This patch provides for interpreting element numbers for the Altivec
vec_insert and vec_extract intrinsics as big-endian (left to right in a
vector register) when targeting a little endian machine and specifying
-maltivec=be. New test cases are added to test this functionality on
all supported vector types.
Bootstrapped and tested with no new regressions on
powerpc64{,le}-unknown-linux-gnu. Ok for trunk?
Thanks,
Bill
gcc:
2014-01-12 Bill Schmidt <wschm...@linux.vnet.ibm.com>
* config/rs6000/rs6000-c.c (altivec_resolve_overloaded_builtin):
Implement -maltivec=be for vec_insert and vec_extract.
gcc/testsuite:
2014-01-12 Bill Schmidt <wschm...@linux.vnet.ibm.com>
* gcc.dg/vmx/insert.c: New.
* gcc.dg/vmx/insert-be-order.c: New.
* gcc.dg/vmx/extract.c: New.
* gcc.dg/vmx/extract-be-order.c: New.
Index: gcc/testsuite/gcc.dg/vmx/insert-be-order.c
===================================================================
--- gcc/testsuite/gcc.dg/vmx/insert-be-order.c (revision 0)
+++ gcc/testsuite/gcc.dg/vmx/insert-be-order.c (revision 0)
@@ -0,0 +1,65 @@
+/* { dg-options "-maltivec=be -mabi=altivec -std=gnu99 -mno-vsx" } */
+
+#include "harness.h"
+
+static void test()
+{
+ vector unsigned char va = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
+ vector signed char vb = {-8,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7};
+ vector unsigned short vc = {0,1,2,3,4,5,6,7};
+ vector signed short vd = {-4,-3,-2,-1,0,1,2,3};
+ vector unsigned int ve = {0,1,2,3};
+ vector signed int vf = {-2,-1,0,1};
+ vector float vg = {-2.0f,-1.0f,0.0f,1.0f};
+
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+ check (vec_all_eq (vec_insert (16, va, 5),
+ ((vector unsigned char)
+ {0,1,2,3,4,5,6,7,8,9,16,11,12,13,14,15})),
+ "vec_insert (va LE)");
+ check (vec_all_eq (vec_insert (-16, vb, 0),
+ ((vector signed char)
+ {-8,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,-16})),
+ "vec_insert (vb LE)");
+ check (vec_all_eq (vec_insert (16, vc, 7),
+ ((vector unsigned short){16,1,2,3,4,5,6,7})),
+ "vec_insert (vc LE)");
+ check (vec_all_eq (vec_insert (-16, vd, 3),
+ ((vector signed short){-4,-3,-2,-1,-16,1,2,3})),
+ "vec_insert (vd LE)");
+ check (vec_all_eq (vec_insert (16, ve, 2),
+ ((vector unsigned int){0,16,2,3})),
+ "vec_insert (ve LE)");
+ check (vec_all_eq (vec_insert (-16, vf, 1),
+ ((vector signed int){-2,-1,-16,1})),
+ "vec_insert (vf LE)");
+ check (vec_all_eq (vec_insert (-16.0f, vg, 0),
+ ((vector float){-2.0f,-1.0f,0.0f,-16.0f})),
+ "vec_insert (vg LE)");
+#else
+ check (vec_all_eq (vec_insert (16, va, 5),
+ ((vector unsigned char)
+ {0,1,2,3,4,16,6,7,8,9,10,11,12,13,14,15})),
+ "vec_insert (va BE)");
+ check (vec_all_eq (vec_insert (-16, vb, 0),
+ ((vector signed char)
+ {-16,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7})),
+ "vec_insert (vb BE)");
+ check (vec_all_eq (vec_insert (16, vc, 7),
+ ((vector unsigned short){0,1,2,3,4,5,6,16})),
+ "vec_insert (vc BE)");
+ check (vec_all_eq (vec_insert (-16, vd, 3),
+ ((vector signed short){-4,-3,-2,-16,0,1,2,3})),
+ "vec_insert (vd BE)");
+ check (vec_all_eq (vec_insert (16, ve, 2),
+ ((vector unsigned int){0,1,16,3})),
+ "vec_insert (ve BE)");
+ check (vec_all_eq (vec_insert (-16, vf, 1),
+ ((vector signed int){-2,-16,0,1})),
+ "vec_insert (vf BE)");
+ check (vec_all_eq (vec_insert (-16.0f, vg, 0),
+ ((vector float){-16.0f,-1.0f,0.0f,1.0f})),
+ "vec_insert (vg BE)");
