[OpenCL] Add support for the cl_arm_integer_dot_product extensions

[kpet]

Summary

The specification is available in the Khronos OpenCL registry:

https://www.khronos.org/registry/OpenCL/extensions/arm/cl_arm_integer_dot_product.txt

[Anastasia]

This deserves at least a link to the published spec in the description!

[kpet]

Yes, indeed :).

[Anastasia]

LGTM! Thanks!

[kpet]

Committed as r361641.

clang/include/clang/Basic/OpenCLExtensions.def

 
 // Clang Extensions.
 OPENCLEXT_INTERNAL(cl_clang_storage_class_specifiers, 100, ~0U)
 
 // AMD OpenCL extensions
 OPENCLEXT_INTERNAL(cl_amd_media_ops, 100, ~0U)
 OPENCLEXT_INTERNAL(cl_amd_media_ops2, 100, ~0U)
 
+// ARM OpenCL extensions
+OPENCLEXT_INTERNAL(cl_arm_integer_dot_product_int8, 120, ~0U)
+OPENCLEXT_INTERNAL(cl_arm_integer_dot_product_accumulate_int8, 120, ~0U)
+OPENCLEXT_INTERNAL(cl_arm_integer_dot_product_accumulate_int16, 120, ~0U)
+OPENCLEXT_INTERNAL(cl_arm_integer_dot_product_accumulate_saturate_int8, 120, ~0U)
+
 // Intel OpenCL extensions
 OPENCLEXT_INTERNAL(cl_intel_subgroups, 120, ~0U)
 OPENCLEXT_INTERNAL(cl_intel_subgroups_short, 120, ~0U)
 OPENCLEXT_INTERNAL(cl_intel_device_side_avc_motion_estimation, 120, ~0U)
 
 #undef OPENCLEXT_INTERNAL
 
 #ifdef OPENCLEXT
 #undef OPENCLEXT
 #endif

clang/lib/Headers/opencl-c.h

 uint __ovld amd_sadw(uint src0, uint src1, uint src2);
 uint2 __ovld amd_sadw(uint2 src0, uint2 src1, uint2 src2);
 uint3 __ovld amd_sadw(uint3 src0, uint3 src1, uint3 src2);
 uint4 __ovld amd_sadw(uint4 src0, uint4 src1, uint4 src2);
 uint8 __ovld amd_sadw(uint8 src0, uint8 src1, uint8 src2);
 uint16 __ovld amd_sadw(uint16 src0, uint16 src1, uint16 src2);
 #endif // cl_amd_media_ops2
 
+#if defined(cl_arm_integer_dot_product_int8)
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : begin
+uint __ovld arm_dot(uchar4 a, uchar4 b);
+int __ovld arm_dot(char4 a, char4 b);
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : end
+#endif // defined(cl_arm_integer_dot_product_int8)
+
+#if defined(cl_arm_integer_dot_product_accumulate_int8)
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : begin
+uint __ovld arm_dot_acc(uchar4 a, uchar4 b, uint c);
+int __ovld arm_dot_acc(char4 a, char4 b, int c);
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : end
+#endif // defined(cl_arm_integer_dot_product_accumulate_int8)
+
+#if defined(cl_arm_integer_dot_product_accumulate_int16)
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int16 : begin
+uint __ovld arm_dot_acc(ushort2 a, ushort2 b, uint c);
+int __ovld arm_dot_acc(short2 a, short2 b, int c);
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int16 : end
+#endif // defined(cl_arm_integer_dot_product_accumulate_int16)
+
+#if defined(cl_arm_integer_dot_product_accumulate_saturate_int8)
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_saturate_int8 : begin
+uint __ovld arm_dot_acc_sat(uchar4 a, uchar4 b, uint c);
+int __ovld arm_dot_acc_sat(char4 a, char4 b, int c);
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_saturate_int8 : end
+#endif // defined(cl_arm_integer_dot_product_accumulate_saturate_int8)
+
 // Disable any extensions we may have enabled previously.
 #pragma OPENCL EXTENSION all : disable
 
