aaron/ispc
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Merge pull request #641 from jbrodman/stdlibshift

Browse Source 
Add a "shift" operator to the stdlib.
This commit is contained in:
Dmitry Babokin
2013-10-28 14:18:31 -07:00
parent ef5d2dc043 e682b19eda
commit 6585a925be
12 changed files with 425 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
builtins.cpp
View File

@@ -536,6 +536,12 @@ lSetInternalFunctions(llvm::Module *module) {
"__set_system_isa",
"__sext_uniform_bool",
"__sext_varying_bool",
"__shift_double",
"__shift_float",
"__shift_i16",
"__shift_i32",
"__shift_i64",
"__shift_i8",
"__shuffle2_double",
"__shuffle2_float",
"__shuffle2_i16",

7
builtins/target-generic-common.ll
View File

@@ -80,6 +80,13 @@ declare <WIDTH x i32> @__rotate_i32(<WIDTH x i32>, i32) nounwind readnone
declare <WIDTH x double> @__rotate_double(<WIDTH x double>, i32) nounwind readnone
declare <WIDTH x i64> @__rotate_i64(<WIDTH x i64>, i32) nounwind readnone
declare <WIDTH x i8> @__shift_i8(<WIDTH x i8>, i32) nounwind readnone
declare <WIDTH x i16> @__shift_i16(<WIDTH x i16>, i32) nounwind readnone
declare <WIDTH x float> @__shift_float(<WIDTH x float>, i32) nounwind readnone
declare <WIDTH x i32> @__shift_i32(<WIDTH x i32>, i32) nounwind readnone
declare <WIDTH x double> @__shift_double(<WIDTH x double>, i32) nounwind readnone
declare <WIDTH x i64> @__shift_i64(<WIDTH x i64>, i32) nounwind readnone
declare <WIDTH x i8> @__shuffle_i8(<WIDTH x i8>, <WIDTH x i32>) nounwind readnone
declare <WIDTH x i8> @__shuffle2_i8(<WIDTH x i8>, <WIDTH x i8>,
<WIDTH x i32>) nounwind readnone

18
builtins/util.m4
View File

@@ -797,6 +797,24 @@ not_const:
ret <WIDTH x $1> %result
}
define <WIDTH x $1> @__shift_$1(<WIDTH x $1>, i32) nounwind readnone alwaysinline {
%ptr = alloca <WIDTH x $1>, i32 3
%ptr0 = getelementptr <WIDTH x $1> * %ptr, i32 0
store <WIDTH x $1> zeroinitializer, <WIDTH x $1> * %ptr0
%ptr1 = getelementptr <WIDTH x $1> * %ptr, i32 1
store <WIDTH x $1> %0, <WIDTH x $1> * %ptr1
%ptr2 = getelementptr <WIDTH x $1> * %ptr, i32 2
store <WIDTH x $1> zeroinitializer, <WIDTH x $1> * %ptr2
%offset = add i32 %1, WIDTH
%ptr_as_elt_array = bitcast <WIDTH x $1> * %ptr to [eval(3*WIDTH) x $1] *
%load_ptr = getelementptr [eval(3*WIDTH) x $1] * %ptr_as_elt_array, i32 0, i32 %offset
%load_ptr_vec = bitcast $1 * %load_ptr to <WIDTH x $1> *
%result = load <WIDTH x $1> * %load_ptr_vec, align $2
ret <WIDTH x $1> %result
}
define <WIDTH x $1> @__shuffle_$1(<WIDTH x $1>, <WIDTH x i32>) nounwind readnone alwaysinline {
forloop(i, 0, eval(WIDTH-1), `
%index_`'i = extractelement <WIDTH x i32> %1, i32 i')

18
docs/ispc.rst
View File

@@ -3719,6 +3719,22 @@ the size of the gang (it is masked to ensure valid offsets).
double rotate(double value, uniform int offset)
The ``shift()`` function allows each program instance to find the value of
the given value that their neighbor ``offset`` steps away has. This is similar
to ``rotate()`` with the exception that values are not circularly shifted.
Instead, zeroes are shifted in where appropriate.
::
int8 shift(int8 value, uniform int offset)
int16 shift(int16 value, uniform int offset)
int32 shift(int32 value, uniform int offset)
int64 shift(int64 value, uniform int offset)
float shift(float value, uniform int offset)
double shift(double value, uniform int offset)
Finally, the ``shuffle()`` functions allow two variants of fully general
shuffling of values among the program instances. For the first version,
each program instance's value of permutation gives the program instance
@@ -3751,7 +3767,7 @@ the last element of ``value1``, etc.)
double shuffle(double value0, double value1, int permutation)
Finally, there are primitive operations that extract and set values in the
SIMD lanes. You can implement all of the broadcast, rotate, and shuffle
SIMD lanes. You can implement all of the broadcast, rotate, shift, and shuffle
operations described above in this section from these routines, though in
general, not as efficiently. These routines are useful for implementing
other reductions and cross-lane communication that isn't included in the

