format abstraction for sparse tensor algebra compilers
play

Format Abstraction for Sparse Tensor Algebra Compilers Stephen Chou , - PowerPoint PPT Presentation

Apr 08, 2023 •218 likes •1.56k views

Format Abstraction for Sparse Tensor Algebra Compilers Stephen Chou , Fredrik Kjolstad, and Saman Amarasinghe Sparse tensors are a natural way of representing real-world data 2 Sparse tensors are a natural way of representing real-world data 2

  1. Computing with di ff erent formats can require very di ff erent code A = B ∘ C Coordinate ✕ Dense array CSR ✕ Dense array CSR ✕ Coordinate for (int pB = B1_pos[0]; for (int i = 0; int pC1 = C1_pos[0]; pB < B1_pos[1]; i < M; while (pC1 < C1_pos[1]) { pB++) { i++) { int i = C1_crd[pC1]; int i = B1_crd[pB]; for (int pB = B2_pos[i]; int C1_segend = pC1 + 1; int j = B2_crd[pB]; pB < B2_pos[i + 1]; while (C1_segend < C1_pos[1] && int pC = i * N + j; pB++) { C1_crd[C1_segend] == i) int pA = i * N + j; int j = B2_crd[pB]; C1_segend++; A[pA] = B[pB] * C[pC]; int pC = i * N + j; int pB2 = B2_pos[i]; } int pA = i * N + j; int pC2 = pC1; A[pA] = B[pB] * C[pC]; while (pB2 < B2_pos[i + 1] && } pC2 < C1_segend) { } int jB2 = B2_crd[pB2]; int jC2 = C2_crd[pC2]; int j = min(jB2, jC2); int pA = i * N + j; if (jB2 == j && jC2 == j) A[pA] = B[pB2] * C[pC2]; if (jB2 == j) pB2++; if (jC2 == j) pC2++; } pC1 = C1_segend; } 10

  2. Hand-coding support for a wide range of formats is infeasible A = B ∘ C Coordinate ✕ Dense array CSR ✕ Dense array CSR ✕ Coordinate 11

  3. Hand-coding support for a wide range of formats is infeasible A = B ∘ C Coordinate ✕ Dense array CSR ✕ Dense array CSR ✕ Coordinate 11

  4. Hand-coding support for a wide range of formats is infeasible A = B ∘ C Coordinate ✕ Dense array CSR ✕ Dense array CSR ✕ Coordinate Dense array ✕ Dense array Coordinate ✕ Coordinate CSR ✕ CSR DIA ✕ DIA DIA ✕ Dense array DIA ✕ Coordinate DIA ✕ CSR ELLPACK ✕ ELLPACK ELLPACK ✕ Dense array ELLPACK ✕ Coordinate ELLPACK ✕ CSR ELLPACK ✕ DIA BCSR ✕ BCSR BCSR ✕ Dense array BCSR ✕ Coordinate 11

  5. Hand-coding support for a wide range of formats is infeasible A = B ∘ C A = B ∘ C ∘ D Coordinate ✕ Dense array Dense array ✕ CSR ✕ CSR Coordinate ✕ CSR ✕ CSR CSR ✕ Dense array CSR ✕ CSR ✕ CSR CSR ✕ Coordinate Dense array ✕ Coordinate ✕ CSR Dense array ✕ Dense array Dense array ✕ Dense array ✕ CSR Coordinate ✕ Coordinate Coordinate ✕ Coordinate ✕ CSR CSR ✕ CSR DIA ✕ Coordinate ✕ Dense array DIA ✕ DIA DIA ✕ Coordinate ✕ CSR DIA ✕ Dense array DIA ✕ Dense array ✕ CSR DIA ✕ Coordinate DIA ✕ CSR ✕ CSR DIA ✕ CSR DIA ✕ Coordinate ✕ Coordinate ELLPACK ✕ ELLPACK DIA ✕ Dense array ✕ Dense array ELLPACK ✕ Dense array DIA ✕ DIA ✕ CSR ELLPACK ✕ Coordinate DIA ✕ DIA ✕ Coordinate ELLPACK ✕ CSR DIA ✕ DIA ✕ Dense array ELLPACK ✕ DIA DIA ✕ DIA ✕ DIA BCSR ✕ BCSR ELLPACK ✕ ELLPACK ✕ DIA BCSR ✕ Dense array ELLPACK ✕ CSR ✕ DIA BCSR ✕ Coordinate 11 ELLPACK ✕ BCSR ✕ DIA

  6. Hand-coding support for a wide range of formats is infeasible A = B ∘ C A = B ∘ C ∘ D y = Ax + z Dense array ✕ Dense array ✕ Dense array Coordinate ✕ Dense array Dense array ✕ CSR ✕ CSR Dense array ✕ Dense array ✕ Sparse vector Coordinate ✕ CSR ✕ CSR CSR ✕ Dense array Dense array ✕ Dense array ✕ Hash map CSR ✕ CSR ✕ CSR CSR ✕ Coordinate Dense array ✕ Sparse vector ✕ Sparse vector Dense array ✕ Coordinate ✕ CSR Dense array ✕ Dense array Dense array ✕ Sparse vector ✕ Hash map Dense array ✕ Dense array ✕ CSR Coordinate ✕ Coordinate Dense array ✕ Hash map ✕ Sparse vector Coordinate ✕ Coordinate ✕ CSR CSR ✕ CSR Dense array ✕ Sparse vector ✕ Dense array DIA ✕ Coordinate ✕ Dense array DIA ✕ DIA Coordinate ✕ Dense array ✕ Dense array DIA ✕ Coordinate ✕ CSR DIA ✕ Dense array Coordinate ✕ Sparse vector ✕ Dense array DIA ✕ Dense array ✕ CSR DIA ✕ Coordinate Coordinate ✕ Dense array ✕ Hash map DIA ✕ CSR ✕ CSR DIA ✕ CSR Coordinate ✕ Sparse vector ✕ Hash map DIA ✕ Coordinate ✕ Coordinate Coordinate ✕ Hash map ✕ Sparse vector ELLPACK ✕ ELLPACK DIA ✕ Dense array ✕ Dense array CSR ✕ Dense array ✕ Dense array ELLPACK ✕ Dense array DIA ✕ DIA ✕ CSR CSR ✕ Dense array ✕ Sparse vector ELLPACK ✕ Coordinate DIA ✕ DIA ✕ Coordinate CSR ✕ Hash map ✕ Sparse vector ELLPACK ✕ CSR DIA ✕ DIA ✕ Dense array CSR ✕ Hash map ✕ Dense array ELLPACK ✕ DIA CSR ✕ Sparse vector ✕ Dense array DIA ✕ DIA ✕ DIA BCSR ✕ BCSR DIA ✕ Dense array ✕ Dense array ELLPACK ✕ ELLPACK ✕ DIA BCSR ✕ Dense array DIA ✕ Hash map ✕ Dense array ELLPACK ✕ CSR ✕ DIA BCSR ✕ Coordinate ELLPACK ✕ Dense array ✕ Sparse vector 11 ELLPACK ✕ BCSR ✕ DIA

