AMDGPU: Define cl_khr_gl_sharing as a supported extension

[arsenm]

Summary

This doesn't seem to be that useful, since it doesn't change any
device side functions. Report it since it seems a macro should be
reported for every extension.

ROCm already reports this. Clover doesn't report it, but since this
isn't useful for device code, it probably isn't harmful to always
define the macro.

[yaxunl]

I am not so sure whether compiler should always report this.

I thought it depends on platform and should let runtime to add it when it is really supported.

But on the other hand, we could report it by default if it is the most case, and let runtime to disable it if it is not supported.

So I will defer this to Brian.

[b-sumner]

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

[arsenm]

In D77910#1976171, @b-sumner wrote:

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

We don't need to worry about theoretical devices. We should know the properties of the driver from -amdhsa, -amdpal, -mesa3d

[arsenm]

Check triple OS

[b-sumner]

In D77910#1981429, @arsenm wrote:

In D77910#1976171, @b-sumner wrote:

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

We don't need to worry about theoretical devices. We should know the properties of the driver from -amdhsa, -amdpal, -mesa3d

It takes more than support in the ISA for some features. The OpenCL driver may not want to support a given optional feature, e.g. images. I'm not opposed to defaults, but if the driver chooses to not support images, it needs to be able to prevent __IMAGE_SUPPORT__ from being defined. Conformance will fail if the runtime and compiler are not consistent.

[arsenm]

In D77910#1981828, @b-sumner wrote:

In D77910#1981429, @arsenm wrote:

In D77910#1976171, @b-sumner wrote:

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

We don't need to worry about theoretical devices. We should know the properties of the driver from -amdhsa, -amdpal, -mesa3d

It takes more than support in the ISA for some features. The OpenCL driver may not want to support a given optional feature, e.g. images. I'm not opposed to defaults, but if the driver chooses to not support images, it needs to be able to prevent __IMAGE_SUPPORT__ from being defined. Conformance will fail if the runtime and compiler are not consistent.

The driver details should be captured by the the triple. If some weird driver decided to do something different, we would need to add a new triple for it. We don't have such a driver, so I don't see why worry about it. It's possible to work around with undef and redef in an implicitly included header. We need to fix properties of the driver based on the target to have perfectly matching offline compilation

[b-sumner]

In D77910#1988807, @arsenm wrote:

In D77910#1981828, @b-sumner wrote:

In D77910#1981429, @arsenm wrote:

In D77910#1976171, @b-sumner wrote:

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

We don't need to worry about theoretical devices. We should know the properties of the driver from -amdhsa, -amdpal, -mesa3d

It takes more than support in the ISA for some features. The OpenCL driver may not want to support a given optional feature, e.g. images. I'm not opposed to defaults, but if the driver chooses to not support images, it needs to be able to prevent __IMAGE_SUPPORT__ from being defined. Conformance will fail if the runtime and compiler are not consistent.

The driver details should be captured by the the triple. If some weird driver decided to do something different, we would need to add a new triple for it. We don't have such a driver, so I don't see why worry about it. It's possible to work around with undef and redef in an implicitly included header. We need to fix properties of the driver based on the target to have perfectly matching offline compilation

I don't see anywhere in the triple talking about driver specific details, unless you would use the environment? That seems like overkill to me. But again, I'm not opposed to defaults, and as long as the driver can override them, this should be OK.

[arsenm]

In D77910#1989163, @b-sumner wrote:

In D77910#1988807, @arsenm wrote:

In D77910#1981828, @b-sumner wrote:

In D77910#1981429, @arsenm wrote:

In D77910#1976171, @b-sumner wrote:

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

We don't need to worry about theoretical devices. We should know the properties of the driver from -amdhsa, -amdpal, -mesa3d

It takes more than support in the ISA for some features. The OpenCL driver may not want to support a given optional feature, e.g. images. I'm not opposed to defaults, but if the driver chooses to not support images, it needs to be able to prevent __IMAGE_SUPPORT__ from being defined. Conformance will fail if the runtime and compiler are not consistent.