17
examples/intrinsics/generic-16.h
View File

@@ -311,6 +311,17 @@ static FORCEINLINE VTYPE __rotate_##NAME(VTYPE v, int index) { \
return ret; \
} \
#define SHIFT(VTYPE, NAME, STYPE) \
static FORCEINLINE VTYPE __shift_##NAME(VTYPE v, int index) { \
VTYPE ret; \
for (int i = 0; i < 16; ++i) { \
int modIndex = i+index; \
STYPE val = ((modIndex >= 0) && (modIndex < 16)) ? v.v[modIndex] : 0; \
ret.v[i] = val; \
} \
return ret; \
} \
#define SHUFFLES(VTYPE, NAME, STYPE) \
static FORCEINLINE VTYPE __shuffle_##NAME(VTYPE v, __vec16_i32 index) { \
VTYPE ret; \
@@ -492,6 +503,7 @@ SETZERO(__vec16_i8, i8)
UNDEF(__vec16_i8, i8)
BROADCAST(__vec16_i8, i8, int8_t)
ROTATE(__vec16_i8, i8, int8_t)
SHIFT(__vec16_i8, i8, int8_t)
SHUFFLES(__vec16_i8, i8, int8_t)
LOAD_STORE(__vec16_i8, int8_t)
@@ -537,6 +549,7 @@ SETZERO(__vec16_i16, i16)
UNDEF(__vec16_i16, i16)
BROADCAST(__vec16_i16, i16, int16_t)
ROTATE(__vec16_i16, i16, int16_t)
SHIFT(__vec16_i16, i16, int16_t)
SHUFFLES(__vec16_i16, i16, int16_t)
LOAD_STORE(__vec16_i16, int16_t)
@@ -582,6 +595,7 @@ SETZERO(__vec16_i32, i32)
UNDEF(__vec16_i32, i32)
BROADCAST(__vec16_i32, i32, int32_t)
ROTATE(__vec16_i32, i32, int32_t)
SHIFT(__vec16_i32, i32, int32_t)
SHUFFLES(__vec16_i32, i32, int32_t)
LOAD_STORE(__vec16_i32, int32_t)
@@ -627,6 +641,7 @@ SETZERO(__vec16_i64, i64)
UNDEF(__vec16_i64, i64)
BROADCAST(__vec16_i64, i64, int64_t)
ROTATE(__vec16_i64, i64, int64_t)
SHIFT(__vec16_i64, i64, int64_t)
SHUFFLES(__vec16_i64, i64, int64_t)
LOAD_STORE(__vec16_i64, int64_t)
@@ -672,6 +687,7 @@ SETZERO(__vec16_f, float)
UNDEF(__vec16_f, float)
BROADCAST(__vec16_f, float, float)
ROTATE(__vec16_f, float, float)
SHIFT(__vec16_f, float, float)
SHUFFLES(__vec16_f, float, float)
LOAD_STORE(__vec16_f, float)
@@ -832,6 +848,7 @@ SETZERO(__vec16_d, double)
UNDEF(__vec16_d, double)
BROADCAST(__vec16_d, double, double)
ROTATE(__vec16_d, double, double)
SHIFT(__vec16_d, double, double)
SHUFFLES(__vec16_d, double, double)
LOAD_STORE(__vec16_d, double)

