maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/mlir/test/Transforms/unroll.mlir

505 lines
21 KiB
MLIR
Raw Normal View History

Implement support for constant folding operations and a simple constant folding optimization pass: - Give the ability for operations to implement a constantFold hook (a simple one for single-result ops as well as general support for multi-result ops). - Implement folding support for constant and addf. - Implement support in AbstractOperation and Operation to make this usable by clients. - Implement a very simple constant folding pass that does top down folding on CFG and ML functions, with a testcase that exercises all the above stuff. Random cleanups: - Improve the build APIs for ConstantOp. - Stop passing "-o -" to mlir-opt in the testsuite, since that is the default. PiperOrigin-RevId: 213749809
2018-09-20 12:35:11 +08:00
// RUN: mlir-opt %s -loop-unroll -unroll-full | FileCheck %s
// RUN: mlir-opt %s -loop-unroll -unroll-full -unroll-full-threshold=2 | FileCheck %s --check-prefix SHORT
// RUN: mlir-opt %s -loop-unroll -unroll-factor=4 | FileCheck %s --check-prefix UNROLL-BY-4
Sketch out loop unrolling transformation. - Implement a full loop unroll for innermost loops. - Use it to implement a pass that unroll all the innermost loops of all mlfunction's in a module. ForStmt's parsed currently have constant trip counts (and constant loop bounds). - Implement StmtVisitor based (Visitor pattern) Loop IVs aren't currently parsed and represented as SSA values. Replacing uses of loop IVs in unrolled bodies is thus a TODO. Class comments are sparse at some places - will add them after one round of comments. A cmd-line flag triggers this for now. Original: mlfunc @loops() { for x = 1 to 100 step 2 { for x = 1 to 4 { "Const"(){value: 1} : () -> () } } return } After unrolling: mlfunc @loops() { for x = 1 to 100 step 2 { "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () } return } PiperOrigin-RevId: 205933235
2018-07-25 11:01:16 +08:00
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// CHECK: #map0 = (d0) -> (d0 + 1)
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: #map1 = (d0) -> (d0 + 2)
// CHECK: #map2 = (d0) -> (d0 + 3)
// CHECK: #map3 = (d0) -> (d0 + 4)
// CHECK: #map4 = (d0, d1) -> (d0 + 1, d1 + 2)
// CHECK: #map5 = (d0, d1) -> (d0 + 3, d1 + 4)
// CHECK: #map6 = (d0)[s0] -> (d0 + s0 + 1)
// CHECK: #map7 = (d0) -> (d0 + 5)
// CHECK: #map8 = (d0) -> (d0 + 6)
// CHECK: #map9 = (d0) -> (d0 + 7)
// CHECK: #map10 = (d0, d1) -> (d0 * 16 + d1)
// CHECK: #map11 = (d0) -> (d0 + 8)
// CHECK: #map12 = (d0) -> (d0 + 9)
// CHECK: #map13 = (d0) -> (d0 + 10)
// CHECK: #map14 = (d0) -> (d0 + 15)
// CHECK: #map15 = (d0) -> (d0 + 20)
// CHECK: #map16 = (d0) -> (d0 + 25)
// CHECK: #map17 = (d0) -> (d0 + 30)
// CHECK: #map18 = (d0) -> (d0 + 35)
// SHORT: #map0 = (d0) -> (d0 + 1)
// SHORT: #map1 = (d0) -> (d0 + 2)
// SHORT: #map2 = (d0, d1) -> (d0 + 1, d1 + 2)
// SHORT: #map3 = (d0, d1) -> (d0 + 3, d1 + 4)
// SHORT: #map4 = (d0)[s0] -> (d0 + s0 + 1)
// SHORT: #map5 = (d0, d1) -> (d0 * 16 + d1)
// UNROLL-BY-4: #map0 = (d0) -> (d0 + 1)
// UNROLL-BY-4: #map1 = (d0) -> (d0 + 2)
// UNROLL-BY-4: #map2 = (d0) -> (d0 + 3)
// UNROLL-BY-4: #map3 = (d0, d1) -> (d0 + 1, d1 + 2)
// UNROLL-BY-4: #map4 = (d0, d1) -> (d0 + 3, d1 + 4)
// UNROLL-BY-4: #map5 = (d0)[s0] -> (d0 + s0 + 1)
// UNROLL-BY-4: #map6 = (d0, d1) -> (d0 * 16 + d1)
// UNROLL-BY-4: #map7 = (d0) -> (d0 + 5)
// UNROLL-BY-4: #map8 = (d0) -> (d0 + 10)
// UNROLL-BY-4: #map9 = (d0) -> (d0 + 15)
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-LABEL: mlfunc @loop_nest_simplest() {
mlfunc @loop_nest_simplest() {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK: for %i0 = 1 to 100 step 2 {
Sketch out loop unrolling transformation. - Implement a full loop unroll for innermost loops. - Use it to implement a pass that unroll all the innermost loops of all mlfunction's in a module. ForStmt's parsed currently have constant trip counts (and constant loop bounds). - Implement StmtVisitor based (Visitor pattern) Loop IVs aren't currently parsed and represented as SSA values. Replacing uses of loop IVs in unrolled bodies is thus a TODO. Class comments are sparse at some places - will add them after one round of comments. A cmd-line flag triggers this for now. Original: mlfunc @loops() { for x = 1 to 100 step 2 { for x = 1 to 4 { "Const"(){value: 1} : () -> () } } return } After unrolling: mlfunc @loops() { for x = 1 to 100 step 2 { "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () } return } PiperOrigin-RevId: 205933235
2018-07-25 11:01:16 +08:00
for %i = 1 to 100 step 2 {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK: %c1_i32 = constant 1 : i32
// CHECK-NEXT: %c1_i32_0 = constant 1 : i32
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
// CHECK-NEXT: %c1_i32_1 = constant 1 : i32
// CHECK-NEXT: %c1_i32_2 = constant 1 : i32
Sketch out loop unrolling transformation. - Implement a full loop unroll for innermost loops. - Use it to implement a pass that unroll all the innermost loops of all mlfunction's in a module. ForStmt's parsed currently have constant trip counts (and constant loop bounds). - Implement StmtVisitor based (Visitor pattern) Loop IVs aren't currently parsed and represented as SSA values. Replacing uses of loop IVs in unrolled bodies is thus a TODO. Class comments are sparse at some places - will add them after one round of comments. A cmd-line flag triggers this for now. Original: mlfunc @loops() { for x = 1 to 100 step 2 { for x = 1 to 4 { "Const"(){value: 1} : () -> () } } return } After unrolling: mlfunc @loops() { for x = 1 to 100 step 2 { "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () } return } PiperOrigin-RevId: 205933235
2018-07-25 11:01:16 +08:00
for %j = 1 to 4 {
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
%x = constant 1 : i32
Sketch out loop unrolling transformation. - Implement a full loop unroll for innermost loops. - Use it to implement a pass that unroll all the innermost loops of all mlfunction's in a module. ForStmt's parsed currently have constant trip counts (and constant loop bounds). - Implement StmtVisitor based (Visitor pattern) Loop IVs aren't currently parsed and represented as SSA values. Replacing uses of loop IVs in unrolled bodies is thus a TODO. Class comments are sparse at some places - will add them after one round of comments. A cmd-line flag triggers this for now. Original: mlfunc @loops() { for x = 1 to 100 step 2 { for x = 1 to 4 { "Const"(){value: 1} : () -> () } } return } After unrolling: mlfunc @loops() { for x = 1 to 100 step 2 { "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () "Const"(){value: 1} : () -> () } return } PiperOrigin-RevId: 205933235
2018-07-25 11:01:16 +08:00
}
} // CHECK: }
return // CHECK: return
} // CHECK }
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-LABEL: mlfunc @loop_nest_simple_iv_use() {
mlfunc @loop_nest_simple_iv_use() {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK: %c1 = constant 1 : affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: for %i0 = 1 to 100 step 2 {
for %i = 1 to 100 step 2 {
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: %0 = "addi32"(%c1, %c1) : (affineint, affineint) -> i32
// CHECK: %1 = affine_apply #map0(%c1)
// CHECK-NEXT: %2 = "addi32"(%1, %1) : (affineint, affineint) -> i32
// CHECK: %3 = affine_apply #map1(%c1)
// CHECK-NEXT: %4 = "addi32"(%3, %3) : (affineint, affineint) -> i32
// CHECK: %5 = affine_apply #map2(%c1)
// CHECK-NEXT: %6 = "addi32"(%5, %5) : (affineint, affineint) -> i32
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
for %j = 1 to 4 {
%x = "addi32"(%j, %j) : (affineint, affineint) -> i32
}
} // CHECK: }
return // CHECK: return
} // CHECK }
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// Operations in the loop body have results that are used therein.