The driver details should be captured by the the triple. If some weird driver decided to do something different, we would need to add a new triple for it. We don't have such a driver, so I don't see why worry about it. It's possible to work around with undef and redef in an implicitly included header. We need to fix properties of the driver based on the target to have perfectly matching offline compilation

I don't see anywhere in the triple talking about driver specific details, unless you would use the environment? That seems like overkill to me. But again, I'm not opposed to defaults, and as long as the driver can override them, this should be OK.

The OS is the driver. It doesn't need to specifically encode these details; the OS should encode properties of the driver environment. Anything using -amdhsa should be reporting image support

[b-sumner]

In D77910#2015225, @arsenm wrote:

In D77910#1989163, @b-sumner wrote:

In D77910#1988807, @arsenm wrote:

In D77910#1981828, @b-sumner wrote:

In D77910#1981429, @arsenm wrote:

In D77910#1976171, @b-sumner wrote:

I don't think we can guarantee this is or will be supported on all devices. The language runtime makes this decision.

We don't need to worry about theoretical devices. We should know the properties of the driver from -amdhsa, -amdpal, -mesa3d

It takes more than support in the ISA for some features. The OpenCL driver may not want to support a given optional feature, e.g. images. I'm not opposed to defaults, but if the driver chooses to not support images, it needs to be able to prevent __IMAGE_SUPPORT__ from being defined. Conformance will fail if the runtime and compiler are not consistent.

The driver details should be captured by the the triple. If some weird driver decided to do something different, we would need to add a new triple for it. We don't have such a driver, so I don't see why worry about it. It's possible to work around with undef and redef in an implicitly included header. We need to fix properties of the driver based on the target to have perfectly matching offline compilation

I don't see anywhere in the triple talking about driver specific details, unless you would use the environment? That seems like overkill to me. But again, I'm not opposed to defaults, and as long as the driver can override them, this should be OK.

The OS is the driver. It doesn't need to specifically encode these details; the OS should encode properties of the driver environment. Anything using -amdhsa should be reporting image support

But that's not how things work now or are likely to work in the future. The language runtime is what decides if it is going to report the availability of image support. For offline compilation, I suppose it is OK to assume images are supported, but for online compilation, the language runtime needs to be able to reflect its own decision.

clang/lib/Basic/Targets/AMDGPU.h

void setSupportedOpenCLOpts() override {
     if (IsAMDGCN) {
       Opts.support("cl_khr_fp16");
       Opts.support("cl_khr_int64_base_atomics");
       Opts.support("cl_khr_int64_extended_atomics");
       Opts.support("cl_khr_mipmap_image");
       Opts.support("cl_khr_mipmap_image_writes");
       Opts.support("cl_khr_subgroups");
       Opts.support("cl_khr_3d_image_writes");
+
+      if (getTriple().getOS() == llvm::Triple::AMDHSA ||
+          getTriple().getOS() == llvm::Triple::AMDPAL)
+        Opts.support("cl_khr_gl_sharing");
+
       Opts.support("cl_amd_media_ops");
       Opts.support("cl_amd_media_ops2");
     }
   }
 
   LangAS getOpenCLTypeAddrSpace(OpenCLTypeKind TK) const override {
     switch (TK) {
     case OCLTK_Image:

clang/test/Misc/amdgcn.languageOptsOpenCL.cl

 // REQUIRES: amdgpu-registered-target
 // RUN: %clang_cc1 -x cl -cl-std=CL %s -verify -triple amdgcn-unknown-unknown
 // RUN: %clang_cc1 -x cl -cl-std=CL1.1 %s -verify -triple amdgcn-unknown-unknown
 // RUN: %clang_cc1 -x cl -cl-std=CL1.2 %s -verify -triple amdgcn-unknown-unknown
 // RUN: %clang_cc1 -x cl -cl-std=CL2.0 %s -verify -triple amdgcn-unknown-unknown
 // RUN: %clang_cc1 -x cl -cl-std=CL %s -verify -triple amdgcn-unknown-unknown -Wpedantic-core-features -DTEST_CORE_FEATURES
 // RUN: %clang_cc1 -x cl -cl-std=CL1.1 %s -verify -triple amdgcn-unknown-unknown -Wpedantic-core-features -DTEST_CORE_FEATURES
 // RUN: %clang_cc1 -x cl -cl-std=CL1.2 %s -verify -triple amdgcn-unknown-unknown -Wpedantic-core-features -DTEST_CORE_FEATURES
 // RUN: %clang_cc1 -x cl -cl-std=CL2.0 %s -verify -triple amdgcn-unknown-unknown -Wpedantic-core-features -DTEST_CORE_FEATURES
 