19
examples/intrinsics/knc-i1x16.h
View File

@@ -451,6 +451,17 @@ static FORCEINLINE VTYPE __rotate_##NAME(VTYPE v, int index) { \
return ret; \
} \
#define SHIFT(VTYPE, NAME, STYPE) \
static FORCEINLINE VTYPE __shift_##NAME(VTYPE v, int index) { \
VTYPE ret; \
for (int i = 0; i < 16; ++i) { \
int modIndex = i+index; \
STYPE val = ((modIndex >= 0) && (modIndex < 16)) ? v[modIndex] : 0; \
ret[i] = val; \
} \
return ret; \
} \
/* knc::macro::used */
#define SHUFFLES(VTYPE, NAME, STYPE) \
static FORCEINLINE VTYPE __shuffle_##NAME(VTYPE v, __vec16_i32 index) { \
@@ -566,6 +577,7 @@ SETZERO(__vec16_i8, i8)
UNDEF(__vec16_i8, i8)
BROADCAST(__vec16_i8, i8, int8_t)
ROTATE(__vec16_i8, i8, int8_t)
SHIFT(__vec16_i8, i8, int8_t)
SHUFFLES(__vec16_i8, i8, int8_t)
LOAD_STORE(__vec16_i8, int8_t)
@@ -612,6 +624,7 @@ SETZERO(__vec16_i16, i16)
UNDEF(__vec16_i16, i16)
BROADCAST(__vec16_i16, i16, int16_t)
ROTATE(__vec16_i16, i16, int16_t)
SHIFT(__vec16_i16, i16, int16_t)
SHUFFLES(__vec16_i16, i16, int16_t)
LOAD_STORE(__vec16_i16, int16_t)
@@ -688,6 +701,8 @@ static FORCEINLINE __vec16_i32 __rotate_i32(__vec16_i32 v, int index)
return _mm512_mask_permutevar_epi32(v, 0xFFFF, shuffle, v);
}
SHIFT(__vec16_i32, i32, int32_t)
static FORCEINLINE __vec16_i32 __shuffle_i32 (__vec16_i32 v, __vec16_i32 index)
{
return _mm512_mask_permutevar_epi32(v, 0xFFFF, __and(index, __smear_i32<__vec16_i32>(0xF)), v);
@@ -942,6 +957,8 @@ static FORCEINLINE __vec16_i64 __rotate_i64(const __vec16_i64 _v, const int inde
const __vec16_i32 ret_lo = __rotate_i32(v_lo, index);
return CASTI2L(ret_hi, ret_lo);
}
SHIFT(__vec16_i64, i64, int64_t)
static FORCEINLINE __vec16_i64 __shuffle_double(__vec16_i64 _v, const __vec16_i32 index)
{
CASTL2I(_v, v_hi, v_lo);
@@ -1063,6 +1080,7 @@ static FORCEINLINE __vec16_f __rotate_float(__vec16_f _v, int index)
const __vec16_i32 shuffle = _mm512_and_epi32(_mm512_add_epi32(__ispc_stride1, idx), __smear_i32<__vec16_i32>(0xF));
return _mm512_castsi512_ps(_mm512_mask_permutevar_epi32(v, 0xFFFF, shuffle, v));
}
SHIFT(__vec16_f, float, float)
static FORCEINLINE __vec16_f __shuffle_float(__vec16_f v, __vec16_i32 index)
{
return _mm512_castsi512_ps(_mm512_mask_permutevar_epi32(_mm512_castps_si512(v), 0xffff, index, _mm512_castps_si512(v)));
@@ -1333,6 +1351,7 @@ static FORCEINLINE __vec16_d __rotate_double(const __vec16_d _v, const int index
const __vec16_f ret_lo = __rotate_float(v_lo, index);
return CASTF2D(ret_hi, ret_lo);
}
SHIFT(__vec16_d, double, double)
static FORCEINLINE __vec16_d __shuffle_double(__vec16_d _v, const __vec16_i32 index)
{
CASTD2F(_v, v_hi, v_lo);