// CHECK-LABEL: mlfunc @loop_nest_body_def_use() {
mlfunc @loop_nest_body_def_use() {
// CHECK: %c0 = constant 0 : affineint
// CHECK-NEXT: for %i0 = 1 to 100 step 2 {
for %i = 1 to 100 step 2 {
// CHECK: %c0_0 = constant 0 : affineint
%c0 = constant 0 : affineint
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// CHECK: %0 = affine_apply #map0(%c0)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK-NEXT: %1 = "addi32"(%0, %c0_0) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %2 = affine_apply #map0(%c0)
// CHECK-NEXT: %3 = affine_apply #map0(%2)
// CHECK-NEXT: %4 = "addi32"(%3, %c0_0) : (affineint, affineint) -> affineint
// CHECK-NEXT: %5 = affine_apply #map1(%c0)
// CHECK-NEXT: %6 = affine_apply #map0(%5)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK-NEXT: %7 = "addi32"(%6, %c0_0) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %8 = affine_apply #map2(%c0)
// CHECK-NEXT: %9 = affine_apply #map0(%8)
// CHECK-NEXT: %10 = "addi32"(%9, %c0_0) : (affineint, affineint) -> affineint
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
for %j = 0 to 3 {
%x = "affine_apply" (%j) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
%y = "addi32"(%x, %c0) : (affineint, affineint) -> affineint
}
} // CHECK: }
return // CHECK: return
} // CHECK }
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-LABEL: mlfunc @loop_nest_strided() {
mlfunc @loop_nest_strided() {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK: %c3 = constant 3 : affineint
// CHECK-NEXT: %c3_0 = constant 3 : affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: for %i0 = 1 to 100 {
for %i = 1 to 100 {
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// CHECK: %0 = affine_apply #map0(%c3_0)
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: %1 = "addi32"(%0, %0) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %2 = affine_apply #map1(%c3_0)
// CHECK-NEXT: %3 = affine_apply #map0(%2)
// CHECK-NEXT: %4 = "addi32"(%3, %3) : (affineint, affineint) -> affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
for %j = 3 to 6 step 2 {
%x = "affine_apply" (%j) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
%y = "addi32"(%x, %x) : (affineint, affineint) -> affineint
}
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: %5 = affine_apply #map0(%c3)
// CHECK-NEXT: %6 = "addi32"(%5, %5) : (affineint, affineint) -> affineint
// CHECK-NEXT: %7 = affine_apply #map1(%c3)
// CHECK-NEXT: %8 = affine_apply #map0(%7)
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: %9 = "addi32"(%8, %8) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %10 = affine_apply #map3(%c3)
// CHECK-NEXT: %11 = affine_apply #map0(%10)
// CHECK-NEXT: %12 = "addi32"(%11, %11) : (affineint, affineint) -> affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
for %k = 3 to 7 step 2 {
%z = "affine_apply" (%k) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
%w = "addi32"(%z, %z) : (affineint, affineint) -> affineint
}
} // CHECK: }
return // CHECK: return
} // CHECK }
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK-LABEL: mlfunc @loop_nest_multiple_results() {
mlfunc @loop_nest_multiple_results() {
// CHECK: %c0 = constant 0 : affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: for %i0 = 1 to 100 {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
for %i = 1 to 100 {
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: %0 = affine_apply #map4(%i0, %c0)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK-NEXT: %1 = "addi32"(%0#0, %0#1) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %2 = affine_apply #map5(%i0, %c0)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK-NEXT: %3 = "fma"(%2#0, %2#1, %0#0) : (affineint, affineint, affineint) -> (affineint, affineint)
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %4 = affine_apply #map0(%c0)
// CHECK-NEXT: %5 = affine_apply #map4(%i0, %4)
// CHECK-NEXT: %6 = "addi32"(%5#0, %5#1) : (affineint, affineint) -> affineint
// CHECK-NEXT: %7 = affine_apply #map5(%i0, %4)
// CHECK-NEXT: %8 = "fma"(%7#0, %7#1, %5#0) : (affineint, affineint, affineint) -> (affineint, affineint)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
for %j = 0 to 1 step 1 {
%x = "affine_apply" (%i, %j) { map: (d0, d1) -> (d0 + 1, d1 + 2) } :
(affineint, affineint) -> (affineint, affineint)
%y = "addi32"(%x#0, %x#1) : (affineint, affineint) -> affineint
%z = "affine_apply" (%i, %j) { map: (d0, d1) -> (d0 + 3, d1 + 4) } :
(affineint, affineint) -> (affineint, affineint)
%w = "fma"(%z#0, %z#1, %x#0) : (affineint, affineint, affineint) -> (affineint, affineint)
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
}
} // CHECK: }
return // CHECK: return
} // CHECK }
// Imperfect loop nest. Unrolling innermost here yields a perfect nest.
