/* loop over verts */
for (i = 0; i < count; i += MAX_VERTICES) {
const uint max_vertices = MIN2(MAX_VERTICES, count - i);
- float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4] ALIGN16_ATTRIB;
- float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4] ALIGN16_ATTRIB;
- float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4]);
+ PIPE_ALIGN_VAR(16, float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4]);
+ PIPE_ALIGN_VAR(16, float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4]);
uint attr;
/* convert (up to) four input verts to SoA format */
* Since it's pretty much impossible to form PPC vector immediates, load
* them from memory here:
*/
-const float ppc_builtin_constants[] ALIGN16_ATTRIB = {
+PIPE_ALIGN_VAR(16, const float ppc_builtin_constants[]) = {
1.0f, -128.0f, 128.0, 0.0
};
/** This is the object passed to spe_create_thread() */
+PIPE_ALIGN_TYPE(16,
struct cell_init_info
{
unsigned id;
uint *buffer_status; /**< points at cell_context->buffer_status */
struct cell_spu_function_info *spu_functions;
-} ALIGN16_ATTRIB;
+});
#endif /* CELL_COMMON_H */
*/
struct cell_buffer_list
{
- struct cell_fence fence ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, struct cell_fence fence);
struct cell_buffer_node *head;
};
/** Mapped constant buffers */
void *mapped_constants[PIPE_SHADER_TYPES];
- struct cell_spu_function_info spu_functions ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, struct cell_spu_function_info spu_functions);
uint num_cells, num_spus;
/** Buffers for command batches, vertex/index data */
uint buffer_size[CELL_NUM_BUFFERS];
- ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, ubyte buffer[CELL_NUM_BUFFERS][CELL_BUFFER_SIZE]);
int cur_batch; /**< which buffer is being filled w/ commands */
/** [4] to ensure 16-byte alignment for each status word */
- uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, uint buffer_status[CELL_MAX_SPUS][CELL_NUM_BUFFERS][4]);
/** Associated with each command/batch buffer is a list of pipe_buffers
/**
* Buffers containing dynamically generated SPU code:
*/
-static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS]
- ALIGN16_ATTRIB;
+PIPE_ALIGN_VAR(16, static unsigned char attribute_fetch_code_buffer[136 * PIPE_MAX_ATTRIBS]);
{
const uint buf = (opcode >> 8) & 0xff;
uint size = (opcode >> 16);
- qword buffer[CELL_BUFFER_SIZE / 16] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, qword buffer[CELL_BUFFER_SIZE / 16]);
const unsigned usize = ROUNDUP16(size) / sizeof(buffer[0]);
uint pos;
/* execute declarations (interpolants) */
if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
for (i = 0; i < mach->NumDeclarations; i++) {
+ PIPE_ALIGN_VAR(16,
union {
struct tgsi_full_declaration decl;
qword buffer[ROUNDUP16(sizeof(struct tgsi_full_declaration)) / 16];
- } d ALIGN16_ATTRIB;
+ } d);
unsigned ea = (unsigned) (mach->Declarations + pc);
spu_dcache_fetch_unaligned(d.buffer, ea, sizeof(d.decl));
/* execute instructions, until pc is set to -1 */
while (pc != -1) {
+ PIPE_ALIGN_VAR(16,
union {
struct tgsi_full_instruction inst;
qword buffer[ROUNDUP16(sizeof(struct tgsi_full_instruction)) / 16];
- } i ALIGN16_ATTRIB;
+ } i);
unsigned ea = (unsigned) (mach->Instructions + pc);
spu_dcache_fetch_unaligned(i.buffer, ea, sizeof(i.inst));
* 4 internal temporaries