205
examples/intrinsics/sse4.h
View File

@@ -108,22 +108,21 @@ struct __vec4_i64 {
};
struct __vec4_i32 {
__vec4_i32() { }
FORCEINLINE __vec4_i32() { }
FORCEINLINE __vec4_i32(__m128i vv) : v(vv) { }
FORCEINLINE __vec4_i32(uint32_t a, uint32_t b, uint32_t c, uint32_t d) {
FORCEINLINE __vec4_i32(int32_t a, int32_t b, int32_t c, int32_t d) {
v = _mm_set_epi32(d, c, b, a);
}
FORCEINLINE __vec4_i32(uint32_t *p) {
FORCEINLINE __vec4_i32(int32_t *p) {
v = _mm_loadu_si128((__m128i *)p);
}
FORCEINLINE __vec4_i32(const __vec4_i32 &other) : v(other.v) {}
FORCEINLINE __vec4_i32& operator =(const __vec4_i32 &o) { v=o.v; return *this; }
FORCEINLINE operator __m128() const { return _mm_castsi128_ps(v); }
__m128i v;
};
static inline int32_t __extract_element(__vec4_i32 v, int index);
struct __vec4_i16 {
__vec4_i16() { }
FORCEINLINE __vec4_i16(__m128i vv) : v(vv) { }
@@ -215,6 +214,64 @@ INSERT_EXTRACT(__vec1_i64, int64_t)
INSERT_EXTRACT(__vec1_f, float)
INSERT_EXTRACT(__vec1_d, double)
static FORCEINLINE bool __extract_element(const __vec4_i1 &v, int index) {
return ((int32_t *)&v)[index] ? true : false;
}
static FORCEINLINE void __insert_element(__vec4_i1 *v, int index, bool val) {
((int32_t *)v)[index] = val ? -1 : 0;
}
static FORCEINLINE int8_t __extract_element(const __vec4_i8 &v, int index) {
return ((int8_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i8 *v, int index, int8_t val) {
((int8_t *)v)[index] = val;
}
static FORCEINLINE int16_t __extract_element(const __vec4_i16 &v, int index) {
return ((int16_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i16 *v, int index, int16_t val) {
((int16_t *)v)[index] = val;
}
static FORCEINLINE int32_t __extract_element(const __vec4_i32 &v, int index) {
return ((int32_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i32 *v, int index, int32_t val) {
((int32_t *)v)[index] = val;
}
static FORCEINLINE int64_t __extract_element(const __vec4_i64 &v, int index) {
return ((int64_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i64 *v, int index, int64_t val) {
((int64_t *)v)[index] = val;
}
static FORCEINLINE float __extract_element(const __vec4_f &v, int index) {
return ((float *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_f *v, int index, float val) {
((float *)v)[index] = val;
}
static FORCEINLINE double __extract_element(const __vec4_d &v, int index) {
return ((double *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_d *v, int index, double val) {
((double *)v)[index] = val;
}
#define CAST_BITS_SCALAR(TO, FROM) \
static FORCEINLINE TO __cast_bits(TO, FROM v) { \
union { \
@@ -313,13 +370,6 @@ static FORCEINLINE __vec4_i1 __select(__vec4_i1 mask, __vec4_i1 a, __vec4_i1 b)
return _mm_blendv_ps(b.v, a.v, mask.v);
}
static FORCEINLINE bool __extract_element(__vec4_i1 v, int index) {
return ((int32_t *)&v)[index] ? true : false;
}
static FORCEINLINE void __insert_element(__vec4_i1 *v, int index, bool val) {
((int32_t *)v)[index] = val ? -1 : 0;
}
template <int ALIGN> static FORCEINLINE __vec4_i1 __load(const __vec4_i1 *v) {
// FIXME: handle align of 16...
@@ -564,13 +614,6 @@ static FORCEINLINE __vec4_i8 __select(__vec4_i1 mask, __vec4_i8 a, __vec4_i8 b)
_mm_extract_epi8(b.v, 3));
}
static FORCEINLINE int8_t __extract_element(__vec4_i8 v, int index) {
return ((int8_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i8 *v, int index, int8_t val) {
((int8_t *)v)[index] = val;
}
template <class RetVecType> __vec4_i8 __smear_i8(int8_t v);
template <> FORCEINLINE __vec4_i8 __smear_i8<__vec4_i8>(int8_t v) {
@@ -598,6 +641,20 @@ static FORCEINLINE __vec4_i8 __rotate_i8(__vec4_i8 v, int delta) {