Implement ML function arguments. Add representation for argument list in ML Function using TrailingObjects template. Implement argument iterators, parsing and printing. Unrelated minor change - remove OperationStmt::dropReferences(). Since MLFunction does not have cyclic operand references (it's an AST) destruction can be safely done w/o a special pass to drop references. PiperOrigin-RevId: 207583024
2018-08-07 02:54:39 +08:00
// CHECK-LABEL: mlfunc @loop_nest_seq_imperfect(%arg0 : memref<128x128xf32>) {
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
mlfunc @loop_nest_seq_imperfect(%a : memref<128x128xf32>) {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK: %c1 = constant 1 : affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: %c128 = constant 128 : affineint
%c128 = constant 128 : affineint
// CHECK: for %i0 = 1 to 100 {
for %i = 1 to 100 {
// CHECK: %0 = "vld"(%i0) : (affineint) -> i32
%ld = "vld"(%i) : (affineint) -> i32
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// CHECK: %1 = affine_apply #map0(%c1)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK-NEXT: %2 = "vmulf"(%c1, %1) : (affineint, affineint) -> affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: %3 = "vaddf"(%2, %2) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %4 = affine_apply #map0(%c1)
// CHECK-NEXT: %5 = affine_apply #map0(%4)
// CHECK-NEXT: %6 = "vmulf"(%4, %5) : (affineint, affineint) -> affineint
// CHECK-NEXT: %7 = "vaddf"(%6, %6) : (affineint, affineint) -> affineint
// CHECK-NEXT: %8 = affine_apply #map1(%c1)
// CHECK-NEXT: %9 = affine_apply #map0(%8)
// CHECK-NEXT: %10 = "vmulf"(%8, %9) : (affineint, affineint) -> affineint
// CHECK-NEXT: %11 = "vaddf"(%10, %10) : (affineint, affineint) -> affineint
// CHECK-NEXT: %12 = affine_apply #map2(%c1)
// CHECK-NEXT: %13 = affine_apply #map0(%12)
// CHECK-NEXT: %14 = "vmulf"(%12, %13) : (affineint, affineint) -> affineint
// CHECK-NEXT: %15 = "vaddf"(%14, %14) : (affineint, affineint) -> affineint
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
for %j = 1 to 4 {
%x = "affine_apply" (%j) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
%y = "vmulf"(%j, %x) : (affineint, affineint) -> affineint
%z = "vaddf"(%y, %y) : (affineint, affineint) -> affineint
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
}
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: %16 = "scale"(%c128, %i0) : (affineint, affineint) -> affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
%addr = "scale"(%c128, %i) : (affineint, affineint) -> affineint
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: "vst"(%16, %i0) : (affineint, affineint) -> ()
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
"vst"(%addr, %i) : (affineint, affineint) -> ()
} // CHECK }
return // CHECK: return
}
// CHECK-LABEL: mlfunc @loop_nest_seq_multiple() {
mlfunc @loop_nest_seq_multiple() {
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// CHECK: %c1 = constant 1 : affineint
// CHECK-NEXT: %c0 = constant 0 : affineint
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// CHECK-NEXT: %0 = affine_apply #map0(%c0)
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: "mul"(%0, %0) : (affineint, affineint) -> ()
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %1 = affine_apply #map0(%c0)
// CHECK-NEXT: %2 = affine_apply #map0(%1)
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK-NEXT: "mul"(%2, %2) : (affineint, affineint) -> ()
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK-NEXT: %3 = affine_apply #map1(%c0)
// CHECK-NEXT: %4 = affine_apply #map0(%3)
// CHECK-NEXT: "mul"(%4, %4) : (affineint, affineint) -> ()
// CHECK-NEXT: %5 = affine_apply #map2(%c0)
// CHECK-NEXT: %6 = affine_apply #map0(%5)
// CHECK-NEXT: "mul"(%6, %6) : (affineint, affineint) -> ()
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
for %j = 0 to 3 {
%x = "affine_apply" (%j) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
"mul"(%x, %x) : (affineint, affineint) -> ()
}
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
// CHECK: %c99 = constant 99 : affineint
%k = "constant"(){value: 99} : () -> affineint
Loop unrolling update. - deal with non-operation stmt's (if/for stmt's) in loops being unrolled (unrolling of non-innermost loops works). - update uses in unrolled bodies to use results of new operations that may be introduced in the unrolled bodies. Unrolling now works for all kinds of loop nests - perfect nests, imperfect nests, loops at any depth, and with any kind of operation in the body. (IfStmt support not done, hence untested there). Added missing dump/print method for StmtBlock. TODO: add test case for outer loop unrolling. PiperOrigin-RevId: 207314286
2018-08-04 04:22:26 +08:00
// CHECK: for %i0 = 1 to 100 step 2 {
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
for %m = 1 to 100 step 2 {
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// CHECK: %7 = affine_apply #map0(%c1)
// CHECK-NEXT: %8 = affine_apply #map6(%c1)[%c99]
// CHECK-NEXT: %9 = affine_apply #map0(%c1)
// CHECK-NEXT: %10 = affine_apply #map0(%9)
// CHECK-NEXT: %11 = affine_apply #map6(%9)[%c99]
// CHECK-NEXT: %12 = affine_apply #map1(%c1)
// CHECK-NEXT: %13 = affine_apply #map0(%12)
// CHECK-NEXT: %14 = affine_apply #map6(%12)[%c99]
// CHECK-NEXT: %15 = affine_apply #map2(%c1)
// CHECK-NEXT: %16 = affine_apply #map0(%15)
// CHECK-NEXT: %17 = affine_apply #map6(%15)[%c99]
MLStmt cloning and IV replacement for loop unrolling, add constant pool to MLFunctions. - MLStmt cloning and IV replacement - While at this, fix the innermostLoopGatherer to actually gather all the innermost loops (it was stopping its walk at the first innermost loop it found) - Improve comments for MLFunction statement classes, fix inheritance order. - Fixed StmtBlock destructor. PiperOrigin-RevId: 207049173
2018-08-02 13:36:12 +08:00
for %n = 1 to 4 {
%y = "affine_apply" (%n) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
%z = "affine_apply" (%n, %k) { map: (d0) [s0] -> (d0 + s0 + 1) } :
(affineint, affineint) -> (affineint)
} // CHECK }
} // CHECK }
return // CHECK: return
} // CHECK }
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// SHORT-LABEL: mlfunc @loop_nest_outer_unroll() {
mlfunc @loop_nest_outer_unroll() {
// SHORT: for %i0 = 1 to 4 {
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// SHORT-NEXT: %0 = affine_apply #map0(%i0)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// SHORT-NEXT: %1 = "addi32"(%0, %0) : (affineint, affineint) -> affineint
// SHORT-NEXT: }
// SHORT-NEXT: for %i1 = 1 to 4 {
Escape and unescape strings in the parser and printer so they can roundtrip, print floating point in a structured form that we know can round trip, enumerate attributes in the visitor so we print affine mapping attributes symbolically (the majority of the testcase updates). We still have an issue where the hexadecimal floating point syntax is reparsed as an integer, but that can evolve in subsequent patches. PiperOrigin-RevId: 208828876
2018-08-16 00:09:54 +08:00
// SHORT-NEXT: %2 = affine_apply #map0(%i1)
Loop unrolling pass update - fix/complete forStmt cloning for unrolling to work for outer loops - create IV const's only when needed - test outer loop unrolling by creating a short trip count unroll pass for loops with trip counts <= <parameter> - add unrolling test cases for multiple op results, outer loop unrolling - fix/clean up StmtWalker class while on this - switch unroll loop iterator values from i32 to affineint PiperOrigin-RevId: 207645967
2018-08-07 09:40:34 +08:00
// SHORT-NEXT: %3 = "addi32"(%2, %2) : (affineint, affineint) -> affineint
// SHORT-NEXT: }
for %i = 1 to 2 {
for %j = 1 to 4 {
%x = "affine_apply" (%j) { map: (d0) -> (d0 + 1) } :
(affineint) -> (affineint)
%y = "addi32"(%x, %x) : (affineint, affineint) -> affineint
}
}
return // SHORT: return
} // SHORT }
ShortLoopUnroll - bug fix. Collect loops through a post order walk instead of a pre-order so that loops are collected from inner loops are collected before outer surrounding ones. Add a complex test case. PiperOrigin-RevId: 209041057
2018-08-17 04:51:44 +08:00
// We aren't doing any file check here. We just need this test case to
// successfully run. Both %i0 and i1 will get unrolled here with the min trip
// count threshold set to 2.