  7. Format Abstraction & Code Generation DIA x supports Dense no yes locate? Range y supports y supports Mode-generic tensor yes yes Offset locate? locate? Compressed no Singleton x unordered yes no and y ordered? Dense Dense no co-iterate over x iterate over x and iterate over y and and y locate into y locate into x Evaluation 12

  8. Format Abstraction & Code Generation DIA x supports Dense no yes locate? Range y supports y supports Mode-generic tensor yes yes Offset locate? locate? Compressed no Singleton x unordered yes no and y ordered? Dense Dense no co-iterate over x iterate over x and iterate over y and and y locate into y locate into x Evaluation 12

  9. Format Abstraction & Code Generation DIA x supports Dense no yes locate? Range y supports y supports Mode-generic tensor yes yes Offset locate? locate? Compressed no Singleton x unordered yes no and y ordered? Dense Dense no co-iterate over x iterate over x and iterate over y and and y locate into y locate into x Evaluation 12

  10. Tensor formats can be viewed as compositions of level formats A 0 B C D 1 E F G 0 2 H J 1 0 1 2 3 13

  11. Tensor formats can be viewed as compositions of level formats 3 A 0 B C 0 3 5 9 D 1 0 1 1 0 1 0 0 1 1 E F G 0 0 0 2 3 1 1 3 0 3 2 H J 1 0 1 2 3 A B C D E F G H J 0 1 2 3 4 5 6 7 8 13

  12. Tensor formats can be viewed as compositions of level formats 3 A 0 B C 0 3 5 9 D 0 1 1 0 1 0 0 1 1 1 E F G 0 0 0 2 3 1 1 3 0 3 2 H J 1 0 1 2 3 A B C D E F G H J 0 1 2 3 4 5 6 7 8 13

  13. Tensor formats can be viewed as compositions of level formats 3 Slices A 0 B C 0 3 5 9 D 0 1 1 0 1 0 0 1 1 1 E F G 0 0 0 2 3 1 1 3 0 3 2 H J 1 0 1 2 3 A B C D E F G H J 0 1 2 3 4 5 6 7 8 13

  14. Tensor formats can be viewed as compositions of level formats 3 A 0 B C 0 3 5 9 D Rows 0 1 1 0 1 0 0 1 1 1 E F G 0 0 0 2 3 1 1 3 0 3 2 H J 1 0 1 2 3 A B C D E F G H J 0 1 2 3 4 5 6 7 8 13

  15. Tensor formats can be viewed as compositions of level formats 3 A 0 B C 0 3 5 9 D 0 1 1 0 1 0 0 1 1 1 E F G 0 0 0 2 3 1 1 3 0 3 Columns 2 H J 1 0 1 2 3 A B C D E F G H J 0 1 2 3 4 5 6 7 8 13

  16. Tensor formats can be viewed as compositions of level formats 3 Dense A 0 B C 0 3 5 9 D Compressed 0 1 1 0 1 0 0 1 1 1 E F G 0 0 0 2 3 1 1 3 0 3 Singleton 2 H J 1 0 1 2 3 A B C D E F G H J 0 1 2 3 4 5 6 7 8 13

  17. The same level formats can be composed in many ways Dense Compressed Singleton 14

  18. The same level formats can be composed in many ways Dense Compressed Singleton 0 1 2 3 A B 0 C D E 1 F 2 14

  19. The same level formats can be composed in many ways Dense Compressed Singleton 3 0 1 2 3 Dense A B 0 C D E 1 F 2 14

  20. The same level formats can be composed in many ways Dense Compressed Singleton 3 0 1 2 3 Dense A B 0 0 2 5 6 C D E 1 Compressed 0 2 1 2 3 3 F 2 A B C D E F 14

  21. The same level formats can be composed in many ways Dense Compressed Singleton CSR { 3 0 1 2 3 A B 0 0 2 5 6 C D E 1 0 2 1 2 3 3 F 2 A B C D E F 14

  22. The same level formats can be composed in many ways Dense Compressed Singleton 0 1 2 3 A B 0 C D E 1 F 2 15

  23. The same level formats can be composed in many ways Dense Compressed Singleton 0 6 0 1 2 3 Compressed A B 0 0 1 1 1 2 0 C D E 1 F 2 15

  24. The same level formats can be composed in many ways Dense Compressed Singleton 0 6 0 1 2 3 Compressed A B 0 0 1 1 1 2 0 C D E 1 0 2 1 2 3 3 Singleton F 2 A B C D E F 15

  25. The same level formats can be composed in many ways Dense Compressed Singleton Coordinate { 0 6 0 1 2 3 A B 0 0 1 1 1 2 0 C D E 1 0 2 1 2 3 3 F 2 A B C D E F 15

  26. The same level formats can be composed in many ways Dense Compressed Singleton 16

  27. The same level formats can be composed in many ways Dense Compressed Singleton Level formats Coordinate matrix CSR Compressed Dense Singleton Compressed [Tinney and Walker, 1967] Tensor formats 16

  28. The same level formats can be composed in many ways Dense Compressed Singleton Level formats Coordinate matrix CSR Dense array tensor Coordinate tensor Compressed Dense Dense Compressed Singleton Compressed Dense Singleton Dense Singleton [Tinney and Walker, 1967] Mode-generic tensor Compressed Singleton Tensor Dense formats Dense [Baskaran et al. 2012] 16

