Lines Matching refs:CODE

3 ; RUN: FileCheck -check-prefix=CODE %s
30 ; CODE:        cudaCheckReturn(cudaMemcpy(dev_MemRef_A, MemRef_A, (4096) * (4096) * sizeof(float), …
31 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(dev_MemRef_B, MemRef_B, (4096) * (4096) * sizeof(float), …
32 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(dev_MemRef_D, MemRef_D, (4096) * (4096) * sizeof(float), …
33 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(dev_MemRef_C, MemRef_C, (4096) * (4096) * sizeof(float), …
34 ; CODE-NEXT:   {
35 ; CODE-NEXT:     dim3 k0_dimBlock(16, 32);
36 ; CODE-NEXT:     dim3 k0_dimGrid(128, 128);
37 ; CODE-NEXT:     kernel0 <<<k0_dimGrid, k0_dimBlock>>> (dev_MemRef_tmp, dev_MemRef_A, MemRef_alpha,…
38 ; CODE-NEXT:     cudaCheckKernel();
39 ; CODE-NEXT:   }
41 ; CODE:   {
42 ; CODE-NEXT:     dim3 k1_dimBlock(16, 32);
43 ; CODE-NEXT:     dim3 k1_dimGrid(128, 128);
44 ; CODE-NEXT:     kernel1 <<<k1_dimGrid, k1_dimBlock>>> (dev_MemRef_tmp, dev_MemRef_D, MemRef_beta, …
45 ; CODE-NEXT:     cudaCheckKernel();
46 ; CODE-NEXT:   }
48 ; CODE:   cudaCheckReturn(cudaMemcpy(MemRef_tmp, dev_MemRef_tmp, (4096) * (4096) * sizeof(float), c…
49 ; CODE-NEXT:   cudaCheckReturn(cudaMemcpy(MemRef_D, dev_MemRef_D, (4096) * (4096) * sizeof(float), …
51 ; CODE: # kernel0
52 ; CODE-NEXT: for (int c2 = 0; c2 <= 127; c2 += 1)
53 ; CODE-NEXT:   for (int c4 = 0; c4 <= 1; c4 += 1) {
54 ; CODE-NEXT:     if (c2 == 0)
55 ; CODE-NEXT:       Stmt_for_body6(32 * b0 + t0, 32 * b1 + t1 + 16 * c4);
56 ; CODE-NEXT:     for (int c5 = 0; c5 <= 31; c5 += 1)
57 ; CODE-NEXT:       Stmt_for_body11(32 * b0 + t0, 32 * b1 + t1 + 16 * c4, 32 * c2 + c5);
58 ; CODE-NEXT:   }
60 ; CODE: # kernel1
61 ; CODE-NEXT: for (int c2 = 0; c2 <= 127; c2 += 1)
62 ; CODE-NEXT:   for (int c4 = 0; c4 <= 1; c4 += 1) {
63 ; CODE-NEXT:     if (c2 == 0)
64 ; CODE-NEXT:       Stmt_for_body36(32 * b0 + t0, 32 * b1 + t1 + 16 * c4);
65 ; CODE-NEXT:     for (int c5 = 0; c5 <= 31; c5 += 1)
66 ; CODE-NEXT:       Stmt_for_body44(32 * b0 + t0, 32 * b1 + t1 + 16 * c4, 32 * c2 + c5);
67 ; CODE-NEXT:   }