Skip to content

Commit ed13d19

Browse files
haonanya1haonanya1
authored
[OpenCL] Add SPV_KHR_integer_dot_product support (#302)
* [OpenCL] Add SPV_KHR_integer_dot_product support This is backporting of https://reviews.llvm.org/D106434 Signed-off-by: haonanya <[email protected]> * Add missing tableGen definitions Signed-off-by: haonanya <[email protected]>
1 parent 62547f2 commit ed13d19

File tree

1 file changed

+140
-0
lines changed

1 file changed

+140
-0
lines changed

‎patches/clang/0006-OpenCL-Add-cl_khr_integer_dot_product.patch

Lines changed: 140 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,140 @@
1+
From 9a1886d1b70b4bf452f201ff1268194e21bd686c Mon Sep 17 00:00:00 2001
2+
From: haonanya <[email protected]>
3+
Date: Fri, 24 Dec 2021 10:12:19 +0800
4+
Subject: [PATCH] Add cl_khr_integer_dot_product
5+
6+
This is backporting of https://reviews.llvm.org/D106434
7+
Add the builtins defined by Section 42 "Integer dot product" in the OpenCL
8+
Extension Specification.
9+
10+
See https://www.khronos.org/registry/OpenCL/specs/3.0-unified/html/OpenCL_Ext.html#cl_khr_integer_dot_product
11+
12+
Signed-off-by: haonanya <[email protected]>
13+
---
14+
clang/lib/Headers/opencl-c-base.h | 3 +++
15+
clang/lib/Headers/opencl-c.h | 24 ++++++++++++++++++++++++
16+
clang/lib/Sema/OpenCLBuiltins.td | 25 +++++++++++++++++++++++++
17+
clang/test/Headers/opencl-c-header.cl | 18 ++++++++++++++++++
18+
4 files changed, 70 insertions(+)
19+
20+
diff --git a/clang/lib/Headers/opencl-c-base.h b/clang/lib/Headers/opencl-c-base.h
21+
index 18d367de68ec..f612e85a1797 100644
22+
--- a/clang/lib/Headers/opencl-c-base.h
23+
+++ b/clang/lib/Headers/opencl-c-base.h
24+
@@ -34,6 +34,9 @@
25+
#define __opencl_c_ext_fp32_local_atomic_add 1
26+
#define __opencl_c_ext_fp32_global_atomic_min_max 1
27+
#define __opencl_c_ext_fp32_local_atomic_min_max 1
28+
+#define cl_khr_integer_dot_product 1
29+
+#define __opencl_c_integer_dot_product_input_4x8bit 1
30+
+#define __opencl_c_integer_dot_product_input_4x8bit_packed 1
31+
#endif // (defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200)
32+
33+
// Define features for 2.0 for header backward compatibility
34+
diff --git a/clang/lib/Headers/opencl-c.h b/clang/lib/Headers/opencl-c.h
35+
index 454469991d59..225d9a47abfe 100644
36+
--- a/clang/lib/Headers/opencl-c.h
37+
+++ b/clang/lib/Headers/opencl-c.h
38+
@@ -18428,6 +18428,30 @@ double __ovld sub_group_clustered_reduce_max( double value, uint clustersize );
39+
40+
#endif // cl_khr_subgroup_clustered_reduce
41+
42+
+#if defined(__opencl_c_integer_dot_product_input_4x8bit)
43+
+uint __ovld __cnfn dot(uchar4, uchar4);
44+
+int __ovld __cnfn dot(char4, char4);
45+
+int __ovld __cnfn dot(uchar4, char4);
46+
+int __ovld __cnfn dot(char4, uchar4);
47+
+
48+
+uint __ovld __cnfn dot_acc_sat(uchar4, uchar4, uint);
49+
+int __ovld __cnfn dot_acc_sat(char4, char4, int);
50+
+int __ovld __cnfn dot_acc_sat(uchar4, char4, int);
51+
+int __ovld __cnfn dot_acc_sat(char4, uchar4, int);
52+
+#endif // __opencl_c_integer_dot_product_input_4x8bit
53+
+
54+
+#if defined(__opencl_c_integer_dot_product_input_4x8bit_packed)
55+
+uint __ovld __cnfn dot_4x8packed_uu_uint(uint, uint);
56+
+int __ovld __cnfn dot_4x8packed_ss_int(uint, uint);
57+
+int __ovld __cnfn dot_4x8packed_us_int(uint, uint);
58+
+int __ovld __cnfn dot_4x8packed_su_int(uint, uint);
59+
+
60+
+uint __ovld __cnfn dot_acc_sat_4x8packed_uu_uint(uint, uint, uint);
61+
+int __ovld __cnfn dot_acc_sat_4x8packed_ss_int(uint, uint, int);
62+
+int __ovld __cnfn dot_acc_sat_4x8packed_us_int(uint, uint, int);
63+
+int __ovld __cnfn dot_acc_sat_4x8packed_su_int(uint, uint, int);
64+
+#endif // __opencl_c_integer_dot_product_input_4x8bit_packed
65+
+
66+
#if defined(cl_intel_subgroups)
67+
// Intel-Specific Sub Group Functions
68+
float __ovld __conv intel_sub_group_shuffle( float x, uint c );