  29. The same level formats can be composed in many ways Dense Compressed Singleton Level formats Coordinate matrix CSR Dense array tensor Coordinate tensor Compressed Dense Dense Compressed Singleton Compressed Dense Singleton Dense Singleton [Tinney and Walker, 1967] CSB Mode-generic tensor BCSR ELLPACK Dense Compressed Dense Dense Singleton Compressed Dense Dense Tensor Dense Dense Compressed Singleton formats Dense Dense Singleton [Kincaid et al. 1989] [Baskaran et al. 2012] [Im and Yelick 1998] [Buluç et al. 2009] 16

  30. The same level formats can be composed in many ways Dense Compressed Singleton Level formats Hashed Range Offset Coordinate matrix CSR Dense array tensor Coordinate tensor Compressed Dense Dense Compressed Singleton Compressed Dense Singleton Dense Singleton [Tinney and Walker, 1967] CSB Mode-generic tensor BCSR ELLPACK Dense Compressed Dense Dense Singleton Compressed Dense Dense Tensor Dense Dense Compressed Singleton formats Dense Dense Singleton [Kincaid et al. 1989] [Baskaran et al. 2012] [Im and Yelick 1998] [Buluç et al. 2009] 16

  31. The same level formats can be composed in many ways Dense Compressed Singleton Level formats Hashed Range Offset Coordinate matrix CSR Dense array tensor Coordinate tensor Compressed Dense Dense Compressed Singleton Compressed Dense Singleton Dense Singleton [Tinney and Walker, 1967] CSB Mode-generic tensor BCSR ELLPACK Dense Compressed Dense Dense Singleton Compressed Dense Dense Tensor Dense Dense Compressed Singleton formats Dense Dense Singleton [Kincaid et al. 1989] [Baskaran et al. 2012] [Im and Yelick 1998] [Buluç et al. 2009] Block DIA Hash map vector Hash map matrix DIA Dense Hashed Hashed Dense Range Hashed Range Offset [Patwary et al. 2015] Offset Dense Dense [Saad 2003] 16