// SHORT-LABEL: mlfunc @loop_nest_seq_long() -> i32 {
mlfunc @loop_nest_seq_long() -> i32 {
%A = alloc() : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
%B = alloc() : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
%C = alloc() : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
%zero = constant 0 : i32
%one = constant 1 : i32
%two = constant 2 : i32
%zero_idx = constant 0 : affineint
for %n0 = 0 to 512 {
for %n1 = 0 to 7 {
store %one, %A[%n0, %n1] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
store %two, %B[%n0, %n1] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
store %zero, %C[%n0, %n1] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
}
}
for %i0 = 0 to 1 {
for %i1 = 0 to 1 {
for %i2 = 0 to 7 {
%b2 = "affine_apply" (%i1, %i2) {map: (d0, d1) -> (16*d0 + d1)} : (affineint, affineint) -> affineint
%x = load %B[%i0, %b2] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
"op1"(%x) : (i32) -> ()
}
for %j1 = 0 to 7 {
for %j2 = 0 to 7 {
%a2 = "affine_apply" (%i1, %j2) {map: (d0, d1) -> (16*d0 + d1)} : (affineint, affineint) -> affineint
%v203 = load %A[%j1, %a2] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
"op2"(%v203) : (i32) -> ()
}
for %k2 = 0 to 7 {
%s0 = "op3"() : () -> i32
%c2 = "affine_apply" (%i0, %k2) {map: (d0, d1) -> (16*d0 + d1)} : (affineint, affineint) -> affineint
%s1 = load %C[%j1, %c2] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
%s2 = "addi32"(%s0, %s1) : (i32, i32) -> i32
store %s2, %C[%j1, %c2] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
}
}
"op4"() : () -> ()
}
}
%ret = load %C[%zero_idx, %zero_idx] : memref<512 x 512 x i32, (d0, d1) -> (d0, d1), 2>
return %ret : i32
}
Extend loop unrolling to unroll by a given factor; add builder for affine apply op. - add builder for AffineApplyOp (first one for an operation that has non-zero operands) - add support for loop unrolling by a given factor; uses the affine apply op builder. While on this, change 'step' of ForStmt to be 'unsigned' instead of AffineConstantExpr *. Add setters for ForStmt lb, ub, step. Sample Input: // CHECK-LABEL: mlfunc @loop_nest_unroll_cleanup() { mlfunc @loop_nest_unroll_cleanup() { for %i = 1 to 100 { for %j = 0 to 17 { %x = "addi32"(%j, %j) : (affineint, affineint) -> i32 %y = "addi32"(%x, %x) : (i32, i32) -> i32 } } return } Output: $ mlir-opt -loop-unroll -unroll-factor=4 /tmp/single2.mlir #map0 = (d0) -> (d0 + 1) #map1 = (d0) -> (d0 + 2) #map2 = (d0) -> (d0 + 3) mlfunc @loop_nest_unroll_cleanup() { for %i0 = 1 to 100 { for %i1 = 0 to 17 step 4 { %0 = "addi32"(%i1, %i1) : (affineint, affineint) -> i32 %1 = "addi32"(%0, %0) : (i32, i32) -> i32 %2 = affine_apply #map0(%i1) %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32 %4 = affine_apply #map1(%i1) %5 = "addi32"(%4, %4) : (affineint, affineint) -> i32 %6 = affine_apply #map2(%i1) %7 = "addi32"(%6, %6) : (affineint, affineint) -> i32 } for %i2 = 16 to 17 { %8 = "addi32"(%i2, %i2) : (affineint, affineint) -> i32 %9 = "addi32"(%8, %8) : (i32, i32) -> i32 } } return } PiperOrigin-RevId: 209676220
2018-08-22 07:01:23 +08:00
// UNROLL-BY-4-LABEL: mlfunc @unroll_unit_stride_no_cleanup() {
mlfunc @unroll_unit_stride_no_cleanup() {
// UNROLL-BY-4: for %i0 = 1 to 100 {
for %i = 1 to 100 {
// UNROLL-BY-4: for [[L1:%i[0-9]+]] = 1 to 8 step 4 {
// UNROLL-BY-4-NEXT: %0 = "addi32"([[L1]], [[L1]]) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %1 = "addi32"(%0, %0) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %2 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %4 = "addi32"(%3, %3) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %5 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %6 = "addi32"(%5, %5) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %7 = "addi32"(%6, %6) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %8 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %9 = "addi32"(%8, %8) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %10 = "addi32"(%9, %9) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: }
for %j = 1 to 8 {
%x = "addi32"(%j, %j) : (affineint, affineint) -> i32
%y = "addi32"(%x, %x) : (i32, i32) -> i32
}
// empty loop
// UNROLL-BY-4: for %i2 = 1 to 8 {
for %k = 1 to 8 {
}
}
return
}
// UNROLL-BY-4-LABEL: mlfunc @unroll_unit_stride_cleanup() {
mlfunc @unroll_unit_stride_cleanup() {
// UNROLL-BY-4: for %i0 = 1 to 100 {
for %i = 1 to 100 {
// UNROLL-BY-4: for [[L1:%i[0-9]+]] = 1 to 8 step 4 {
// UNROLL-BY-4-NEXT: %0 = "addi32"([[L1]], [[L1]]) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %1 = "addi32"(%0, %0) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %2 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %4 = "addi32"(%3, %3) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %5 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %6 = "addi32"(%5, %5) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %7 = "addi32"(%6, %6) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %8 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %9 = "addi32"(%8, %8) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %10 = "addi32"(%9, %9) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: }
// UNROLL-BY-4-NEXT: for [[L2:%i[0-9]+]] = 9 to 10 {
// UNROLL-BY-4-NEXT: %11 = "addi32"([[L2]], [[L2]]) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %12 = "addi32"(%11, %11) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: }
for %j = 1 to 10 {
%x = "addi32"(%j, %j) : (affineint, affineint) -> i32
%y = "addi32"(%x, %x) : (i32, i32) -> i32
}
}
return
}
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// UNROLL-BY-4-LABEL: mlfunc @unroll_non_unit_stride_cleanup() {