__extract_element(v, (delta+3) & 0x3));
}
static FORCEINLINE __vec4_i8 __shift_i8(__vec4_i8 v, int delta) {
int8_t v1, v2, v3, v4;
int d1, d2, d3, d4;
d1 = delta+0;
d2 = delta+1;
d3 = delta+2;
d4 = delta+3;
v1 = ((d1 >= 0) && (d1 < 4)) ? __extract_element(v, d1) : 0;
v2 = ((d2 >= 0) && (d2 < 4)) ? __extract_element(v, d2) : 0;
v3 = ((d3 >= 0) && (d3 < 4)) ? __extract_element(v, d3) : 0;
v4 = ((d4 >= 0) && (d4 < 4)) ? __extract_element(v, d4) : 0;
return __vec4_i8(v1, v2, v3, v4);
}
static FORCEINLINE __vec4_i8 __shuffle_i8(__vec4_i8 v, __vec4_i32 index) {
return __vec4_i8(__extract_element(v, __extract_element(index, 0) & 0x3),
__extract_element(v, __extract_element(index, 1) & 0x3),
@@ -836,13 +893,6 @@ static FORCEINLINE __vec4_i16 __select(__vec4_i1 mask, __vec4_i16 a, __vec4_i16
_mm_extract_epi16(b.v, 3));
}
static FORCEINLINE int16_t __extract_element(__vec4_i16 v, int index) {
return ((int16_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i16 *v, int index, int16_t val) {
((int16_t *)v)[index] = val;
}
template <class RetVecType> __vec4_i16 __smear_i16(int16_t v);
template <> FORCEINLINE __vec4_i16 __smear_i16<__vec4_i16>(int16_t v) {
@@ -870,6 +920,20 @@ static FORCEINLINE __vec4_i16 __rotate_i16(__vec4_i16 v, int delta) {
__extract_element(v, (delta+3) & 0x3));
}
static FORCEINLINE __vec4_i16 __shift_i16(__vec4_i16 v, int delta) {
int16_t v1, v2, v3, v4;
int d1, d2, d3, d4;
d1 = delta+0;
d2 = delta+1;
d3 = delta+2;
d4 = delta+3;
v1 = ((d1 >= 0) && (d1 < 4)) ? __extract_element(v, d1) : 0;
v2 = ((d2 >= 0) && (d2 < 4)) ? __extract_element(v, d2) : 0;
v3 = ((d3 >= 0) && (d3 < 4)) ? __extract_element(v, d3) : 0;
v4 = ((d4 >= 0) && (d4 < 4)) ? __extract_element(v, d4) : 0;
return __vec4_i16(v1, v2, v3, v4);
}
static FORCEINLINE __vec4_i16 __shuffle_i16(__vec4_i16 v, __vec4_i32 index) {
return __vec4_i16(__extract_element(v, __extract_element(index, 0) & 0x3),
__extract_element(v, __extract_element(index, 1) & 0x3),
@@ -1109,13 +1173,6 @@ template <> FORCEINLINE __vec4_i32 __undef_i32<__vec4_i32>() {
return __vec4_i32();
}
static FORCEINLINE int32_t __extract_element(__vec4_i32 v, int index) {
return ((int32_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i32 *v, int index, int32_t val) {
((int32_t *)v)[index] = val;
}
static FORCEINLINE __vec4_i32 __broadcast_i32(__vec4_i32 v, int index) {
return _mm_set1_epi32(__extract_element(v, index));
@@ -1128,6 +1185,21 @@ static FORCEINLINE __vec4_i32 __rotate_i32(__vec4_i32 v, int delta) {
__extract_element(v, (delta+3) & 0x3));
}
#include <iostream>
static FORCEINLINE __vec4_i32 __shift_i32(const __vec4_i32 &v, int delta) {
int32_t v1, v2, v3, v4;
int32_t d1, d2, d3, d4;
d1 = delta+0;
d2 = delta+1;
d3 = delta+2;
d4 = delta+3;
v1 = ((d1 >= 0) && (d1 < 4)) ? __extract_element(v, d1) : 0;
v2 = ((d2 >= 0) && (d2 < 4)) ? __extract_element(v, d2) : 0;
v3 = ((d3 >= 0) && (d3 < 4)) ? __extract_element(v, d3) : 0;
v4 = ((d4 >= 0) && (d4 < 4)) ? __extract_element(v, d4) : 0;
return __vec4_i32(v1, v2, v3, v4);
}
static FORCEINLINE __vec4_i32 __shuffle_i32(__vec4_i32 v, __vec4_i32 index) {
return __vec4_i32(__extract_element(v, __extract_element(index, 0) & 0x3),
__extract_element(v, __extract_element(index, 1) & 0x3),
@@ -1383,13 +1455,6 @@ template <> FORCEINLINE __vec4_i64 __undef_i64<__vec4_i64>() {
return __vec4_i64();
}
static FORCEINLINE int64_t __extract_element(__vec4_i64 v, int index) {
return ((int64_t *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_i64 *v, int index, int64_t val) {
((int64_t *)v)[index] = val;
}
static FORCEINLINE __vec4_i64 __broadcast_i64(__vec4_i64 v, int index) {