  32. <latexit sha1_base64="4q1Sk4lxncbqFoMX5S1GLw3t+GQ=">ACIXicbVDLSgMxFM34rPVdekmWIQKIjMiWFyVunFZwT6gLSWTuW2DmcyQ3BHL0F9x46+4caFId+LPmD4W2vZA4OSce29yjx9LYdB1v52V1bX1jc3MVnZ7Z3dvP3dwWDNRojlUeSQj3fCZASkUVFGghEasgYW+hLr/eDv260+gjYjUAw5iaIesp0RXcIZW6uSKZXpDCy2EZ0wDUAaG53R641FoJxkDwTLprJPLuxfuBHSReDOSJzNUOrlRK4h4EoJCLpkxTc+NsZ0yjYJLGZbiYGY8UfWg6alioVg2ulkwyE9tUpAu5G2RyGdqH87UhYaMwh9Wxky7Jt5bywu85oJdovtVKg4QVB8+lA3kRQjOo6LBkIDRzmwhHEt7F8p7zPNONpQszYEb37lRVK7vPAsv7/Kl8qzODLkmJyQAvHINSmRO1IhVcLJC3kjH+TeXenS9nNC1dcWY9R+QfnJ9fTfykRA=</latexit> <latexit sha1_base64="4q1Sk4lxncbqFoMX5S1GLw3t+GQ=">ACIXicbVDLSgMxFM34rPVdekmWIQKIjMiWFyVunFZwT6gLSWTuW2DmcyQ3BHL0F9x46+4caFId+LPmD4W2vZA4OSce29yjx9LYdB1v52V1bX1jc3MVnZ7Z3dvP3dwWDNRojlUeSQj3fCZASkUVFGghEasgYW+hLr/eDv260+gjYjUAw5iaIesp0RXcIZW6uSKZXpDCy2EZ0wDUAaG53R641FoJxkDwTLprJPLuxfuBHSReDOSJzNUOrlRK4h4EoJCLpkxTc+NsZ0yjYJLGZbiYGY8UfWg6alioVg2ulkwyE9tUpAu5G2RyGdqH87UhYaMwh9Wxky7Jt5bywu85oJdovtVKg4QVB8+lA3kRQjOo6LBkIDRzmwhHEt7F8p7zPNONpQszYEb37lRVK7vPAsv7/Kl8qzODLkmJyQAvHINSmRO1IhVcLJC3kjH+TeXenS9nNC1dcWY9R+QfnJ9fTfykRA=</latexit> <latexit sha1_base64="4q1Sk4lxncbqFoMX5S1GLw3t+GQ=">ACIXicbVDLSgMxFM34rPVdekmWIQKIjMiWFyVunFZwT6gLSWTuW2DmcyQ3BHL0F9x46+4caFId+LPmD4W2vZA4OSce29yjx9LYdB1v52V1bX1jc3MVnZ7Z3dvP3dwWDNRojlUeSQj3fCZASkUVFGghEasgYW+hLr/eDv260+gjYjUAw5iaIesp0RXcIZW6uSKZXpDCy2EZ0wDUAaG53R641FoJxkDwTLprJPLuxfuBHSReDOSJzNUOrlRK4h4EoJCLpkxTc+NsZ0yjYJLGZbiYGY8UfWg6alioVg2ulkwyE9tUpAu5G2RyGdqH87UhYaMwh9Wxky7Jt5bywu85oJdovtVKg4QVB8+lA3kRQjOo6LBkIDRzmwhHEt7F8p7zPNONpQszYEb37lRVK7vPAsv7/Kl8qzODLkmJyQAvHINSmRO1IhVcLJC3kjH+TeXenS9nNC1dcWY9R+QfnJ9fTfykRA=</latexit> <latexit sha1_base64="4q1Sk4lxncbqFoMX5S1GLw3t+GQ=">ACIXicbVDLSgMxFM34rPVdekmWIQKIjMiWFyVunFZwT6gLSWTuW2DmcyQ3BHL0F9x46+4caFId+LPmD4W2vZA4OSce29yjx9LYdB1v52V1bX1jc3MVnZ7Z3dvP3dwWDNRojlUeSQj3fCZASkUVFGghEasgYW+hLr/eDv260+gjYjUAw5iaIesp0RXcIZW6uSKZXpDCy2EZ0wDUAaG53R641FoJxkDwTLprJPLuxfuBHSReDOSJzNUOrlRK4h4EoJCLpkxTc+NsZ0yjYJLGZbiYGY8UfWg6alioVg2ulkwyE9tUpAu5G2RyGdqH87UhYaMwh9Wxky7Jt5bywu85oJdovtVKg4QVB8+lA3kRQjOo6LBkIDRzmwhHEt7F8p7zPNONpQszYEb37lRVK7vPAsv7/Kl8qzODLkmJyQAvHINSmRO1IhVcLJC3kjH+TeXenS9nNC1dcWY9R+QfnJ9fTfykRA=</latexit> <latexit sha1_base64="znx8HB3iu6URMn3Pu43WiyqTY=">ACAHicbVC7SgNBFJ31GeNr1cLCZjAIsQm7IihWQRvLCOYBSQizk7vJkNmdZeauGJY0/oqNhSK2foadf+PkUWjigYHDOfdw54gkcKg5307S8srq2vruY385tb2zq67t18zKtUcqlxJpRsBMyBFDFUKGRaGBRIKEeDG7Gfv0BtBEqvsdhAu2I9WIRCs7QSh3kNMrWmwhPGLGVWSzxkB3dNpxC17Jm4AuEn9GCmSGSsf9anUVTyOIkUtmTNP3EmxnTKPgEkb5VmogYXzAetC0NGYRmHY2OWBET6zSpaHS9sVIJ+rvRMYiY4ZRYCcjhn0z743F/7xmiuFlOxNxkiLEfLoTCVFRcdt0K7QwFEOLWFcC/tXyvtM462s7wtwZ8/eZHUzkq+5XfnhfL1rI4cOSLHpEh8ckHK5JZUSJVwMiLP5JW8OU/Oi/PufExHl5xZ5oD8gfP5Awmdlg0=</latexit> <latexit sha1_base64="znx8HB3iu6URMn3Pu43WiyqTY=">ACAHicbVC7SgNBFJ31GeNr1cLCZjAIsQm7IihWQRvLCOYBSQizk7vJkNmdZeauGJY0/oqNhSK2foadf+PkUWjigYHDOfdw54gkcKg5307S8srq2vruY385tb2zq67t18zKtUcqlxJpRsBMyBFDFUKGRaGBRIKEeDG7Gfv0BtBEqvsdhAu2I9WIRCs7QSh3kNMrWmwhPGLGVWSzxkB3dNpxC17Jm4AuEn9GCmSGSsf9anUVTyOIkUtmTNP3EmxnTKPgEkb5VmogYXzAetC0NGYRmHY2OWBET6zSpaHS9sVIJ+rvRMYiY4ZRYCcjhn0z743F/7xmiuFlOxNxkiLEfLoTCVFRcdt0K7QwFEOLWFcC/tXyvtM462s7wtwZ8/eZHUzkq+5XfnhfL1rI4cOSLHpEh8ckHK5JZUSJVwMiLP5JW8OU/Oi/PufExHl5xZ5oD8gfP5Awmdlg0=</latexit> <latexit sha1_base64="znx8HB3iu6URMn3Pu43WiyqTY=">ACAHicbVC7SgNBFJ31GeNr1cLCZjAIsQm7IihWQRvLCOYBSQizk7vJkNmdZeauGJY0/oqNhSK2foadf+PkUWjigYHDOfdw54gkcKg5307S8srq2vruY385tb2zq67t18zKtUcqlxJpRsBMyBFDFUKGRaGBRIKEeDG7Gfv0BtBEqvsdhAu2I9WIRCs7QSh3kNMrWmwhPGLGVWSzxkB3dNpxC17Jm4AuEn9GCmSGSsf9anUVTyOIkUtmTNP3EmxnTKPgEkb5VmogYXzAetC0NGYRmHY2OWBET6zSpaHS9sVIJ+rvRMYiY4ZRYCcjhn0z743F/7xmiuFlOxNxkiLEfLoTCVFRcdt0K7QwFEOLWFcC/tXyvtM462s7wtwZ8/eZHUzkq+5XfnhfL1rI4cOSLHpEh8ckHK5JZUSJVwMiLP5JW8OU/Oi/PufExHl5xZ5oD8gfP5Awmdlg0=</latexit> <latexit sha1_base64="znx8HB3iu6URMn3Pu43WiyqTY=">ACAHicbVC7SgNBFJ31GeNr1cLCZjAIsQm7IihWQRvLCOYBSQizk7vJkNmdZeauGJY0/oqNhSK2foadf+PkUWjigYHDOfdw54gkcKg5307S8srq2vruY385tb2zq67t18zKtUcqlxJpRsBMyBFDFUKGRaGBRIKEeDG7Gfv0BtBEqvsdhAu2I9WIRCs7QSh3kNMrWmwhPGLGVWSzxkB3dNpxC17Jm4AuEn9GCmSGSsf9anUVTyOIkUtmTNP3EmxnTKPgEkb5VmogYXzAetC0NGYRmHY2OWBET6zSpaHS9sVIJ+rvRMYiY4ZRYCcjhn0z743F/7xmiuFlOxNxkiLEfLoTCVFRcdt0K7QwFEOLWFcC/tXyvtM462s7wtwZ8/eZHUzkq+5XfnhfL1rI4cOSLHpEh8ckHK5JZUSJVwMiLP5JW8OU/Oi/PufExHl5xZ5oD8gfP5Awmdlg0=</latexit> X A ij = B ijk c k for (int i = 0; i < m; i++) { k for (int pB2 = B2_pos[pB1]; pB2 < B2_pos[pB1 + 1]; pB2++) { int j = B2_idx[pB2]; int pA2 = (i * n) + j; int pB3 = B3_pos[pB2]; int pc1 = c1_pos[0]; Tensor Algebra Compiler while (pB3 < B3_pos[pB2 + 1] && pc1 < c1_pos[1]) { int kB = B3_idx[pB3]; (taco) int kc = c1_idx[pc1]; int k = min(kB, kc); if (kB == k && kc == k) { a[pA2] += b[pB3] * c[pc1]; [Kjolstad et al. 2017] } if (kB == k) pB3++; if (kc == k) pc1++; } } } A : (dense , dense) B : (dense , compressed , compressed) c : (compressed) 17