 #undef __cnfn
 #undef __ovld
 #endif //_OPENCL_H_

clang/test/CodeGenOpenCL/arm-integer-dot-product.cl

+// RUN: %clang_cc1 %s -triple spir-unknown-unknown -finclude-default-header -cl-std=CL1.2 -emit-llvm -o - -O0 | FileCheck %s
+
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable
+void test_int8(uchar4 ua, uchar4 ub, char4 sa, char4 sb) {
+    uint ur = arm_dot(ua, ub);
+    // CHECK: call spir_func i32 @_Z7arm_dotDv4_hS_
+    int sr = arm_dot(sa, sb);
+    // CHECK: call spir_func i32 @_Z7arm_dotDv4_cS_
+}
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : disable
+
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable
+void test_accumulate_int8(uchar4 ua, uchar4 ub, uint uc, char4 sa, char4 sb, int c) {
+    uint ur = arm_dot_acc(ua, ub, uc);
+    // CHECK: call spir_func i32 @_Z11arm_dot_accDv4_hS_j
+    int sr = arm_dot_acc(sa, sb, c);
+    // CHECK: call spir_func i32 @_Z11arm_dot_accDv4_cS_i
+}
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : disable
+
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int16 : enable
+void test_accumulate_int16(ushort2 ua, ushort2 ub, uint uc, short2 sa, short2 sb, int c) {
+    uint ur = arm_dot_acc(ua, ub, uc);
+    // CHECK: call spir_func i32 @_Z11arm_dot_accDv2_tS_j
+    int sr = arm_dot_acc(sa, sb, c);
+    // CHECK: call spir_func i32 @_Z11arm_dot_accDv2_sS_i
+}
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int16 : disable
+
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_saturate_int8 : enable
+void test_accumulate_saturate_int8(uchar4 ua, uchar4 ub, uint uc, char4 sa, char4 sb, int c) {
+    uint ur = arm_dot_acc_sat(ua, ub, uc);
+    // CHECK: call spir_func i32 @_Z15arm_dot_acc_satDv4_hS_j
+    int sr = arm_dot_acc_sat(sa, sb, c);
+    // CHECK: call spir_func i32 @_Z15arm_dot_acc_satDv4_cS_i
+}
+#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_saturate_int8 : disable
+

clang/test/SemaOpenCL/arm-integer-dot-product.cl

+// RUN: %clang_cc1 %s -triple spir-unknown-unknown -finclude-default-header -verify -cl-std=CL1.2 -emit-llvm -o - -O0
+
+void test_negative() {
+    uchar4 ua8, ub8;
+    char4 sa8, sb8;
+    ushort2 ua16, ub16;
+    short2 sa16, sb16;
+    uint ur;
+    int sr;
+    ur = arm_dot(ua8, ub8); // expected-error{{no matching function for call to 'arm_dot'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_int8' to be enabled}}
+    sr = arm_dot(sa8, sb8); // expected-error{{no matching function for call to 'arm_dot'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_int8' to be enabled}}
+    ur = arm_dot_acc(ua8, ub8, ur); // expected-error{{no matching function for call to 'arm_dot_acc'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_accumulate_int8' to be enabled}}
+    sr = arm_dot_acc(sa8, sb8, sr); // expected-error{{no matching function for call to 'arm_dot_acc'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_accumulate_int8' to be enabled}}
+    ur = arm_dot_acc(ua16, ub16, ur); // expected-error{{no matching function for call to 'arm_dot_acc'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_accumulate_int16' to be enabled}}
+    sr = arm_dot_acc(sa16, sb16, sr); // expected-error{{no matching function for call to 'arm_dot_acc'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_accumulate_int16' to be enabled}}
+    ur = arm_dot_acc_sat(ua8, ub8, ur); // expected-error{{no matching function for call to 'arm_dot_acc_sat'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_accumulate_saturate_int8' to be enabled}}
+    sr = arm_dot_acc_sat(sa8, sb8, sr); // expected-error{{no matching function for call to 'arm_dot_acc_sat'}}
+    // expected-note@opencl-c.h:* {{candidate function not viable}}
+    // expected-note@opencl-c.h:* {{candidate unavailable as it requires OpenCL extension 'cl_arm_integer_dot_product_accumulate_saturate_int8' to be enabled}}
+}
+