+// Test extensions supported by amdhsa and amdpal
+// RUN: %clang_cc1 -x cl -cl-std=CL1.1 %s -verify -triple amdgcn-unknown-amdhsa -Wpedantic-core-features -DTEST_CORE_FEATURES -DHSAPAL
+// RUN: %clang_cc1 -x cl -cl-std=CL1.2 %s -verify -triple amdgcn-unknown-amdhsa -Wpedantic-core-features -DTEST_CORE_FEATURES -DHSAPAL
+// RUN: %clang_cc1 -x cl -cl-std=CL2.0 %s -verify -triple amdgcn-unknown-amdhsa -Wpedantic-core-features -DTEST_CORE_FEATURES -DHSAPAL
+// RUN: %clang_cc1 -x cl -cl-std=CL1.1 %s -verify -triple amdgcn-unknown-amdpal -Wpedantic-core-features -DTEST_CORE_FEATURES -DHSAPAL
+// RUN: %clang_cc1 -x cl -cl-std=CL1.2 %s -verify -triple amdgcn-unknown-amdpal -Wpedantic-core-features -DTEST_CORE_FEATURES -DHSAPAL
+// RUN: %clang_cc1 -x cl -cl-std=CL2.0 %s -verify -triple amdgcn-unknown-amdpal -Wpedantic-core-features -DTEST_CORE_FEATURES -DHSAPAL
+
+// Test extensions supported by mesa3d/clover
+// RUN: %clang_cc1 -x cl -cl-std=CL1.1 %s -verify -triple amdgcn-unknown-mesa3d -Wpedantic-core-features -DTEST_CORE_FEATURES
+// RUN: %clang_cc1 -x cl -cl-std=CL1.2 %s -verify -triple amdgcn-unknown-mesa3d -Wpedantic-core-features -DTEST_CORE_FEATURES
+// RUN: %clang_cc1 -x cl -cl-std=CL2.0 %s -verify -triple amdgcn-unknown-mesa3d -Wpedantic-core-features -DTEST_CORE_FEATURES
+
+
 // Extensions in all versions
 #ifndef cl_clang_storage_class_specifiers
 #error "Missing cl_clang_storage_class_specifiers define"
 #endif
 #pragma OPENCL EXTENSION cl_clang_storage_class_specifiers: enable
 
 #ifndef cl_khr_fp16
 #error "Missing cl_khr_fp16 define"
 #endif
 #pragma OPENCL EXTENSION cl_khr_fp16: enable
 
 #ifndef cl_khr_int64_base_atomics
 #error "Missing cl_khr_int64_base_atomics define"
 #endif
 #pragma OPENCL EXTENSION cl_khr_int64_base_atomics: enable
 
 #ifndef cl_khr_int64_extended_atomics
 #error "Missing cl_khr_int64_extended_atomics define"
 #endif
 #pragma OPENCL EXTENSION cl_khr_int64_extended_atomics: enable
 
+#ifdef HSAPAL
+#ifndef cl_khr_gl_sharing
+#error "Incorrect cl_khr_gl_sharing define"
+#endif
+#pragma OPENCL EXTENSION cl_khr_gl_sharing: enable
+
+#else
+
 #ifdef cl_khr_gl_sharing
 #error "Incorrect cl_khr_gl_sharing define"
 #endif
 #pragma OPENCL EXTENSION cl_khr_gl_sharing: enable
 // expected-warning@-1{{unsupported OpenCL extension 'cl_khr_gl_sharing' - ignoring}}
+#endif
 
 #ifndef cl_khr_icd
 #error "Missing cl_khr_icd define"
 #endif
 #pragma OPENCL EXTENSION cl_khr_icd: enable
 
 // Core features in CL 1.1