  33. <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="ck8pdC+ekZH4nUmSP+ZG7r8lEyk=">AB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i7+gsTLXTzQrMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GS7KDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RtRxzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D7odOn4MoA7ncAFXEMIN3MEDdKALAhJ4hXdv4r15H6uat6tDP4I+/zBzjGijg=</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">ACf3icbZFda9swFIZl76Ndlm1pb3sjWga7aINkN7FNGXTbTS87aNpCbIysKkW+QNJLgShH7O/tLv+m8pClu6A4KH9+jVOTqnaARXGqEHz3/1+s3bnd13vf9Dx8/Dfb616puJWUTWota3hZEMcErNtFcC3bSEbKQrCbYvmjy9/cM6l4XV3pVcOykiwqPueUaCflg98mXT8ylYsiM2iI4hEKT4/RcBwmCMcOEI7COLDfcsN/WQu/wlS1ZW6W1mxZ8Sg8TQLnCKMxDpGDOE5QgO3zrp03pTOag23bBEaB3HSFVqHgwiHUTyhubOZPB0XMOvgS8gSOwict8Ced1bQtWaWpIEpNMWp0ZojUnApme2mrWEPokizY1GFSqYys+7Jws9OmcF5Ld2pNFyrfzsMKZValYW7WRJ9p7Zznfi/3LTV8zgzvGpazSr6VGjeCqhr2O0FzrhkVIuVA0Ild71Cekckodptr+eGgLe/BKugyF2/BOBXADsEXgEzsEFuAQTQL0d78Qbe5Hf9wM/eRqX723mtg/+Cf/sEWMFufg=</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">ACf3icbZFda9swFIZl76Ndlm1pb3sjWga7aINkN7FNGXTbTS87aNpCbIysKkW+QNJLgShH7O/tLv+m8pClu6A4KH9+jVOTqnaARXGqEHz3/1+s3bnd13vf9Dx8/Dfb616puJWUTWota3hZEMcErNtFcC3bSEbKQrCbYvmjy9/cM6l4XV3pVcOykiwqPueUaCflg98mXT8ylYsiM2iI4hEKT4/RcBwmCMcOEI7COLDfcsN/WQu/wlS1ZW6W1mxZ8Sg8TQLnCKMxDpGDOE5QgO3zrp03pTOag23bBEaB3HSFVqHgwiHUTyhubOZPB0XMOvgS8gSOwict8Ced1bQtWaWpIEpNMWp0ZojUnApme2mrWEPokizY1GFSqYys+7Jws9OmcF5Ld2pNFyrfzsMKZValYW7WRJ9p7Zznfi/3LTV8zgzvGpazSr6VGjeCqhr2O0FzrhkVIuVA0Ild71Cekckodptr+eGgLe/BKugyF2/BOBXADsEXgEzsEFuAQTQL0d78Qbe5Hf9wM/eRqX723mtg/+Cf/sEWMFufg=</latexit> <latexit sha1_base64="9QtCk7M63G3IchN+Z5Vs/Le7raY=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> taco generates code dimension by dimension X A ij = B ijk · c k k for (int i = 0; i < m; i++) { for (int pB2 = B2_pos[pB1]; pB2 < B2_pos[pB1 + 1]; pB2++) { int j = B2_idx[pB2]; int pA2 = (i * n) + j; int pB3 = B3_pos[pB2]; int pc1 = c1_pos[0]; while (pB3 < B3_pos[pB2 + 1] && pc1 < c1_pos[1]) { int kB = B3_crd[pB3]; int kc = c1_crd[pc1]; int k = min(kB, kc); if (kB == k && kc == k) { A[pA2] += B[pB3] * c[pc1]; } if (kB == k) pB3++; if (kc == k) pc1++; } } } 18

  34. <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="ck8pdC+ekZH4nUmSP+ZG7r8lEyk=">AB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i7+gsTLXTzQrMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GS7KDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RtRxzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D7odOn4MoA7ncAFXEMIN3MEDdKALAhJ4hXdv4r15H6uat6tDP4I+/zBzjGijg=</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">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</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">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</latexit> <latexit sha1_base64="9QtCk7M63G3IchN+Z5Vs/Le7raY=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> taco generates code dimension by dimension X A ij = B ijk · c k k for (int i = 0; i < m; i++) { for (int pB2 = B2_pos[pB1]; pB2 < B2_pos[pB1 + 1]; pB2++) { int j = B2_idx[pB2]; int pA2 = (i * n) + j; int pB3 = B3_pos[pB2]; int pc1 = c1_pos[0]; while (pB3 < B3_pos[pB2 + 1] && pc1 < c1_pos[1]) { int kB = B3_crd[pB3]; int kc = c1_crd[pc1]; int k = min(kB, kc); if (kB == k && kc == k) { A[pA2] += B[pB3] * c[pc1]; } if (kB == k) pB3++; if (kc == k) pc1++; } } } 18

  35. <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="ck8pdC+ekZH4nUmSP+ZG7r8lEyk=">AB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i7+gsTLXTzQrMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GS7KDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RtRxzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D7odOn4MoA7ncAFXEMIN3MEDdKALAhJ4hXdv4r15H6uat6tDP4I+/zBzjGijg=</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">ACf3icbZFda9swFIZl76Ndlm1pb3sjWga7aINkN7FNGXTbTS87aNpCbIysKkW+QNJLgShH7O/tLv+m8pClu6A4KH9+jVOTqnaARXGqEHz3/1+s3bnd13vf9Dx8/Dfb616puJWUTWota3hZEMcErNtFcC3bSEbKQrCbYvmjy9/cM6l4XV3pVcOykiwqPueUaCflg98mXT8ylYsiM2iI4hEKT4/RcBwmCMcOEI7COLDfcsN/WQu/wlS1ZW6W1mxZ8Sg8TQLnCKMxDpGDOE5QgO3zrp03pTOag23bBEaB3HSFVqHgwiHUTyhubOZPB0XMOvgS8gSOwict8Ced1bQtWaWpIEpNMWp0ZojUnApme2mrWEPokizY1GFSqYys+7Jws9OmcF5Ld2pNFyrfzsMKZValYW7WRJ9p7Zznfi/3LTV8zgzvGpazSr6VGjeCqhr2O0FzrhkVIuVA0Ild71Cekckodptr+eGgLe/BKugyF2/BOBXADsEXgEzsEFuAQTQL0d78Qbe5Hf9wM/eRqX723mtg/+Cf/sEWMFufg=</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">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</latexit> <latexit sha1_base64="9QtCk7M63G3IchN+Z5Vs/Le7raY=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> taco generates code dimension by dimension X A ij = B ijk · c k k for (int i = 0; i < m; i++) { for (int pB2 = B2_pos[pB1]; pB2 < B2_pos[pB1 + 1]; pB2++) { int j = B2_idx[pB2]; int pA2 = (i * n) + j; int pB3 = B3_pos[pB2]; int pc1 = c1_pos[0]; while (pB3 < B3_pos[pB2 + 1] && pc1 < c1_pos[1]) { int kB = B3_crd[pB3]; int kc = c1_crd[pc1]; int k = min(kB, kc); if (kB == k && kc == k) { A[pA2] += B[pB3] * c[pc1]; } if (kB == k) pB3++; if (kc == k) pc1++; } } } 18