* 1 address
*/
+ PIPE_ALIGN_VAR(16,
struct spu_exec_vector Temps[TGSI_EXEC_NUM_TEMPS
- + TGSI_EXEC_NUM_TEMP_EXTRAS + 1]
- ALIGN16_ATTRIB;
+ + TGSI_EXEC_NUM_TEMP_EXTRAS + 1]);
struct spu_exec_vector *Addrs;
void
return_function_info(void)
{
- struct cell_spu_function_info funcs ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, struct cell_spu_function_info funcs);
int tag = TAG_MISC;
ASSERT(sizeof(funcs) == 256); /* must be multiple of 16 bytes */
vector float *constants);
+PIPE_ALIGN_TYPE(16,
struct spu_framebuffer
{
void *color_start; /**< addr of color surface in main memory */
uint zsize; /**< 0, 2 or 4 bytes per Z */
float zscale; /**< 65535.0, 2^24-1 or 2^32-1 */
-} ALIGN16_ATTRIB;
+});
/** per-texture level info */
+PIPE_ALIGN_TYPE(16,
struct spu_texture_level
{
void *start;
vector signed int mask_s, mask_t, mask_r;
/** texcoord clamp limits */
vector signed int max_s, max_t, max_r;
-} ALIGN16_ATTRIB;
+});
+PIPE_ALIGN_TYPE(16,
struct spu_texture
{
struct spu_texture_level level[CELL_MAX_TEXTURE_LEVELS];
uint max_level;
uint target; /**< PIPE_TEXTURE_x */
-} ALIGN16_ATTRIB;
+});
/**
* All SPU global/context state will be in a singleton object of this type:
*/
+PIPE_ALIGN_TYPE(16,
struct spu_global
{
/** One-time init/constant info */
struct vertex_info vertex_info;
/** Current color and Z tiles */
- tile_t ctile ALIGN16_ATTRIB;
- tile_t ztile ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, tile_t ctile);
+ PIPE_ALIGN_VAR(16, tile_t ztile);
/** Read depth/stencil tiles? */
boolean read_depth_stencil;
ubyte cur_ctile_status, cur_ztile_status;
/** Status of all tiles in framebuffer */
- ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
- ubyte ztile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, ubyte ctile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE]);
+ PIPE_ALIGN_VAR(16, ubyte ztile_status[CELL_MAX_HEIGHT/TILE_SIZE][CELL_MAX_WIDTH/TILE_SIZE]);
/** Current fragment ops machine code, at 8-byte boundary */
uint *fragment_ops_code;
spu_fragment_ops_func fragment_ops[2];
/** Current fragment program machine code, at 8-byte boundary */
- uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS] ALIGN8_ATTRIB;
+ PIPE_ALIGN_VAR(8, uint fragment_program_code[SPU_MAX_FRAGMENT_PROGRAM_INSTS]);
/** Current fragment ops function */
spu_fragment_program_func fragment_program;
/** Fragment program constants */
vector float constants[4 * CELL_MAX_CONSTANTS];
-} ALIGN16_ATTRIB;
+});
extern struct spu_global spu;
cmd_render(const struct cell_command_render *render, uint *pos_incr)
{
/* we'll DMA into these buffers */
- ubyte vertex_data[CELL_BUFFER_SIZE] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, ubyte vertex_data[CELL_BUFFER_SIZE]);
const uint vertex_size = render->vertex_size; /* in bytes */
/*const*/ uint total_vertex_bytes = render->num_verts * vertex_size;
uint index_bytes;
const qword *shuffle_data);
-static const qword fetch_shuffle_data[5] ALIGN16_ATTRIB = {
+PIPE_ALIGN_VAR(16, static const qword fetch_shuffle_data[5]) = {
/* Shuffle used by CVT_64_FLOAT
*/
{
unsigned idx;
const unsigned bytes_per_entry = draw->vertex_fetch.size[attr];
const unsigned quads_per_entry = (bytes_per_entry + 15) / 16;
- qword in[2 * 4] ALIGN16_ATTRIB;
+ PIPE_ALIGN_VAR(16, qword in[2 * 4]);
/* Fetch four attributes for four vertices.
