File: inline-asm-amdgcn.cl

package info (click to toggle)
llvm-toolchain-15 1%3A15.0.6-4
  • links: PTS, VCS
  • area: main
  • in suites: bookworm
  • size: 1,554,644 kB
  • sloc: cpp: 5,922,452; ansic: 1,012,136; asm: 674,362; python: 191,568; objc: 73,855; f90: 42,327; lisp: 31,913; pascal: 11,973; javascript: 10,144; sh: 9,421; perl: 7,447; ml: 5,527; awk: 3,523; makefile: 2,520; xml: 885; cs: 573; fortran: 567
file content (49 lines) | stat: -rw-r--r-- 2,108 bytes parent folder | download | duplicates (14)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
// REQUIRES: amdgpu-registered-target
// RUN: %clang_cc1 -emit-llvm -O0 -o - -triple amdgcn %s | FileCheck %s

typedef float float32 __attribute__((ext_vector_type(32)));

kernel void test_long(int arg0) {
  long v15_16;
  // CHECK: call i64 asm sideeffect "v_lshlrev_b64 v[15:16], 0, $0", "={v[15:16]},v"
  __asm volatile("v_lshlrev_b64 v[15:16], 0, %0" : "={v[15:16]}"(v15_16) : "v"(arg0));
}

kernel void test_agpr() {
  float32 acc_c;
  float reg_a;
  float reg_b;
  float32 reg_c;
  // CHECK:  call <32 x float> asm "v_mfma_f32_32x32x1f32 $0, $1, $2, $3", "=a,v,v,a,~{a0},~{a1},~{a2},~{a3},~{a4},~{a5},~{a6},~{a7},~{a8},~{a9},~{a10},~{a11},~{a12},~{a13},~{a14},~{a15},~{a16},~{a17},~{a18},~{a19},~{a20},~{a21},~{a22},~{a23},~{a24},~{a25},~{a26},~{a27},~{a28},~{a29},~{a30},~{a31}"
  __asm ("v_mfma_f32_32x32x1f32 %0, %1, %2, %3"
         : "=a"(acc_c)
         : "v"(reg_a), "v"(reg_b), "a"(reg_c)
         : "a0", "a1", "a2", "a3", "a4", "a5", "a6", "a7",
           "a8", "a9", "a10", "a11", "a12", "a13", "a14", "a15",
           "a16", "a17", "a18", "a19", "a20", "a21", "a22", "a23",
           "a24", "a25", "a26", "a27", "a28", "a29", "a30", "a31");

   // CHECK: call <32 x float> asm sideeffect "v_mfma_f32_32x32x1f32 a[0:31], $0, $1, a[0:31]", "={a[0:31]},v,v,{a[0:31]}"
  __asm volatile("v_mfma_f32_32x32x1f32 a[0:31], %0, %1, a[0:31]"
                 : "={a[0:31]}"(acc_c)
                 : "v"(reg_a),"v"(reg_b), "{a[0:31]}"(reg_c));

  // CHECK: call float asm "v_accvgpr_read_b32 $0, $1", "={a1},{a1}"
  __asm ("v_accvgpr_read_b32 %0, %1"
         : "={a1}"(reg_a)
         : "{a1}"(reg_b));
}

kernel void test_constraint_DA() {
  const long x = 0x200000001;
  int res;
  // CHECK: call i32 asm sideeffect "v_mov_b32 $0, $1 & 0xFFFFFFFF", "=v,^DA"(i64 8589934593)
  __asm volatile("v_mov_b32 %0, %1 & 0xFFFFFFFF" : "=v"(res) : "DA"(x));
}

kernel void test_constraint_DB() {
  const long x = 0x200000001;
  int res;
  // CHECK: call i32 asm sideeffect "v_mov_b32 $0, $1 & 0xFFFFFFFF", "=v,^DB"(i64 8589934593)
  __asm volatile("v_mov_b32 %0, %1 & 0xFFFFFFFF" : "=v"(res) : "DB"(x));
}