Extend loop unrolling to unroll by a given factor; add builder for affine apply op. - add builder for AffineApplyOp (first one for an operation that has non-zero operands) - add support for loop unrolling by a given factor; uses the affine apply op builder. While on this, change 'step' of ForStmt to be 'unsigned' instead of AffineConstantExpr *. Add setters for ForStmt lb, ub, step. Sample Input: // CHECK-LABEL: mlfunc @loop_nest_unroll_cleanup() { mlfunc @loop_nest_unroll_cleanup() { for %i = 1 to 100 { for %j = 0 to 17 { %x = "addi32"(%j, %j) : (affineint, affineint) -> i32 %y = "addi32"(%x, %x) : (i32, i32) -> i32 } } return } Output: $ mlir-opt -loop-unroll -unroll-factor=4 /tmp/single2.mlir #map0 = (d0) -> (d0 + 1) #map1 = (d0) -> (d0 + 2) #map2 = (d0) -> (d0 + 3) mlfunc @loop_nest_unroll_cleanup() { for %i0 = 1 to 100 { for %i1 = 0 to 17 step 4 { %0 = "addi32"(%i1, %i1) : (affineint, affineint) -> i32 %1 = "addi32"(%0, %0) : (i32, i32) -> i32 %2 = affine_apply #map0(%i1) %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32 %4 = affine_apply #map1(%i1) %5 = "addi32"(%4, %4) : (affineint, affineint) -> i32 %6 = affine_apply #map2(%i1) %7 = "addi32"(%6, %6) : (affineint, affineint) -> i32 } for %i2 = 16 to 17 { %8 = "addi32"(%i2, %i2) : (affineint, affineint) -> i32 %9 = "addi32"(%8, %8) : (i32, i32) -> i32 } } return } PiperOrigin-RevId: 209676220
2018-08-22 07:01:23 +08:00
mlfunc @unroll_non_unit_stride_cleanup() {
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// UNROLL-BY-4: for %i0 = 1 to 100 {
Extend loop unrolling to unroll by a given factor; add builder for affine apply op. - add builder for AffineApplyOp (first one for an operation that has non-zero operands) - add support for loop unrolling by a given factor; uses the affine apply op builder. While on this, change 'step' of ForStmt to be 'unsigned' instead of AffineConstantExpr *. Add setters for ForStmt lb, ub, step. Sample Input: // CHECK-LABEL: mlfunc @loop_nest_unroll_cleanup() { mlfunc @loop_nest_unroll_cleanup() { for %i = 1 to 100 { for %j = 0 to 17 { %x = "addi32"(%j, %j) : (affineint, affineint) -> i32 %y = "addi32"(%x, %x) : (i32, i32) -> i32 } } return } Output: $ mlir-opt -loop-unroll -unroll-factor=4 /tmp/single2.mlir #map0 = (d0) -> (d0 + 1) #map1 = (d0) -> (d0 + 2) #map2 = (d0) -> (d0 + 3) mlfunc @loop_nest_unroll_cleanup() { for %i0 = 1 to 100 { for %i1 = 0 to 17 step 4 { %0 = "addi32"(%i1, %i1) : (affineint, affineint) -> i32 %1 = "addi32"(%0, %0) : (i32, i32) -> i32 %2 = affine_apply #map0(%i1) %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32 %4 = affine_apply #map1(%i1) %5 = "addi32"(%4, %4) : (affineint, affineint) -> i32 %6 = affine_apply #map2(%i1) %7 = "addi32"(%6, %6) : (affineint, affineint) -> i32 } for %i2 = 16 to 17 { %8 = "addi32"(%i2, %i2) : (affineint, affineint) -> i32 %9 = "addi32"(%8, %8) : (i32, i32) -> i32 } } return } PiperOrigin-RevId: 209676220
2018-08-22 07:01:23 +08:00
for %i = 1 to 100 {
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// UNROLL-BY-4: for [[L1:%i[0-9]+]] = 2 to 37 step 20 {
// UNROLL-BY-4-NEXT: %0 = "addi32"([[L1]], [[L1]]) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %1 = "addi32"(%0, %0) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %2 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %4 = "addi32"(%3, %3) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %5 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %6 = "addi32"(%5, %5) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %7 = "addi32"(%6, %6) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: %8 = affine_apply #map{{[0-9]+}}([[L1]])
// UNROLL-BY-4-NEXT: %9 = "addi32"(%8, %8) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %10 = "addi32"(%9, %9) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: }
// UNROLL-BY-4-NEXT: for [[L2:%i[0-9]+]] = 42 to 48 step 5 {
// UNROLL-BY-4-NEXT: %11 = "addi32"([[L2]], [[L2]]) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %12 = "addi32"(%11, %11) : (i32, i32) -> i32
// UNROLL-BY-4-NEXT: }
for %j = 2 to 48 step 5 {
Extend loop unrolling to unroll by a given factor; add builder for affine apply op. - add builder for AffineApplyOp (first one for an operation that has non-zero operands) - add support for loop unrolling by a given factor; uses the affine apply op builder. While on this, change 'step' of ForStmt to be 'unsigned' instead of AffineConstantExpr *. Add setters for ForStmt lb, ub, step. Sample Input: // CHECK-LABEL: mlfunc @loop_nest_unroll_cleanup() { mlfunc @loop_nest_unroll_cleanup() { for %i = 1 to 100 { for %j = 0 to 17 { %x = "addi32"(%j, %j) : (affineint, affineint) -> i32 %y = "addi32"(%x, %x) : (i32, i32) -> i32 } } return } Output: $ mlir-opt -loop-unroll -unroll-factor=4 /tmp/single2.mlir #map0 = (d0) -> (d0 + 1) #map1 = (d0) -> (d0 + 2) #map2 = (d0) -> (d0 + 3) mlfunc @loop_nest_unroll_cleanup() { for %i0 = 1 to 100 { for %i1 = 0 to 17 step 4 { %0 = "addi32"(%i1, %i1) : (affineint, affineint) -> i32 %1 = "addi32"(%0, %0) : (i32, i32) -> i32 %2 = affine_apply #map0(%i1) %3 = "addi32"(%2, %2) : (affineint, affineint) -> i32 %4 = affine_apply #map1(%i1) %5 = "addi32"(%4, %4) : (affineint, affineint) -> i32 %6 = affine_apply #map2(%i1) %7 = "addi32"(%6, %6) : (affineint, affineint) -> i32 } for %i2 = 16 to 17 { %8 = "addi32"(%i2, %i2) : (affineint, affineint) -> i32 %9 = "addi32"(%8, %8) : (i32, i32) -> i32 } } return } PiperOrigin-RevId: 209676220
2018-08-22 07:01:23 +08:00
%x = "addi32"(%j, %j) : (affineint, affineint) -> i32
%y = "addi32"(%x, %x) : (i32, i32) -> i32
}
}
return
}
Add utility to promote single iteration loops. Add methods for getting constant loop counts. Improve / refactor loop unroll / loop unroll and jam. - add utility to remove single iteration loops. - use this utility to promote single iteration loops after unroll/unroll-and-jam - use loopUnrollByFactor for loopUnrollFull and remove most of the latter. - add methods for getting constant loop trip count PiperOrigin-RevId: 212039569
2018-09-08 05:47:21 +08:00
// Both the unrolled loop and the cleanup loop are single iteration loops.