  36. <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="ck8pdC+ekZH4nUmSP+ZG7r8lEyk=">AB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i7+gsTLXTzQrMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GS7KDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RtRxzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D7odOn4MoA7ncAFXEMIN3MEDdKALAhJ4hXdv4r15H6uat6tDP4I+/zBzjGijg=</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">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</latexit> <latexit sha1_base64="n5e1ptEuTL43cmV9t1ImHPSoNvM=">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</latexit> <latexit sha1_base64="9QtCk7M63G3IchN+Z5Vs/Le7raY=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">ACinicbZHfbtMwFMadMNgoAwq75MZahcQFVHbSNonQpLHtYpdDotukJoc1+1MnT+yHaTK8sPwStzxNjhdkVjHkSz9J3z+Rz7FI3gSiP02/Of7D19tn/wvPfi8OWr1/03b69V3UrKprQWtbwtiGKCV2yquRbstpGMlIVgN8XqvMvf/GBS8br6ptcNy0qyrPiCU6KdlPd/mnRzyUwui8ygIYrHKBx9RMNJmCAcO0A4CuPAfskN/24tPIGpasvcrKzZseJxOEoC5wijCQ6RgzhOUIDtWdOW9K57WGO7YITYI46RptwkGEwygeW0NzZ7J5f/A3Bx8D3sIAbOMq7/9K5zVtS1ZpKohSM4wanRkiNaeC2V7aKtYQuiJLNnNYkZKpzGxmsvC9U+ZwUt3Kg036r8OQ0ql1mXhKkui79RurhP/l5u1ehFnhldNq1lF7xstWgF1Dbu9wDmXjGqxdkCo5G5WSO+IJFS7fXcJ+DdJz+G62CIHX8dDU7Pt9xAN6BY/ABYBCBU3AJrsAUG/f+RNvMg/9AM/8T/fl/re1nMEHoR/8QeH4Lru</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> <latexit sha1_base64="N+Gz0mErqtYlWzSRJfkrAyiSckE=">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</latexit> taco generates code dimension by dimension X A ij = B ijk · c k k for (int i = 0; i < m; i++) { for (int pB2 = B2_pos[pB1]; pB2 < B2_pos[pB1 + 1]; pB2++) { int j = B2_idx[pB2]; int pA2 = (i * n) + j; int pB3 = B3_pos[pB2]; int pc1 = c1_pos[0]; while (pB3 < B3_pos[pB2 + 1] && pc1 < c1_pos[1]) { int kB = B3_crd[pB3]; int kc = c1_crd[pc1]; int k = min(kB, kc); if (kB == k && kc == k) { A[pA2] += B[pB3] * c[pc1]; } if (kB == k) pB3++; if (kc == k) pc1++; } } } 18

  37. Hand-coding support for a wide range of level formats is also infeasible Compressed Dense × + Dense Compressed 19

  38. Hand-coding support for a wide range of level formats is also infeasible Compressed Dense × + Dense Compressed Compressed Hashed × + Compressed Singleton + Dense Range + Compressed Offset Dense Compressed Hashed × × Singleton Compressed Hashed × × + Singleton Compressed Dense × + + Singleton Compressed Dense . . . 19

  39. Code generation is performed in two stages 20

  40. Code generation is performed in two stages Compressed Hashed × 20

  41. Code generation is performed in two stages Compressed Hashed × High-level algorithm 20

  42. Code generation is performed in two stages Compressed Hashed × High-level algorithm Runnable code 20

  43. Code generation is performed in two stages Compressed Hashed × High-level algorithm How to compute with di ff erent data structures Runnable code 20

  44. Code generation is performed in two stages Compressed Hashed × How to compute with multiple operands High-level algorithm How to compute with di ff erent data structures Runnable code 20

  45. Tensor algebra computations can be expressed in terms of high-level operations on tensor operands B C A B K L C D E F M N O P ∘ G H Q R S I J T U 21

  46. Tensor algebra computations can be expressed in terms of high-level operations on tensor operands B C A B K L C D E F M N O P ∘ G H Q R S I J T U 21

  47. Tensor algebra computations can be expressed in terms of high-level operations on tensor operands B C A B K L C D E F M N O P ∘ G H Q R S I J T U 21

  48. Tensor algebra computations can be expressed in terms of high-level operations on tensor operands B C A B K L C D E F M N O P ∘ G H Q R S I J T U 21

  49. Level formats declare whether they support various high-level operations Dense Range Compressed Singleton Offset Hashed 22

  50. Level formats declare whether they support various high-level operations Random access Iteration Dense Range Compressed Singleton Offset Hashed 22

  51. Level formats declare whether they support various high-level operations Random access Iteration Dense Range Compressed Singleton Offset Hashed 22

  52. Level formats declare whether they support various high-level operations Random access Iteration Dense Range Compressed Singleton Offset Hashed 22

  53. Level formats declare whether they support various high-level operations Random access Iteration Dense Range Compressed Singleton Offset Hashed 22

  54. Compiler constructs e ffi cient algorithm by reasoning about whether operands support required high-level operations Random access Dense Range Compressed Singleton Offset Hashed 23

