
index 1f86b5e..0000000
--- a/amdgpu/lib/workitem/get_work_dim.ll
+++ /dev/null
@@ -1,8 +0,0 @@
-declare i32 @llvm.AMDGPU.read.workdim() nounwind readnone
-
-define i32 @get_work_dim() nounwind readnone alwaysinline {
- %x = call i32 @llvm.AMDGPU.read.workdim() nounwind readnone , !range !0
- ret i32 %x
-}
-
-!0 = !{ i32 1, i32 4 }
diff --git a/r600/lib/SOURCES b/r600/lib/SOURCES
index 49c8dd5..4178d70 100644
--- a/r600/lib/SOURCES
+++ b/r600/lib/SOURCES
@@ -1,3 +1,4 @@
synchronization/barrier_impl.ll
-workitem/get_group_id.ll
-workitem/get_local_id.ll
+workitem/get_group_id.cl
+workitem/get_local_id.cl
+workitem/get_work_dim.cl
diff --git a/r600/lib/workitem/get_group_id.cl b/r600/lib/workitem/get_group_id.cl
new file mode 100644
index 0000000..e5efc0a
--- /dev/null
+++ b/r600/lib/workitem/get_group_id.cl
@@ -0,0 +1,11 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_group_id(uint dim)
+{
+ switch(dim) {
+ case 0: return __builtin_r600_read_tgid_x();
+ case 1: return __builtin_r600_read_tgid_y();
+ case 2: return __builtin_r600_read_tgid_z();
+ default: return 1;
+ }
+}
diff --git a/r600/lib/workitem/get_group_id.ll b/r600/lib/workitem/get_group_id.ll
deleted file mode 100644
index 837c799..0000000
--- a/r600/lib/workitem/get_group_id.ll
+++ /dev/null
@@ -1,29 +0,0 @@
-declare i32 @llvm.r600.read.tgid.x() #0
-declare i32 @llvm.r600.read.tgid.y() #0
-declare i32 @llvm.r600.read.tgid.z() #0
-
-define i32 @get_group_id(i32 %dim) #1 {
- switch i32 %dim, label %default [
- i32 0, label %x_dim
- i32 1, label %y_dim
- i32 2, label %z_dim
- ]
-
-x_dim:
- %x = tail call i32 @llvm.r600.read.tgid.x()
- ret i32 %x
-
-y_dim:
- %y = tail call i32 @llvm.r600.read.tgid.y()
- ret i32 %y
-
-z_dim:
- %z = tail call i32 @llvm.r600.read.tgid.z()
- ret i32 %z
-
-default:
- ret i32 0
-}
-
-attributes #0 = { nounwind readnone }
-attributes #1 = { alwaysinline norecurse nounwind readnone }
diff --git a/r600/lib/workitem/get_local_id.cl b/r600/lib/workitem/get_local_id.cl
new file mode 100644
index 0000000..a871a5d
--- /dev/null
+++ b/r600/lib/workitem/get_local_id.cl
@@ -0,0 +1,11 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_local_id(uint dim)
+{
+ switch(dim) {
+ case 0: return __builtin_r600_read_tidig_x();
+ case 1: return __builtin_r600_read_tidig_y();
+ case 2: return __builtin_r600_read_tidig_z();
+ default: return 1;
+ }
+}
diff --git a/r600/lib/workitem/get_local_id.ll b/r600/lib/workitem/get_local_id.ll
deleted file mode 100644
index da37ca0..0000000
--- a/r600/lib/workitem/get_local_id.ll
+++ /dev/null
@@ -1,31 +0,0 @@
-declare i32 @llvm.r600.read.tidig.x() #0
-declare i32 @llvm.r600.read.tidig.y() #0
-declare i32 @llvm.r600.read.tidig.z() #0
-
-define i32 @get_local_id(i32 %dim) #1 {
- switch i32 %dim, label %default [
- i32 0, label %x_dim
- i32 1, label %y_dim
- i32 2, label %z_dim
- ]
-
-x_dim:
- %x = tail call i32 @llvm.r600.read.tidig.x(), !range !0
- ret i32 %x
-
-y_dim:
- %y = tail call i32 @llvm.r600.read.tidig.y(), !range !0
- ret i32 %y
-z_dim:
-
- %z = tail call i32 @llvm.r600.read.tidig.z(), !range !0
- ret i32 %z
-
-default:
- ret i32 0
-}
-
-attributes #0 = { nounwind readnone }
-attributes #1 = { alwaysinline norecurse nounwind readnone }
-
-!0 = !{ i32 0, i32 2048 }
diff --git a/r600/lib/workitem/get_work_dim.cl b/r600/lib/workitem/get_work_dim.cl
new file mode 100644
index 0000000..826a655
--- /dev/null
+++ b/r600/lib/workitem/get_work_dim.cl
@@ -0,0 +1,9 @@
+#include <clc/clc.h>
+
+_CLC_DEF uint get_work_dim()
+{
+ __attribute__((address_space(7))) uint * ptr =
+ (__attribute__((address_space(7))) uint *)
+ __builtin_r600_implicitarg_ptr();
+ return ptr[0];
+}
--
2.7.4
More information about the Libclc-dev
mailing list