+#endif
+}
+
Index: gcc/testsuite/gcc.dg/vmx/extract-be-order.c
===================================================================
--- gcc/testsuite/gcc.dg/vmx/extract-be-order.c (revision 0)
+++ gcc/testsuite/gcc.dg/vmx/extract-be-order.c (revision 0)
@@ -0,0 +1,33 @@
+/* { dg-options "-maltivec=be -mabi=altivec -std=gnu99 -mno-vsx" } */
+
+#include "harness.h"
+
+static void test()
+{
+ vector unsigned char va = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
+ vector signed char vb = {-8,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7};
+ vector unsigned short vc = {0,1,2,3,4,5,6,7};
+ vector signed short vd = {-4,-3,-2,-1,0,1,2,3};
+ vector unsigned int ve = {0,1,2,3};
+ vector signed int vf = {-2,-1,0,1};
+ vector float vg = {-2.0f,-1.0f,0.0f,1.0f};
+
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+ check (vec_extract (va, 5) == 10, "vec_extract (va, 5)");
+ check (vec_extract (vb, 0) == 7, "vec_extract (vb, 0)");
+ check (vec_extract (vc, 7) == 0, "vec_extract (vc, 7)");
+ check (vec_extract (vd, 3) == 0, "vec_extract (vd, 3)");
+ check (vec_extract (ve, 2) == 1, "vec_extract (ve, 2)");
+ check (vec_extract (vf, 1) == 0, "vec_extract (vf, 1)");
+ check (vec_extract (vg, 0) == 1.0f, "vec_extract (vg, 0)");
+#else
+ check (vec_extract (va, 5) == 5, "vec_extract (va, 5)");
+ check (vec_extract (vb, 0) == -8, "vec_extract (vb, 0)");
+ check (vec_extract (vc, 7) == 7, "vec_extract (vc, 7)");
+ check (vec_extract (vd, 3) == -1, "vec_extract (vd, 3)");
+ check (vec_extract (ve, 2) == 2, "vec_extract (ve, 2)");
+ check (vec_extract (vf, 1) == -1, "vec_extract (vf, 1)");
+ check (vec_extract (vg, 0) == -2.0f, "vec_extract (vg, 0)");
+#endif
+}
+
Index: gcc/testsuite/gcc.dg/vmx/extract.c
===================================================================
--- gcc/testsuite/gcc.dg/vmx/extract.c (revision 0)
+++ gcc/testsuite/gcc.dg/vmx/extract.c (revision 0)
@@ -0,0 +1,21 @@
+#include "harness.h"
+
+static void test()
+{
+ vector unsigned char va = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
+ vector signed char vb = {-8,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7};
+ vector unsigned short vc = {0,1,2,3,4,5,6,7};
+ vector signed short vd = {-4,-3,-2,-1,0,1,2,3};
+ vector unsigned int ve = {0,1,2,3};
+ vector signed int vf = {-2,-1,0,1};
+ vector float vg = {-2.0f,-1.0f,0.0f,1.0f};
+
+ check (vec_extract (va, 5) == 5, "vec_extract (va, 5)");
+ check (vec_extract (vb, 0) == -8, "vec_extract (vb, 0)");
+ check (vec_extract (vc, 7) == 7, "vec_extract (vc, 7)");
+ check (vec_extract (vd, 3) == -1, "vec_extract (vd, 3)");
+ check (vec_extract (ve, 2) == 2, "vec_extract (ve, 2)");
+ check (vec_extract (vf, 1) == -1, "vec_extract (vf, 1)");
+ check (vec_extract (vg, 0) == -2.0f, "vec_extract (vg, 0)");
+}
+
Index: gcc/testsuite/gcc.dg/vmx/insert.c
===================================================================
--- gcc/testsuite/gcc.dg/vmx/insert.c (revision 0)
+++ gcc/testsuite/gcc.dg/vmx/insert.c (revision 0)
@@ -0,0 +1,37 @@
+#include "harness.h"
+
+static void test()
+{
+ vector unsigned char va = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15};
+ vector signed char vb = {-8,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7};
+ vector unsigned short vc = {0,1,2,3,4,5,6,7};
+ vector signed short vd = {-4,-3,-2,-1,0,1,2,3};
+ vector unsigned int ve = {0,1,2,3};