uint64_t val = __extract_element(v, index);
@@ -1403,6 +1468,20 @@ static FORCEINLINE __vec4_i64 __rotate_i64(__vec4_i64 v, int delta) {
__extract_element(v, (delta+3) & 0x3));
}
static FORCEINLINE __vec4_i64 __shift_i64(__vec4_i64 v, int delta) {
int64_t v1, v2, v3, v4;
int d1, d2, d3, d4;
d1 = delta+0;
d2 = delta+1;
d3 = delta+2;
d4 = delta+3;
v1 = ((d1 >= 0) && (d1 < 4)) ? __extract_element(v, d1) : 0;
v2 = ((d2 >= 0) && (d2 < 4)) ? __extract_element(v, d2) : 0;
v3 = ((d3 >= 0) && (d3 < 4)) ? __extract_element(v, d3) : 0;
v4 = ((d4 >= 0) && (d4 < 4)) ? __extract_element(v, d4) : 0;
return __vec4_i64(v1, v2, v3, v4);
}
static FORCEINLINE __vec4_i64 __shuffle_i64(__vec4_i64 v, __vec4_i32 index) {
return __vec4_i64(__extract_element(v, __extract_element(index, 0) & 0x3),
__extract_element(v, __extract_element(index, 1) & 0x3),
@@ -1504,13 +1583,6 @@ template <> FORCEINLINE __vec4_f __undef_float<__vec4_f>() {
return __vec4_f();
}
static FORCEINLINE float __extract_element(__vec4_f v, int index) {
return ((float *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_f *v, int index, float val) {
((float *)v)[index] = val;
}
static FORCEINLINE __vec4_f __broadcast_float(__vec4_f v, int index) {
return _mm_set1_ps(__extract_element(v, index));
@@ -1523,6 +1595,20 @@ static FORCEINLINE __vec4_f __rotate_float(__vec4_f v, int delta) {
__extract_element(v, (delta+3) & 0x3));
}
static FORCEINLINE __vec4_f __shift_float(__vec4_f v, int delta) {
float v1, v2, v3, v4;
int d1, d2, d3, d4;
d1 = delta+0;
d2 = delta+1;
d3 = delta+2;
d4 = delta+3;
v1 = ((d1 >= 0) && (d1 < 4)) ? __extract_element(v, d1) : 0.f;
v2 = ((d2 >= 0) && (d2 < 4)) ? __extract_element(v, d2) : 0.f;
v3 = ((d3 >= 0) && (d3 < 4)) ? __extract_element(v, d3) : 0.f;
v4 = ((d4 >= 0) && (d4 < 4)) ? __extract_element(v, d4) : 0.f;
return __vec4_f(v1, v2, v3, v4);
}
static FORCEINLINE __vec4_f __shuffle_float(__vec4_f v, __vec4_i32 index) {
return __vec4_f(__extract_element(v, __extract_element(index, 0) & 0x3),
__extract_element(v, __extract_element(index, 1) & 0x3),
@@ -1656,13 +1742,6 @@ template <> FORCEINLINE __vec4_d __undef_double<__vec4_d>() {
return __vec4_d();
}
static FORCEINLINE double __extract_element(__vec4_d v, int index) {
return ((double *)&v)[index];
}
static FORCEINLINE void __insert_element(__vec4_d *v, int index, double val) {
((double *)v)[index] = val;
}
static FORCEINLINE __vec4_d __broadcast_double(__vec4_d v, int index) {
return __vec4_d(_mm_set1_pd(__extract_element(v, index)),
@@ -1676,6 +1755,20 @@ static FORCEINLINE __vec4_d __rotate_double(__vec4_d v, int delta) {
__extract_element(v, (delta+3) & 0x3));
}
static FORCEINLINE __vec4_d __shift_double(__vec4_d v, int delta) {
double v1, v2, v3, v4;
int d1, d2, d3, d4;
d1 = delta+0;
d2 = delta+1;
d3 = delta+2;
d4 = delta+3;
v1 = ((d1 >= 0) && (d1 < 4)) ? __extract_element(v, d1) : 0;
v2 = ((d2 >= 0) && (d2 < 4)) ? __extract_element(v, d2) : 0;
v3 = ((d3 >= 0) && (d3 < 4)) ? __extract_element(v, d3) : 0;
v4 = ((d4 >= 0) && (d4 < 4)) ? __extract_element(v, d4) : 0;
return __vec4_d(v1, v2, v3, v4);
}
static FORCEINLINE __vec4_d __shuffle_double(__vec4_d v, __vec4_i32 index) {
return __vec4_d(__extract_element(v, __extract_element(index, 0) & 0x3),
__extract_element(v, __extract_element(index, 1) & 0x3),
@@ -1889,7 +1982,7 @@ static FORCEINLINE __vec4_f __cast_sitofp(__vec4_f, __vec4_i16 val) {
(float)((int16_t)_mm_extract_epi16(val.v, 3)));
}
static FORCEINLINE __vec4_f __cast_sitofp(__vec4_f, __vec4_i32 val) {
static FORCEINLINE __vec4_f __cast_sitofp(__vec4_f, const __vec4_i32 &val) {
return _mm_cvtepi32_ps(val.v);
}

