llvm-project/polly/test/ScopInfo/multidim_only_ivs_3d_cast.ll

90 lines
2.7 KiB
LLVM
Raw Normal View History

; RUN: opt %loadPolly -polly-scops -analyze < %s | FileCheck %s
; void foo(int n, int m, int o, double A[n][m][o]) {
;
; for (int i = 0; i < n; i++)
; for (int j = 0; j < m; j++)
; for (int k = 0; k < o; k++)
; A[i][j][k] = 1.0;
; }
; CHECK: Assumed Context:
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
; CHECK-NEXT: [o, m, n] -> { : }
; CHECK-NEXT: Invalid Context:
; CHECK-NEXT: [o, m, n] -> { : o < 0 or m < 0 or n <= 0 or (m <= 0 and n > 0) or (o <= 0 and m > 0 and n > 0) }
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
;
; CHECK: p0: %o
; CHECK-NEXT: p1: %m
; CHECK-NEXT: p2: %n
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
; CHECK-NOT: p3
;
; CHECK: Arrays {
; CHECK-NEXT: double MemRef_A[*][(zext i32 %m to i64)][(zext i32 %o to i64)]; // Element size 8
; CHECK-NEXT: }
;
; CHECK: Arrays (Bounds as pw_affs) {
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
; CHECK-NEXT: double MemRef_A[*][ [m] -> { [] -> [(m)] } ][ [o] -> { [] -> [(o)] } ]; // Element size 8
; CHECK-NEXT: }
;
; CHECK: Statements {
; CHECK-NEXT: Stmt_for_k
; CHECK-NEXT: Domain :=
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
; CHECK-NEXT: [o, m, n] -> { Stmt_for_k[i0, i1, i2] : 0 <= i0 < n and 0 <= i1 < m and 0 <= i2 < o };
; CHECK-NEXT: Schedule :=
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
; CHECK-NEXT: [o, m, n] -> { Stmt_for_k[i0, i1, i2] -> [i0, i1, i2] };
; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
Model zext-extend instructions A zero-extended value can be interpreted as a piecewise defined signed value. If the value was non-negative it stays the same, otherwise it is the sum of the original value and 2^n where n is the bit-width of the original (or operand) type. Examples: zext i8 127 to i32 -> { [127] } zext i8 -1 to i32 -> { [256 + (-1)] } = { [255] } zext i8 %v to i32 -> [v] -> { [v] | v >= 0; [256 + v] | v < 0 } However, LLVM/Scalar Evolution uses zero-extend (potentially lead by a truncate) to represent some forms of modulo computation. The left-hand side of the condition in the code below would result in the SCEV "zext i1 <false, +, true>for.body" which is just another description of the C expression "i & 1 != 0" or, equivalently, "i % 2 != 0". for (i = 0; i < N; i++) if (i & 1 != 0 /* == i % 2 */) /* do something */ If we do not make the modulo explicit but only use the mechanism described above we will get the very restrictive assumption "N < 3", because for all values of N >= 3 the SCEVAddRecExpr operand of the zero-extend would wrap. Alternatively, we can make the modulo in the operand explicit in the resulting piecewise function and thereby avoid the assumption on N. For the example this would result in the following piecewise affine function: { [i0] -> [(1)] : 2*floor((-1 + i0)/2) = -1 + i0; [i0] -> [(0)] : 2*floor((i0)/2) = i0 } To this end we can first determine if the (immediate) operand of the zero-extend can wrap and, in case it might, we will use explicit modulo semantic to compute the result instead of emitting non-wrapping assumptions. Note that operands with large bit-widths are less likely to be negative because it would result in a very large access offset or loop bound after the zero-extend. To this end one can optimistically assume the operand to be positive and avoid the piecewise definition if the bit-width is bigger than some threshold (here MaxZextSmallBitWidth). We choose to go with a hybrid solution of all modeling techniques described above. For small bit-widths (up to MaxZextSmallBitWidth) we will model the wrapping explicitly and use a piecewise defined function. However, if the bit-width is bigger than MaxZextSmallBitWidth we will employ overflow assumptions and assume the "former negative" piece will not exist. llvm-svn: 267408
2016-04-25 22:01:36 +08:00
; CHECK-NEXT: [o, m, n] -> { Stmt_for_k[i0, i1, i2] -> MemRef_A[i0, i1, i2] };
; CHECK-NEXT: }
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
define void @foo(i32 %n, i32 %m, i32 %o, double* %A) {
entry:
%m_zext = zext i32 %m to i64
%n_zext = zext i32 %o to i64
br label %for.i
for.i:
%i = phi i64 [ %i.inc, %for.i.inc ], [ 0, %entry ]
br label %for.j
for.j:
%j = phi i64 [ %j.inc, %for.j.inc ], [ 0, %for.i ]
br label %for.k
for.k:
%k = phi i64 [ %k.inc, %for.k.inc ], [ 0, %for.j ]
%tmp = mul i64 %i, %m_zext
%tmp1 = trunc i64 %j to i32
%tmp2 = trunc i64 %i to i32
%mul.us.us = mul nsw i32 %tmp1, %tmp2
%tmp.us.us = add i64 %j, %tmp
%tmp17.us.us = mul i64 %tmp.us.us, %n_zext
%subscript = add i64 %tmp17.us.us, %k
%idx = getelementptr inbounds double, double* %A, i64 %subscript
store double 1.0, double* %idx
br label %for.k.inc
for.k.inc:
%k.inc = add i64 %k, 1
%k.inc.trunc = trunc i64 %k.inc to i32
%k.exitcond = icmp eq i32 %k.inc.trunc, %o
br i1 %k.exitcond, label %for.j.inc, label %for.k
for.j.inc:
%j.inc = add i64 %j, 1
%j.inc.trunc = trunc i64 %j.inc to i32
%j.exitcond = icmp eq i32 %j.inc.trunc, %m
br i1 %j.exitcond, label %for.i.inc, label %for.j
for.i.inc:
%i.inc = add i64 %i, 1
%i.inc.trunc = trunc i64 %i.inc to i32
%i.exitcond = icmp eq i32 %i.inc.trunc, %n
br i1 %i.exitcond, label %end, label %for.i
end:
ret void
}