mlfunc @loop_nest_single_iteration_after_unroll(%N: affineint) {
// UNROLL-BY-4: %c0 = constant 0 : affineint
// UNROLL-BY-4: %c4 = constant 4 : affineint
// UNROLL-BY-4: for %i0 = 1 to %arg0 {
for %i = 1 to %N {
// UNROLL-BY-4: %0 = "addi32"(%c0, %c0) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %1 = affine_apply #map0(%c0)
// UNROLL-BY-4-NEXT: %2 = "addi32"(%1, %1) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %3 = affine_apply #map1(%c0)
// UNROLL-BY-4-NEXT: %4 = "addi32"(%3, %3) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %5 = affine_apply #map2(%c0)
// UNROLL-BY-4-NEXT: %6 = "addi32"(%5, %5) : (affineint, affineint) -> i32
// UNROLL-BY-4-NEXT: %7 = "addi32"(%c4, %c4) : (affineint, affineint) -> i32
// UNROLL-BY-4-NOT: for
for %j = 0 to 4 {
%x = "addi32"(%j, %j) : (affineint, affineint) -> i32
} // UNROLL-BY-4-NOT: }
} // UNROLL-BY-4: }
return
}
Extend getConstantTripCount to deal with a larger subset of loop bounds; make loop unroll/unroll-and-jam more powerful; add additional affine expr builder methods - use previously added analysis/simplification to infer multiple of unroll factor trip counts, making loop unroll/unroll-and-jam more general. - for loop unroll, support bounds that are single result affine map's with the same set of operands. For unknown loop bounds, loop unroll will now work as long as trip count can be determined to be a multiple of unroll factor. - extend getConstantTripCount to deal with single result affine map's with the same operands. move it to mlir/Analysis/LoopAnalysis.cpp - add additional builder utility methods for affine expr arithmetic (difference, mod/floordiv/ceildiv w.r.t postitive constant). simplify code to use the utility methods. - move affine analysis routines to AffineAnalysis.cpp/.h from AffineStructures.cpp/.h. - Rename LoopUnrollJam to LoopUnrollAndJam to match class name. - add an additional simplification for simplifyFloorDiv, simplifyCeilDiv - Rename AffineMap::getNumOperands() getNumInputs: an affine map by itself does not have operands. Operands are passed to it through affine_apply, from loop bounds/if condition's, etc., operands are stored in the latter. This should be sufficiently powerful for now as far as unroll/unroll-and-jam go for TPU code generation, and can move to other analyses/transformations. Loop nests like these are now unrolled without any cleanup loop being generated. for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) { %x = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (d0 - d mod 4 - 1) (%i) { %y = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 128) (%i) { %x = "foo"() : () -> i32 } } TODO(bondhugula): extend this to LoopUnrollAndJam as well in the next CL (with minor changes). PiperOrigin-RevId: 212661212
2018-09-13 01:21:23 +08:00
// Test cases with loop bound operands.
// No cleanup will be generated here.
// UNROLL-BY-4-LABEL: mlfunc @loop_nest_operand1() {
mlfunc @loop_nest_operand1() {
// UNROLL-BY-4: for %i0 = 1 to 100 step 2 {
for %i = 1 to 100 step 2 {
// UNROLL-BY-4: %0 = "foo"() : () -> i32
// UNROLL-BY-4: %1 = "foo"() : () -> i32
// UNROLL-BY-4: %2 = "foo"() : () -> i32
// UNROLL-BY-4: %3 = "foo"() : () -> i32
for %j = (d0) -> (0) (%i) to (d0) -> (d0 - d0 mod 4 - 1) (%i) {
%x = "foo"() : () -> i32
}
}
return
}
// No cleanup will be generated here.
// UNROLL-BY-4-LABEL: mlfunc @loop_nest_operand2() {
mlfunc @loop_nest_operand2() {
// UNROLL-BY-4: for %i0 = 1 to 100 step 2 {
for %i = 1 to 100 step 2 {
// UNROLL-BY-4: %0 = "foo"() : () -> i32
// UNROLL-BY-4: %1 = "foo"() : () -> i32
// UNROLL-BY-4: %2 = "foo"() : () -> i32
// UNROLL-BY-4: %3 = "foo"() : () -> i32
for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) {
%x = "foo"() : () -> i32
}
}
return
}
// Difference between loop bounds is constant, but not a multiple of unroll
Extend loop unroll/unroll-and-jam to affine bounds + refactor related code. - extend loop unroll-jam similar to loop unroll for affine bounds - extend both loop unroll/unroll-jam to deal with cleanup loop for non multiple of unroll factor. - extend promotion of single iteration loops to work with affine bounds - fix typo bugs in loop unroll - refactor common code b/w loop unroll and loop unroll-jam - move prototypes of non-pass transforms to LoopUtils.h - add additional builder methods. - introduce loopUnrollUpTo(factor) to unroll by either factor or trip count, whichever is less. - remove Statement::isInnermost (not used for now - will come back at the right place/in right form later) PiperOrigin-RevId: 213471227
2018-09-19 01:22:03 +08:00
// factor. The cleanup loop happens to be a single iteration one and is promoted.