69+
diff --git a/clang/lib/Sema/OpenCLBuiltins.td b/clang/lib/Sema/OpenCLBuiltins.td
70+
index d352d35f1e46..d2f8e0a407d4 100644
71+
--- a/clang/lib/Sema/OpenCLBuiltins.td
72+
+++ b/clang/lib/Sema/OpenCLBuiltins.td
73+
@@ -1366,3 +1366,28 @@ let Extension = FuncExtKhrGlMsaaSharing in {
74+
def : Builtin<"get_image_array_size", [Size, ImageType<Image2dArrayMsaaDepth, aQual>], Attr.Const>;
75+
}
76+
}
77+
+
78+
+// Section 42.3 - cl_khr_integer_dot_product
79+
+let Extension = FunctionExtension<"__opencl_c_integer_dot_product_input_4x8bit"> in {
80+
+ def : Builtin<"dot", [UInt, VectorType<UChar, 4>, VectorType<UChar, 4>], Attr.Const>;
81+
+ def : Builtin<"dot", [Int, VectorType<Char, 4>, VectorType<Char, 4>], Attr.Const>;
82+
+ def : Builtin<"dot", [Int, VectorType<UChar, 4>, VectorType<Char, 4>], Attr.Const>;
83+
+ def : Builtin<"dot", [Int, VectorType<Char, 4>, VectorType<UChar, 4>], Attr.Const>;
84+
+
85+
+ def : Builtin<"dot_acc_sat", [UInt, VectorType<UChar, 4>, VectorType<UChar, 4>, UInt], Attr.Const>;
86+
+ def : Builtin<"dot_acc_sat", [Int, VectorType<Char, 4>, VectorType<Char, 4>, Int], Attr.Const>;
87+
+ def : Builtin<"dot_acc_sat", [Int, VectorType<UChar, 4>, VectorType<Char, 4>, Int], Attr.Const>;
88+
+ def : Builtin<"dot_acc_sat", [Int, VectorType<Char, 4>, VectorType<UChar, 4>, Int], Attr.Const>;
89+
+}
90+
+
91+
+let Extension = FunctionExtension<"__opencl_c_integer_dot_product_input_4x8bit_packed"> in {
92+
+ def : Builtin<"dot_4x8packed_uu_uint", [UInt, UInt, UInt], Attr.Const>;
93+
+ def : Builtin<"dot_4x8packed_ss_int", [Int, UInt, UInt], Attr.Const>;
94+
+ def : Builtin<"dot_4x8packed_us_int", [Int, UInt, UInt], Attr.Const>;
95+
+ def : Builtin<"dot_4x8packed_su_int", [Int, UInt, UInt], Attr.Const>;
96+
+
97+
+ def : Builtin<"dot_acc_sat_4x8packed_uu_uint", [UInt, UInt, UInt, UInt], Attr.Const>;
98+
+ def : Builtin<"dot_acc_sat_4x8packed_ss_int", [Int, UInt, UInt, Int], Attr.Const>;
99+
+ def : Builtin<"dot_acc_sat_4x8packed_us_int", [Int, UInt, UInt, Int], Attr.Const>;
100+
+ def : Builtin<"dot_acc_sat_4x8packed_su_int", [Int, UInt, UInt, Int], Attr.Const>;
101+
+}
102+
diff --git a/clang/test/Headers/opencl-c-header.cl b/clang/test/Headers/opencl-c-header.cl
103+
index 7f720cf28142..f2771cd6c6ca 100644
104+
--- a/clang/test/Headers/opencl-c-header.cl
105+
+++ b/clang/test/Headers/opencl-c-header.cl
106+
@@ -146,6 +146,15 @@ global atomic_int z = ATOMIC_VAR_INIT(99);
107+
#if __opencl_c_ext_fp64_local_atomic_min_max != 1
108+
#error "Incorrectly defined __opencl_c_ext_fp64_local_atomic_min_max"
109+
#endif
110+
+#if cl_khr_integer_dot_product != 1
111+
+#error "Incorrectly defined cl_khr_integer_dot_product"
112+
+#endif
113+
+#if __opencl_c_integer_dot_product_input_4x8bit != 1
114+
+#error "Incorrectly defined __opencl_c_integer_dot_product_input_4x8bit"
115+
+#endif
116+
+#if __opencl_c_integer_dot_product_input_4x8bit_packed != 1
117+
+#error "Incorrectly defined __opencl_c_integer_dot_product_input_4x8bit_packed"
118+
+#endif
119+
#else
120+
121+
#ifdef __opencl_c_ext_fp16_global_atomic_load_store
122+
@@ -190,6 +199,15 @@ global atomic_int z = ATOMIC_VAR_INIT(99);
123+
#ifdef __opencl_c_ext_fp64_local_atomic_min_max
124+
#error "Incorrectly __opencl_c_ext_fp64_local_atomic_min_max defined"
125+
#endif
126+
+#ifdef cl_khr_integer_dot_product
127+
+#error "Incorrect cl_khr_integer_dot_product define"
128+
+#endif
129+
+#ifdef __opencl_c_integer_dot_product_input_4x8bit
130+
+#error "Incorrect __opencl_c_integer_dot_product_input_4x8bit define"
131+
+#endif
132+
+#ifdef __opencl_c_integer_dot_product_input_4x8bit_packed
133+
+#error "Incorrect __opencl_c_integer_dot_product_input_4x8bit_packed define"
134+
+#endif
135+
136+
#endif //(defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200)
137+
138+
--
139+
2.17.1
140+

0 commit comments

Comments
 (0)