  55. Compiler constructs e ffi cient algorithm by reasoning about whether operands support required high-level operations Random access B C ∘ Dense Range Compressed Singleton Offset Hashed 23

  56. Compiler constructs e ffi cient algorithm by reasoning about whether operands support required high-level operations Random access B C ∘ Dense Compressed Hashed × Range Compressed Singleton Offset Hashed 23

  57. Compiler constructs e ffi cient algorithm by reasoning about whether operands support required high-level operations Random access B C ∘ Dense Compressed Hashed × Range Compressed Singleton Offset Iterate over B and random access C Hashed 23

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

A Sparse Tensor Format and a Benchmark Suite Jiajia Li Pacific Northwest National Laboratory

A Sparse Tensor Format and a Benchmark Suite Jiajia Li Pacific Northwest National Laboratory

A Sparse Tensor Format and a Benchmark Suite Jiajia Li Pacific Northwest National Laboratory January 25, 2019 @ MIT Figure sources: A brief survey of tensors by Berton Earnshaw and NVIDIA Tensor Cores HiCOO: Hierarchical Storage of Sparse

272 views • 23 slides
Invited Workshop on Compiler Techniques for Sparse Tensor Algebra Cambridge MA, January 26th

Invited Workshop on Compiler Techniques for Sparse Tensor Algebra Cambridge MA, January 26th

Invited Workshop on Compiler Techniques for Sparse Tensor Algebra Cambridge MA, January 26th Sparse Tensors are Everywhere Data Analytics Machine Learning Science and Engineering Movie ratings Convolutional Networks Product Reviews QCD

427 views • 29 slides
Efficient compilation of complex tensor algebra expressions Martin Sandve Alns Center for

Efficient compilation of complex tensor algebra expressions Martin Sandve Alns Center for

Efficient compilation of complex tensor algebra expressions Martin Sandve Alns Center for Biomedical Computing June 5th, FEniCS 2012 UFL is a DSL, symbolic backend, and frontend to form compilers Users recently reported scalabiliy

311 views • 18 slides
TENSOR ALGEBRA Continuum Mechanics Course (MMC) - ETSECCPB - UPC Introduction to Tensors Tensor

TENSOR ALGEBRA Continuum Mechanics Course (MMC) - ETSECCPB - UPC Introduction to Tensors Tensor

TENSOR ALGEBRA Continuum Mechanics Course (MMC) - ETSECCPB - UPC Introduction to Tensors Tensor Algebra 2 Introduction SCALAR , , ... v VECTOR v f , , ... , , ... MATRIX ? , ... C 3 Concept of Tensor A TENSOR

885 views • 88 slides
Tensor-Matrix Products with a Compressed Sparse Tensor Shaden Smith George Karypis University

Tensor-Matrix Products with a Compressed Sparse Tensor Shaden Smith George Karypis University

Tensor-Matrix Products with a Compressed Sparse Tensor Shaden Smith George Karypis University of Minnesota Department of Computer Science &amp; Engineering shaden@cs.umn.edu Tensor-Matrix Products with a Compressed Sparse Tensor 1 / 47

1.09k views • 47 slides
Sparse Tensor Factorization: Algorithms, Data Structures, and Challenges Shaden Smith &amp;

Sparse Tensor Factorization: Algorithms, Data Structures, and Challenges Shaden Smith &amp;

Sparse Tensor Factorization: Algorithms, Data Structures, and Challenges Shaden Smith &amp; George Karypis University of Minnesota Department of Computer Science &amp; Engineering shaden@cs.umn.edu Sparse Tensor Factorization: Algorithms,

844 views • 80 slides
A Medium-Grained Algorithm for Distributed Sparse Tensor Factorization Shaden Smith George

A Medium-Grained Algorithm for Distributed Sparse Tensor Factorization Shaden Smith George

A Medium-Grained Algorithm for Distributed Sparse Tensor Factorization Shaden Smith George Karypis University of Minnesota Department of Computer Science &amp; Engineering shaden@cs.umn.edu Medium-Grained Sparse Tensor Factorization 1 / 24

424 views • 25 slides
Feeding of the Thousands Leveraging the GPU's Computing Power for Sparse Linear Algebra

Feeding of the Thousands Leveraging the GPU's Computing Power for Sparse Linear Algebra

SPPEXA Annual Meeting 2016, January 25 th , 2016, Garching, Germany Feeding of the Thousands Leveraging the GPU's Computing Power for Sparse Linear Algebra Hartwig Anzt Sparse Linear Algebra on GPUs In Inherently parallel operations

336 views • 21 slides
Tensor Field Techniques Lecture 11 March 5, 2020 Outline Basics of tensor algebra Tensor

Tensor Field Techniques Lecture 11 March 5, 2020 Outline Basics of tensor algebra Tensor

CS53000 - Spring 2020 Introduction to Scientific Visualization Tensor Field Techniques Lecture 11 March 5, 2020 Outline Basics of tensor algebra Tensor glyphs Hyperstreamlines DTI visualization CS53000 / Spring 2020 : Introduction to

1.07k views • 84 slides
Multilinear Algebra and Tensor Decomposition Qibin Zhao Tensor Learning Unit RIKEN AIP

Multilinear Algebra and Tensor Decomposition Qibin Zhao Tensor Learning Unit RIKEN AIP

Multilinear Algebra and Tensor Decomposition Qibin Zhao Tensor Learning Unit RIKEN AIP 2018-6-2 @ Waseda University Self-Introduction 2009, Ph.D. in Computer Science, Shanghai Jiao Tong University 2009 - 2017, RIKEN Brain Science

938 views • 57 slides
Tensor learning approach to sparse QMC sampling of two-particle Greens function in DMFT

Tensor learning approach to sparse QMC sampling of two-particle Greens function in DMFT

Tensor learning approach to sparse QMC sampling of two-particle Greens function in DMFT Hiroshi SHINAOKA Collaborators N. Chikano, J. Otsuki, M. Ohzeki, K. Yoshimi, K. Haule, M. Wallerberger, J. Li, E. Gull, D. Geffroy, J. Kune Matsubara

337 views • 17 slides
Tensor MUSIC in Multidimensional Sparse Arrays Chun-Lin Liu 1 and P . Vaidyanathan 2 . P Dept. of