Extend getConstantTripCount to deal with a larger subset of loop bounds; make loop unroll/unroll-and-jam more powerful; add additional affine expr builder methods - use previously added analysis/simplification to infer multiple of unroll factor trip counts, making loop unroll/unroll-and-jam more general. - for loop unroll, support bounds that are single result affine map's with the same set of operands. For unknown loop bounds, loop unroll will now work as long as trip count can be determined to be a multiple of unroll factor. - extend getConstantTripCount to deal with single result affine map's with the same operands. move it to mlir/Analysis/LoopAnalysis.cpp - add additional builder utility methods for affine expr arithmetic (difference, mod/floordiv/ceildiv w.r.t postitive constant). simplify code to use the utility methods. - move affine analysis routines to AffineAnalysis.cpp/.h from AffineStructures.cpp/.h. - Rename LoopUnrollJam to LoopUnrollAndJam to match class name. - add an additional simplification for simplifyFloorDiv, simplifyCeilDiv - Rename AffineMap::getNumOperands() getNumInputs: an affine map by itself does not have operands. Operands are passed to it through affine_apply, from loop bounds/if condition's, etc., operands are stored in the latter. This should be sufficiently powerful for now as far as unroll/unroll-and-jam go for TPU code generation, and can move to other analyses/transformations. Loop nests like these are now unrolled without any cleanup loop being generated. for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) { %x = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (d0 - d mod 4 - 1) (%i) { %y = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 128) (%i) { %x = "foo"() : () -> i32 } } TODO(bondhugula): extend this to LoopUnrollAndJam as well in the next CL (with minor changes). PiperOrigin-RevId: 212661212
2018-09-13 01:21:23 +08:00
// UNROLL-BY-4-LABEL: mlfunc @loop_nest_operand3() {
mlfunc @loop_nest_operand3() {
// UNROLL-BY-4: for %i0 = 1 to 100 step 2 {
for %i = 1 to 100 step 2 {
[MLIR] Fix AsmPrinter for short-hand bound notation This CL retricts shorthand notation printing to only the bounds that can be roundtripped unambiguously; i.e.: 1. ()[]->(%some_cst) ()[] 2. ()[s0]->(s0) ()[%some_symbol] Upon inspection it turns out that the constant case was lossy so this CL also updates it. Note however that fixing this issue exhibits a potential issues in unroll.mlir. L488 exhibits a map ()[s0] -> (1)()[%arg0] which could be simplified down to ()[]->(1)()[]. This does not seem like a bug but maybe an undesired complexity in the maps generated by unrolling. bondhugula@, care to take a look? PiperOrigin-RevId: 214531410
2018-09-26 08:15:54 +08:00
// UNROLL-BY-4: for %i1 = (d0) -> (d0)(%i0) to #map{{[0-9]+}}(%i0) step 4 {
Extend getConstantTripCount to deal with a larger subset of loop bounds; make loop unroll/unroll-and-jam more powerful; add additional affine expr builder methods - use previously added analysis/simplification to infer multiple of unroll factor trip counts, making loop unroll/unroll-and-jam more general. - for loop unroll, support bounds that are single result affine map's with the same set of operands. For unknown loop bounds, loop unroll will now work as long as trip count can be determined to be a multiple of unroll factor. - extend getConstantTripCount to deal with single result affine map's with the same operands. move it to mlir/Analysis/LoopAnalysis.cpp - add additional builder utility methods for affine expr arithmetic (difference, mod/floordiv/ceildiv w.r.t postitive constant). simplify code to use the utility methods. - move affine analysis routines to AffineAnalysis.cpp/.h from AffineStructures.cpp/.h. - Rename LoopUnrollJam to LoopUnrollAndJam to match class name. - add an additional simplification for simplifyFloorDiv, simplifyCeilDiv - Rename AffineMap::getNumOperands() getNumInputs: an affine map by itself does not have operands. Operands are passed to it through affine_apply, from loop bounds/if condition's, etc., operands are stored in the latter. This should be sufficiently powerful for now as far as unroll/unroll-and-jam go for TPU code generation, and can move to other analyses/transformations. Loop nests like these are now unrolled without any cleanup loop being generated. for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) { %x = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (d0 - d mod 4 - 1) (%i) { %y = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 128) (%i) { %x = "foo"() : () -> i32 } } TODO(bondhugula): extend this to LoopUnrollAndJam as well in the next CL (with minor changes). PiperOrigin-RevId: 212661212
2018-09-13 01:21:23 +08:00
// UNROLL-BY-4-NEXT: %0 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: %1 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: %2 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: %3 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: }
// UNROLL-BY-4-NEXT: %4 = "foo"() : () -> i32
Extend loop unroll/unroll-and-jam to affine bounds + refactor related code. - extend loop unroll-jam similar to loop unroll for affine bounds - extend both loop unroll/unroll-jam to deal with cleanup loop for non multiple of unroll factor. - extend promotion of single iteration loops to work with affine bounds - fix typo bugs in loop unroll - refactor common code b/w loop unroll and loop unroll-jam - move prototypes of non-pass transforms to LoopUtils.h - add additional builder methods. - introduce loopUnrollUpTo(factor) to unroll by either factor or trip count, whichever is less. - remove Statement::isInnermost (not used for now - will come back at the right place/in right form later) PiperOrigin-RevId: 213471227
2018-09-19 01:22:03 +08:00
for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 8) (%i) {
Extend getConstantTripCount to deal with a larger subset of loop bounds; make loop unroll/unroll-and-jam more powerful; add additional affine expr builder methods - use previously added analysis/simplification to infer multiple of unroll factor trip counts, making loop unroll/unroll-and-jam more general. - for loop unroll, support bounds that are single result affine map's with the same set of operands. For unknown loop bounds, loop unroll will now work as long as trip count can be determined to be a multiple of unroll factor. - extend getConstantTripCount to deal with single result affine map's with the same operands. move it to mlir/Analysis/LoopAnalysis.cpp - add additional builder utility methods for affine expr arithmetic (difference, mod/floordiv/ceildiv w.r.t postitive constant). simplify code to use the utility methods. - move affine analysis routines to AffineAnalysis.cpp/.h from AffineStructures.cpp/.h. - Rename LoopUnrollJam to LoopUnrollAndJam to match class name. - add an additional simplification for simplifyFloorDiv, simplifyCeilDiv - Rename AffineMap::getNumOperands() getNumInputs: an affine map by itself does not have operands. Operands are passed to it through affine_apply, from loop bounds/if condition's, etc., operands are stored in the latter. This should be sufficiently powerful for now as far as unroll/unroll-and-jam go for TPU code generation, and can move to other analyses/transformations. Loop nests like these are now unrolled without any cleanup loop being generated. for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) { %x = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (d0 - d mod 4 - 1) (%i) { %y = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 128) (%i) { %x = "foo"() : () -> i32 } } TODO(bondhugula): extend this to LoopUnrollAndJam as well in the next CL (with minor changes). PiperOrigin-RevId: 212661212
2018-09-13 01:21:23 +08:00