95
opt.cpp
View File

@@ -72,6 +72,7 @@
#include <llvm/Analysis/ConstantFolding.h>
#include <llvm/Target/TargetLibraryInfo.h>
#include <llvm/ADT/Triple.h>
#include <llvm/ADT/SmallSet.h>
#include <llvm/Transforms/Scalar.h>
#include <llvm/Transforms/IPO.h>
#include <llvm/Transforms/Utils/BasicBlockUtils.h>
@@ -124,6 +125,8 @@ static llvm::Pass *CreateMakeInternalFuncsStaticPass();
static llvm::Pass *CreateDebugPass(char * output);
static llvm::Pass *CreateReplaceStdlibShiftPass();
#define DEBUG_START_PASS(NAME) \
if (g->debugPrint && \
(getenv("FUNC") == NULL || \
@@ -521,6 +524,7 @@ Optimize(llvm::Module *module, int optLevel) {
optPM.add(llvm::createPromoteMemoryToRegisterPass());
optPM.add(llvm::createAggressiveDCEPass());
if (g->opt.disableGatherScatterOptimizations == false &&
g->target->getVectorWidth() > 1) {
optPM.add(llvm::createInstructionCombiningPass(), 210);
@@ -546,7 +550,8 @@ Optimize(llvm::Module *module, int optLevel) {
optPM.add(llvm::createGlobalOptimizerPass());
optPM.add(llvm::createReassociatePass());
optPM.add(llvm::createIPConstantPropagationPass());
optPM.add(llvm::createDeadArgEliminationPass());
optPM.add(CreateReplaceStdlibShiftPass(),229);
optPM.add(llvm::createDeadArgEliminationPass(),230);
optPM.add(llvm::createInstructionCombiningPass());
optPM.add(llvm::createCFGSimplificationPass());
optPM.add(llvm::createPruneEHPass());
@@ -4879,6 +4884,7 @@ lMatchAvgDownInt16(llvm::Value *inst) {
}
#endif // !LLVM_3_1 && !LLVM_3_2
bool
PeepholePass::runOnBasicBlock(llvm::BasicBlock &bb) {
DEBUG_START_PASS("PeepholePass");
@@ -4923,3 +4929,90 @@ static llvm::Pass *
CreatePeepholePass() {
return new PeepholePass;
}
#include <iostream>
/** Given an llvm::Value known to be an integer, return its value as
an int64_t.
*/
static int64_t
lGetIntValue(llvm::Value *offset) {
llvm::ConstantInt *intOffset = llvm::dyn_cast<llvm::ConstantInt>(offset);
Assert(intOffset && (intOffset->getBitWidth() == 32 ||
intOffset->getBitWidth() == 64));
return intOffset->getSExtValue();
}
///////////////////////////////////////////////////////////////////////////
// ReplaceStdlibShiftPass
class ReplaceStdlibShiftPass : public llvm::BasicBlockPass {
public:
static char ID;
ReplaceStdlibShiftPass() : BasicBlockPass(ID) {
}
const char *getPassName() const { return "Resolve \"replace extract insert chains\""; }
bool runOnBasicBlock(llvm::BasicBlock &BB);
};
char ReplaceStdlibShiftPass::ID = 0;
bool
ReplaceStdlibShiftPass::runOnBasicBlock(llvm::BasicBlock &bb) {
DEBUG_START_PASS("ReplaceStdlibShiftPass");
bool modifiedAny = false;
llvm::Function *shifts[6];
shifts[0] = m->module->getFunction("__shift_i8");
shifts[1] = m->module->getFunction("__shift_i16");
shifts[2] = m->module->getFunction("__shift_i32");
shifts[3] = m->module->getFunction("__shift_i64");
shifts[4] = m->module->getFunction("__shift_float");
shifts[5] = m->module->getFunction("__shift_double");
for (llvm::BasicBlock::iterator iter = bb.begin(), e = bb.end(); iter != e; ++iter) {
llvm::Instruction *inst = &*iter;
if (llvm::CallInst *ci = llvm::dyn_cast<llvm::CallInst>(inst)) {
llvm::Function *func = ci->getCalledFunction();
for (int i = 0; i < 6; i++) {
if (shifts[i] == func) {
// we matched a call
llvm::Value *shiftedVec = ci->getArgOperand(0);
llvm::Value *shiftAmt = ci->getArgOperand(1);
if (llvm::isa<llvm::Constant>(shiftAmt)) {
int vectorWidth = g->target->getVectorWidth();
int shuffleVals[vectorWidth];
int shiftInt = lGetIntValue(shiftAmt);
for (int i = 0; i < vectorWidth; i++) {
int s = i + shiftInt;
s = (s < 0) ? vectorWidth : s;
s = (s >= vectorWidth) ? vectorWidth : s;
shuffleVals[i] = s;
}
llvm::Value *shuffleIdxs = LLVMInt32Vector(shuffleVals);
llvm::Value *zeroVec = llvm::ConstantAggregateZero::get(shiftedVec->getType());
llvm::Value *shuffle = new llvm::ShuffleVectorInst(shiftedVec, zeroVec,
shuffleIdxs, "vecShift", ci);
ci->replaceAllUsesWith(shuffle);
modifiedAny = true;
} else {
PerformanceWarning(SourcePos(), "Stdlib shift() called without constant shift amount.");
}
}
}
}
}
DEBUG_END_PASS("ReplaceStdlibShiftPass");
return modifiedAny;
}
static llvm::Pass *
CreateReplaceStdlibShiftPass() {
return new ReplaceStdlibShiftPass();
}