Tensor MUSIC in Multidimensional Sparse Arrays Chun-Lin Liu 1 and P . Vaidyanathan 2 . P Dept. of

Tensor MUSIC in Multidimensional Sparse Arrays Chun-Lin Liu 1 and P . Vaidyanathan 2 . P Dept. of Electrical Engineering, MC 136-93 California Institute of Technology, Pasadena, CA 91125, USA cl.liu@caltech.edu 1 , ppvnath@systems.caltech.edu 2

523 views • 24 slides
Sparse Tensor Factorization on Many-Core Processors with High-Bandwidth Memory Shaden Smith 1

Sparse Tensor Factorization on Many-Core Processors with High-Bandwidth Memory Shaden Smith 1

Sparse Tensor Factorization on Many-Core Processors with High-Bandwidth Memory Shaden Smith 1 , Jongsoo Park 2 , and George Karypis 1 1 Department of Computer Science &amp; Engineering, University of Minnesota 2 Parallel Computing Lab, Intel

802 views • 37 slides
(Some) Challenges in (Some) Challenges in Tensor Mining Tensor Mining Evrim Acar Sandia

(Some) Challenges in (Some) Challenges in Tensor Mining Tensor Mining Evrim Acar Sandia

(Some) Challenges in (Some) Challenges in Tensor Mining Tensor Mining Evrim Acar Sandia National Labs., Livermore, CA Tensor Mining Tensor Mining Parafac Parafac unsupervised unsupervised = + + X X E X X = + dense or sparse

340 views • 17 slides
Parallel Sparse Tensor Decomposition in Chapel Thomas B. Rolinger , Tyler A. Simon, Christopher

Parallel Sparse Tensor Decomposition in Chapel Thomas B. Rolinger , Tyler A. Simon, Christopher

Parallel Sparse Tensor Decomposition in Chapel Thomas B. Rolinger , Tyler A. Simon, Christopher D. Krieger IPDPSW 2018 CHIUW Outline 1. Motivation and Background 2. Porting SPLATT to Chapel 3. Performance Evaluation: Experiments,

855 views • 52 slides
Parallel Numerical Algorithms Chapter 7 Differential Equations Section 7.5 Tensor

Parallel Numerical Algorithms Chapter 7 Differential Equations Section 7.5 Tensor

Tensor Algebra Tensor Decompositions Fast Algorithms Parallel Numerical Algorithms Chapter 7 Differential Equations Section 7.5 Tensor Analysis Edgar Solomonik Department of Computer Science University of Illinois at Urbana-Champaign

722 views • 26 slides
+ Bellringer What is the overall topic of your essay? Your THEME 1. Name the five types of

+ Bellringer What is the overall topic of your essay? Your THEME 1. Name the five types of

+ Bellringer What is the overall topic of your essay? Your THEME 1. Name the five types of paragraphs that should make up 2. your essay. Introduction, Body Paragraph 1, 2, 3, Conclusion What are the main ideas of each of your body paragraphs? 3.

187 views • 8 slides
ClausIE: Clause-Based Open Information Extraction Luciano Del Corro Rainer Gemulla

ClausIE: Clause-Based Open Information Extraction Luciano Del Corro Rainer Gemulla

ClausIE: Clause-Based Open Information Extraction Luciano Del Corro Rainer Gemulla Max-Planck-Institut fr Informatik May 2013 Del Corro, Gemulla (MPI) ClausIE May 2013 1 / 18 Open Information Extraction: From sentences to propositions

758 views • 72 slides
Building to Scale Shaping Digital Coursework with Tools and Partnerships Carol Lasser Oberlin

Building to Scale Shaping Digital Coursework with Tools and Partnerships Carol Lasser Oberlin

Building to Scale Shaping Digital Coursework with Tools and Partnerships Carol Lasser Oberlin College Katie Holt // @holtkatie The College of Wooster Jon Breitenbucher // @j_breitenbucher The College of Wooster John Krygier //

665 views • 34 slides
Welcome Todays Agenda Welcome Remarks by President Karbhari Overview by Jim Grover

Welcome Todays Agenda Welcome Remarks by President Karbhari Overview by Jim Grover

THE UNIVERSITY OF TEXAS AT ARLINGTON Forum: Controls and Uses of Autonomous Vehicles' Welcome Todays Agenda Welcome Remarks by President Karbhari Overview by Jim Grover Presentations by Frank Lewis Yan Wan Animesh Chakravarthy

835 views • 71 slides
Hearing #11 on Competition and Consumer Protection in the 21st Century Federal Trade Commission

Hearing #11 on Competition and Consumer Protection in the 21st Century Federal Trade Commission

Hearing #11 on Competition and Consumer Protection in the 21st Century Federal Trade Commission Headquarters March 25, 2019 1 Welcome We Will Be Starting Shortly 2 Welcome and Introductory Remarks Randolph Tritell Federal Trade Commission

519 views • 35 slides
Device-independent Randomness Expansion with Entangled Photons Yanbao Zhang NTT Research Center

Device-independent Randomness Expansion with Entangled Photons Yanbao Zhang NTT Research Center

Device-independent Randomness Expansion with Entangled Photons Yanbao Zhang NTT Research Center for Theoretical Quantum Physics NTT Basic Research Lab, Japan Based on the joint work arXiv:1912.11158 with Krister Shalm, Joshua Bienfang, Collin

420 views • 23 slides
Combining Estimates from Related Surveys via Bivariate Models (Application: using ACS estimates

Combining Estimates from Related Surveys via Bivariate Models (Application: using ACS estimates

Combining Estimates from Related Surveys via Bivariate Models (Application: using ACS estimates to improve estimates from smaller U.S. surveys) William R. Bell and Carolina Franco, U.S. Census Bureau 2016 Ross-Royall Symposium February 26,

631 views • 28 slides
Quantum Entanglement and the Bell Matrix Marco Pedicini (Roma Tre University) in collaboration

Quantum Entanglement and the Bell Matrix Marco Pedicini (Roma Tre University) in collaboration

Quantum Entanglement and the Bell Matrix Marco Pedicini (Roma Tre University) in collaboration with Anna Chiara Lai and Silvia Rognone (La Sapienza University of Rome) SIMAI2018 - MS27: Discrete Mathematics, Number Theory and Applications to

714 views • 45 slides

More recommend