diff -r 000000000000 -r 6474c204b198 gfx/qcms/transform-altivec.c --- /dev/null Thu Jan 01 00:00:00 1970 +0000 +++ b/gfx/qcms/transform-altivec.c Wed Dec 31 06:09:35 2014 +0100 @@ -0,0 +1,269 @@ +/* vim: set ts=8 sw=8 noexpandtab: */ +// qcms +// Copyright (C) 2009 Mozilla Corporation +// Copyright (C) 1998-2007 Marti Maria +// +// Permission is hereby granted, free of charge, to any person obtaining +// a copy of this software and associated documentation files (the "Software"), +// to deal in the Software without restriction, including without limitation +// the rights to use, copy, modify, merge, publish, distribute, sublicense, +// and/or sell copies of the Software, and to permit persons to whom the Software +// is furnished to do so, subject to the following conditions: +// +// The above copyright notice and this permission notice shall be included in +// all copies or substantial portions of the Software. +// +// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, +// EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO +// THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND +// NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE +// LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION +// OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +// WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +#include + +#include "qcmsint.h" + +#define FLOATSCALE (float)(PRECACHE_OUTPUT_SIZE) +#define CLAMPMAXVAL (((float) (PRECACHE_OUTPUT_SIZE - 1)) / PRECACHE_OUTPUT_SIZE) +static const ALIGN float floatScaleX4 = FLOATSCALE; +static const ALIGN float clampMaxValueX4 = CLAMPMAXVAL; + +inline vector float load_aligned_float(float *dataPtr) +{ + vector float data = vec_lde(0, dataPtr); + vector unsigned char moveToStart = vec_lvsl(0, dataPtr); + return vec_perm(data, data, moveToStart); +} + +void qcms_transform_data_rgb_out_lut_altivec(qcms_transform *transform, + unsigned char *src, + unsigned char *dest, + size_t length) +{ + unsigned int i; + float (*mat)[4] = transform->matrix; + char input_back[32]; + /* Ensure we have a buffer that's 16 byte aligned regardless of the original + * stack alignment. We can't use __attribute__((aligned(16))) or __declspec(align(32)) + * because they don't work on stack variables. gcc 4.4 does do the right thing + * on x86 but that's too new for us right now. For more info: gcc bug #16660 */ + float const *input = (float*)(((uintptr_t)&input_back[16]) & ~0xf); + /* share input and output locations to save having to keep the + * locations in separate registers */ + uint32_t const *output = (uint32_t*)input; + + /* deref *transform now to avoid it in loop */ + const float *igtbl_r = transform->input_gamma_table_r; + const float *igtbl_g = transform->input_gamma_table_g; + const float *igtbl_b = transform->input_gamma_table_b; + + /* deref *transform now to avoid it in loop */ + const uint8_t *otdata_r = &transform->output_table_r->data[0]; + const uint8_t *otdata_g = &transform->output_table_g->data[0]; + const uint8_t *otdata_b = &transform->output_table_b->data[0]; + + /* input matrix values never change */ + const vector float mat0 = vec_ldl(0, (vector float*)mat[0]); + const vector float mat1 = vec_ldl(0, (vector float*)mat[1]); + const vector float mat2 = vec_ldl(0, (vector float*)mat[2]); + + /* these values don't change, either */ + const vector float max = vec_splat(vec_lde(0, (float*)&clampMaxValueX4), 0); + const vector float min = (vector float)vec_splat_u32(0); + const vector float scale = vec_splat(vec_lde(0, (float*)&floatScaleX4), 0); + + /* working variables */ + vector float vec_r, vec_g, vec_b, result; + + /* CYA */ + if (!length) + return; + + /* one pixel is handled outside of the loop */ + length--; + + /* setup for transforming 1st pixel */ + vec_r = load_aligned_float((float*)&igtbl_r[src[0]]); + vec_g = load_aligned_float((float*)&igtbl_r[src[1]]); + vec_b = load_aligned_float((float*)&igtbl_r[src[2]]); + src += 3; + + /* transform all but final pixel */ + + for (i=0; imatrix; + char input_back[32]; + /* Ensure we have a buffer that's 16 byte aligned regardless of the original + * stack alignment. We can't use __attribute__((aligned(16))) or __declspec(align(32)) + * because they don't work on stack variables. gcc 4.4 does do the right thing + * on x86 but that's too new for us right now. For more info: gcc bug #16660 */ + float const *input = (float*)(((uintptr_t)&input_back[16]) & ~0xf); + /* share input and output locations to save having to keep the + * locations in separate registers */ + uint32_t const *output = (uint32_t*)input; + + /* deref *transform now to avoid it in loop */ + const float *igtbl_r = transform->input_gamma_table_r; + const float *igtbl_g = transform->input_gamma_table_g; + const float *igtbl_b = transform->input_gamma_table_b; + + /* deref *transform now to avoid it in loop */ + const uint8_t *otdata_r = &transform->output_table_r->data[0]; + const uint8_t *otdata_g = &transform->output_table_g->data[0]; + const uint8_t *otdata_b = &transform->output_table_b->data[0]; + + /* input matrix values never change */ + const vector float mat0 = vec_ldl(0, (vector float*)mat[0]); + const vector float mat1 = vec_ldl(0, (vector float*)mat[1]); + const vector float mat2 = vec_ldl(0, (vector float*)mat[2]); + + /* these values don't change, either */ + const vector float max = vec_splat(vec_lde(0, (float*)&clampMaxValueX4), 0); + const vector float min = (vector float)vec_splat_u32(0); + const vector float scale = vec_splat(vec_lde(0, (float*)&floatScaleX4), 0); + + /* working variables */ + vector float vec_r, vec_g, vec_b, result; + unsigned char alpha; + + /* CYA */ + if (!length) + return; + + /* one pixel is handled outside of the loop */ + length--; + + /* setup for transforming 1st pixel */ + vec_r = load_aligned_float((float*)&igtbl_r[src[0]]); + vec_g = load_aligned_float((float*)&igtbl_r[src[1]]); + vec_b = load_aligned_float((float*)&igtbl_r[src[2]]); + alpha = src[3]; + src += 4; + + /* transform all but final pixel */ + + for (i=0; i