54
stdlib.ispc
View File

@@ -170,6 +170,60 @@ static inline int64 rotate(int64 v, uniform int i) {
return __rotate_i64(v, i);
}
__declspec(safe)
static inline float shift(float v, uniform int i) {
varying float result;
unmasked {
result = __shift_float(v, i);
}
return result;
}
__declspec(safe)
static inline int8 shift(int8 v, uniform int i) {
varying int8 result;
unmasked {
result = __shift_i8(v, i);
}
return result;
}
__declspec(safe)
static inline int16 shift(int16 v, uniform int i) {
varying int16 result;
unmasked {
result = __shift_i16(v, i);
}
return result;
}
__declspec(safe)
static inline int32 shift(int32 v, uniform int i) {
varying int32 result;
unmasked {
result = __shift_i32(v, i);
}
return result;
}
__declspec(safe)
static inline double shift(double v, uniform int i) {
varying double result;
unmasked {
result = __shift_double(v, i);
}
return result;
}
__declspec(safe)
static inline int64 shift(int64 v, uniform int i) {
varying int64 result;
unmasked {
result = __shift_i64(v, i);
}
return result;
}
__declspec(safe)
static inline float shuffle(float v, int i) {
return __shuffle_float(v, i);

14
tests/shift-1.ispc Normal file
View File

@@ -0,0 +1,14 @@
export uniform int width() { return programCount; }
export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
int a = aFOO[programIndex];
int rot = shift(a, -1);
RET[programIndex] = rot;
}
export void result(uniform float RET[]) {
varying int val = programIndex;
if (val < 0) val = 0;
RET[programIndex] = val;
}

15
tests/shift-2.ispc Normal file
View File

@@ -0,0 +1,15 @@
export uniform int width() { return programCount; }
export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
int a = aFOO[programIndex];
uniform int delta = b - 6; // -1
int rot = shift(a, delta);
RET[programIndex] = rot;
}
export void result(uniform float RET[]) {
varying int val = programIndex;
if (val < 0) val = 0;
RET[programIndex] = val;
}

14
tests/shift-3.ispc Normal file
View File

@@ -0,0 +1,14 @@
export uniform int width() { return programCount; }
export void f_fu(uniform float RET[], uniform float aFOO[], uniform float b) {
int a = aFOO[programIndex];
int rot = shift(a, 1);
RET[programIndex] = rot;
}
export void result(uniform float RET[]) {
varying int val = 2 + programIndex;
if (val > programCount) val = 0;
RET[programIndex] = val;
}