+ vector signed int vf = {-2,-1,0,1};
+ vector float vg = {-2.0f,-1.0f,0.0f,1.0f};
+
+ check (vec_all_eq (vec_insert (16, va, 5),
+ ((vector unsigned char)
+ {0,1,2,3,4,16,6,7,8,9,10,11,12,13,14,15})),
+ "vec_insert (va)");
+ check (vec_all_eq (vec_insert (-16, vb, 0),
+ ((vector signed char)
+ {-16,-7,-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6,7})),
+ "vec_insert (vb)");
+ check (vec_all_eq (vec_insert (16, vc, 7),
+ ((vector unsigned short){0,1,2,3,4,5,6,16})),
+ "vec_insert (vc)");
+ check (vec_all_eq (vec_insert (-16, vd, 3),
+ ((vector signed short){-4,-3,-2,-16,0,1,2,3})),
+ "vec_insert (vd)");
+ check (vec_all_eq (vec_insert (16, ve, 2),
+ ((vector unsigned int){0,1,16,3})),
+ "vec_insert (ve)");
+ check (vec_all_eq (vec_insert (-16, vf, 1),
+ ((vector signed int){-2,-16,0,1})),
+ "vec_insert (vf)");
+ check (vec_all_eq (vec_insert (-16.0f, vg, 0),
+ ((vector float){-16.0f,-1.0f,0.0f,1.0f})),
+ "vec_insert (vg)");
+}
+
Index: gcc/config/rs6000/rs6000-c.c
===================================================================
--- gcc/config/rs6000/rs6000-c.c (revision 206571)
+++ gcc/config/rs6000/rs6000-c.c (working copy)
@@ -4172,7 +4172,7 @@ altivec_resolve_overloaded_builtin (location_t loc
return build_constructor (type, vec);
}
- /* For now use pointer tricks to do the extaction, unless we are on VSX
+ /* For now use pointer tricks to do the extraction, unless we are on VSX
extracting a double from a constant offset. */
if (fcode == ALTIVEC_BUILTIN_VEC_EXTRACT)
{
@@ -4200,6 +4200,19 @@ altivec_resolve_overloaded_builtin (location_t loc
if (!INTEGRAL_TYPE_P (TREE_TYPE (arg2)))
goto bad;
+ /* If we are targeting little-endian, but -maltivec=be has been
+ specified to override the element order, adjust the element
+ number accordingly. */
+ if (!BYTES_BIG_ENDIAN && rs6000_altivec_element_order == 2)
+ {
+ int last_elem = TYPE_VECTOR_SUBPARTS (arg1_type) - 1;
+ double_int di_last_elem = double_int::from_uhwi (last_elem);
+ arg2 = fold_build2_loc (loc, MINUS_EXPR, TREE_TYPE (arg2),
+ double_int_to_tree (TREE_TYPE (arg2),
+ di_last_elem),
+ arg2);
+ }
+
/* If we can use the VSX xxpermdi instruction, use that for extract. */
mode = TYPE_MODE (arg1_type);
if ((mode == V2DFmode || mode == V2DImode) && VECTOR_MEM_VSX_P (mode)
@@ -4256,7 +4269,7 @@ altivec_resolve_overloaded_builtin (location_t loc
return stmt;
}
- /* For now use pointer tricks to do the insertation, unless we are on VSX
+ /* For now use pointer tricks to do the insertion, unless we are on VSX
inserting a double to a constant offset.. */
if (fcode == ALTIVEC_BUILTIN_VEC_INSERT)
{
@@ -4286,6 +4299,19 @@ altivec_resolve_overloaded_builtin (location_t loc
if (!INTEGRAL_TYPE_P (TREE_TYPE (arg2)))
goto bad;
+ /* If we are targeting little-endian, but -maltivec=be has been
+ specified to override the element order, adjust the element
+ number accordingly. */
+ if (!BYTES_BIG_ENDIAN && rs6000_altivec_element_order == 2)
+ {
+ int last_elem = TYPE_VECTOR_SUBPARTS (arg1_type) - 1;
+ double_int di_last_elem = double_int::from_uhwi (last_elem);
+ arg2 = fold_build2_loc (loc, MINUS_EXPR, TREE_TYPE (arg2),
+ double_int_to_tree (TREE_TYPE (arg2),
+ di_last_elem),
+ arg2);
+ }
+
/* If we can use the VSX xxpermdi instruction, use that for insert. */
mode = TYPE_MODE (arg1_type);
if ((mode == V2DFmode || mode == V2DImode) && VECTOR_UNIT_VSX_P (mode)