%x = "foo"() : () -> i32
}
} // UNROLL-BY-4: }
return
}
Extend loop unroll/unroll-and-jam to affine bounds + refactor related code. - extend loop unroll-jam similar to loop unroll for affine bounds - extend both loop unroll/unroll-jam to deal with cleanup loop for non multiple of unroll factor. - extend promotion of single iteration loops to work with affine bounds - fix typo bugs in loop unroll - refactor common code b/w loop unroll and loop unroll-jam - move prototypes of non-pass transforms to LoopUtils.h - add additional builder methods. - introduce loopUnrollUpTo(factor) to unroll by either factor or trip count, whichever is less. - remove Statement::isInnermost (not used for now - will come back at the right place/in right form later) PiperOrigin-RevId: 213471227
2018-09-19 01:22:03 +08:00
// UNROLL-BY-4-LABEL: mlfunc @loop_nest_operand4(%arg0 : affineint) {
Extend getConstantTripCount to deal with a larger subset of loop bounds; make loop unroll/unroll-and-jam more powerful; add additional affine expr builder methods - use previously added analysis/simplification to infer multiple of unroll factor trip counts, making loop unroll/unroll-and-jam more general. - for loop unroll, support bounds that are single result affine map's with the same set of operands. For unknown loop bounds, loop unroll will now work as long as trip count can be determined to be a multiple of unroll factor. - extend getConstantTripCount to deal with single result affine map's with the same operands. move it to mlir/Analysis/LoopAnalysis.cpp - add additional builder utility methods for affine expr arithmetic (difference, mod/floordiv/ceildiv w.r.t postitive constant). simplify code to use the utility methods. - move affine analysis routines to AffineAnalysis.cpp/.h from AffineStructures.cpp/.h. - Rename LoopUnrollJam to LoopUnrollAndJam to match class name. - add an additional simplification for simplifyFloorDiv, simplifyCeilDiv - Rename AffineMap::getNumOperands() getNumInputs: an affine map by itself does not have operands. Operands are passed to it through affine_apply, from loop bounds/if condition's, etc., operands are stored in the latter. This should be sufficiently powerful for now as far as unroll/unroll-and-jam go for TPU code generation, and can move to other analyses/transformations. Loop nests like these are now unrolled without any cleanup loop being generated. for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) { %x = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (d0 - d mod 4 - 1) (%i) { %y = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 128) (%i) { %x = "foo"() : () -> i32 } } TODO(bondhugula): extend this to LoopUnrollAndJam as well in the next CL (with minor changes). PiperOrigin-RevId: 212661212
2018-09-13 01:21:23 +08:00
mlfunc @loop_nest_operand4(%N : affineint) {
Extend loop unroll/unroll-and-jam to affine bounds + refactor related code. - extend loop unroll-jam similar to loop unroll for affine bounds - extend both loop unroll/unroll-jam to deal with cleanup loop for non multiple of unroll factor. - extend promotion of single iteration loops to work with affine bounds - fix typo bugs in loop unroll - refactor common code b/w loop unroll and loop unroll-jam - move prototypes of non-pass transforms to LoopUtils.h - add additional builder methods. - introduce loopUnrollUpTo(factor) to unroll by either factor or trip count, whichever is less. - remove Statement::isInnermost (not used for now - will come back at the right place/in right form later) PiperOrigin-RevId: 213471227
2018-09-19 01:22:03 +08:00
// UNROLL-BY-4: for %i0 = 1 to 100 {
for %i = 1 to 100 {
[MLIR] Fix AsmPrinter for short-hand bound notation This CL retricts shorthand notation printing to only the bounds that can be roundtripped unambiguously; i.e.: 1. ()[]->(%some_cst) ()[] 2. ()[s0]->(s0) ()[%some_symbol] Upon inspection it turns out that the constant case was lossy so this CL also updates it. Note however that fixing this issue exhibits a potential issues in unroll.mlir. L488 exhibits a map ()[s0] -> (1)()[%arg0] which could be simplified down to ()[]->(1)()[]. This does not seem like a bug but maybe an undesired complexity in the maps generated by unrolling. bondhugula@, care to take a look? PiperOrigin-RevId: 214531410
2018-09-26 08:15:54 +08:00
// UNROLL-BY-4: for %i1 = ()[s0] -> (1)()[%arg0] to #map{{[0-9]+}}()[%arg0] step 4 {
Extend loop unroll/unroll-and-jam to affine bounds + refactor related code. - extend loop unroll-jam similar to loop unroll for affine bounds - extend both loop unroll/unroll-jam to deal with cleanup loop for non multiple of unroll factor. - extend promotion of single iteration loops to work with affine bounds - fix typo bugs in loop unroll - refactor common code b/w loop unroll and loop unroll-jam - move prototypes of non-pass transforms to LoopUtils.h - add additional builder methods. - introduce loopUnrollUpTo(factor) to unroll by either factor or trip count, whichever is less. - remove Statement::isInnermost (not used for now - will come back at the right place/in right form later) PiperOrigin-RevId: 213471227
2018-09-19 01:22:03 +08:00
// UNROLL-BY-4: %0 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: %1 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: %2 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: %3 = "foo"() : () -> i32
// UNROLL-BY-4-NEXT: }
// A cleanup loop will be be generated here.
// UNROLL-BY-4-NEXT: for %i2 = #map{{[0-9]+}}()[%arg0] to %arg0 {
// UNROLL-BY-4-NEXT: %4 = "foo"() : () -> i32
// UNROLL-BY-4_NEXT: }
// Specify the lower bound so that both lb and ub operands match.
for %j = ()[s0] -> (1)()[%N] to %N {
Extend getConstantTripCount to deal with a larger subset of loop bounds; make loop unroll/unroll-and-jam more powerful; add additional affine expr builder methods - use previously added analysis/simplification to infer multiple of unroll factor trip counts, making loop unroll/unroll-and-jam more general. - for loop unroll, support bounds that are single result affine map's with the same set of operands. For unknown loop bounds, loop unroll will now work as long as trip count can be determined to be a multiple of unroll factor. - extend getConstantTripCount to deal with single result affine map's with the same operands. move it to mlir/Analysis/LoopAnalysis.cpp - add additional builder utility methods for affine expr arithmetic (difference, mod/floordiv/ceildiv w.r.t postitive constant). simplify code to use the utility methods. - move affine analysis routines to AffineAnalysis.cpp/.h from AffineStructures.cpp/.h. - Rename LoopUnrollJam to LoopUnrollAndJam to match class name. - add an additional simplification for simplifyFloorDiv, simplifyCeilDiv - Rename AffineMap::getNumOperands() getNumInputs: an affine map by itself does not have operands. Operands are passed to it through affine_apply, from loop bounds/if condition's, etc., operands are stored in the latter. This should be sufficiently powerful for now as far as unroll/unroll-and-jam go for TPU code generation, and can move to other analyses/transformations. Loop nests like these are now unrolled without any cleanup loop being generated. for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (5*d0 + 3) (%i) { %x = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { // unroll factor 4: no cleanup loop will be generated. for %j = (d0) -> (d0) (%i) to (d0) -> (d0 - d mod 4 - 1) (%i) { %y = "foo"(%j) : (affineint) -> i32 } } for %i = 1 to 100 { for %j = (d0) -> (d0) (%i) to (d0) -> (d0 + 128) (%i) { %x = "foo"() : () -> i32 } } TODO(bondhugula): extend this to LoopUnrollAndJam as well in the next CL (with minor changes). PiperOrigin-RevId: 212661212
2018-09-13 01:21:23 +08:00
%x = "foo"() : () -> i32
}
}
return
}