struct spu_exec_machine *machine = &draw->machine;
unsigned int j;
- ALIGN16_DECL(struct spu_exec_vector, inputs, PIPE_MAX_ATTRIBS);
- ALIGN16_DECL(struct spu_exec_vector, outputs, PIPE_MAX_ATTRIBS);
+ PIPE_ALIGN_VAR(16, struct spu_exec_vector inputs[PIPE_MAX_ATTRIBS]);
+ PIPE_ALIGN_VAR(16, struct spu_exec_vector outputs[PIPE_MAX_ATTRIBS]);
const float *scale = draw->viewport.scale;
const float *trans = draw->viewport.translate;
ASSERT_ALIGN16(draw->constants);
machine->Consts = (float (*)[4]) draw->constants;
- machine->Inputs = ALIGN16_ASSIGN(inputs);
- machine->Outputs = ALIGN16_ASSIGN(outputs);
+ machine->Inputs = inputs;
+ machine->Outputs = outputs;
spu_vertex_fetch( draw, machine, elts, count );
for (j = 0; j < count; j++) {
unsigned slot;
float x, y, z, w;
+ PIPE_ALIGN_VAR(16,
unsigned char buffer[sizeof(struct vertex_header)
- + MAX_VERTEX_SIZE] ALIGN16_ATTRIB;
+ + MAX_VERTEX_SIZE]);
struct vertex_header *const tmpOut =
(struct vertex_header *) buffer;
const unsigned vert_size = ROUNDUP16(sizeof(struct vertex_header)
}
-unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]
- ALIGN16_ATTRIB;
+PIPE_ALIGN_VAR(16,
+unsigned char immediates[(sizeof(float) * 4 * TGSI_EXEC_NUM_IMMEDIATES) + 32]);
void
#ifndef LP_QUAD_H
#define LP_QUAD_H
+#include "pipe/p_compiler.h"
#include "pipe/p_state.h"
#include "tgsi/tgsi_exec.h"
struct quad_header_output
{
/** colors in SOA format (rrrr, gggg, bbbb, aaaa) */
- float ALIGN16_ATTRIB color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE];
+ PIPE_ALIGN_VAR(16, float color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE]);
};
*/
struct quad_interp_coef
{
- float ALIGN16_ATTRIB a0[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
- float ALIGN16_ATTRIB dadx[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
- float ALIGN16_ATTRIB dady[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
+ PIPE_ALIGN_VAR(16, float a0[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]);
+ PIPE_ALIGN_VAR(16, float dadx[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]);
+ PIPE_ALIGN_VAR(16, float dady[1 + PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS]);
};
uint8_t *tile;
uint8_t *color;
void *depth;
- uint32_t ALIGN16_ATTRIB mask[4][NUM_CHANNELS];
+ PIPE_ALIGN_VAR(16, uint32_t mask[4][NUM_CHANNELS]);
unsigned chan_index;
unsigned q;
success = TRUE;
for(i = 0; i < n && success; ++i) {
if(mode == AoS) {
- ALIGN16_ATTRIB uint8_t src[LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t dst[LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t con[LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t res[LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t ref[LP_NATIVE_VECTOR_WIDTH/8];
+ PIPE_ALIGN_VAR(16, uint8_t src[LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t dst[LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t con[LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t res[LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t ref[LP_NATIVE_VECTOR_WIDTH/8]);
int64_t start_counter = 0;
int64_t end_counter = 0;
if(mode == SoA) {
const unsigned stride = type.length*type.width/8;
- ALIGN16_ATTRIB uint8_t src[4*LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t dst[4*LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t con[4*LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t res[4*LP_NATIVE_VECTOR_WIDTH/8];
- ALIGN16_ATTRIB uint8_t ref[4*LP_NATIVE_VECTOR_WIDTH/8];
+ PIPE_ALIGN_VAR(16, uint8_t src[4*LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t dst[4*LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t con[4*LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t res[4*LP_NATIVE_VECTOR_WIDTH/8]);
+ PIPE_ALIGN_VAR(16, uint8_t ref[4*LP_NATIVE_VECTOR_WIDTH/8]);
int64_t start_counter = 0;
int64_t end_counter = 0;
boolean mismatch;
for(i = 0; i < n && success; ++i) {
unsigned src_stride = src_type.length*src_type.width/8;
unsigned dst_stride = dst_type.length*dst_type.width/8;
- ALIGN16_ATTRIB uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
- ALIGN16_ATTRIB uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
+ PIPE_ALIGN_VAR(16, uint8_t src[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]);
+ PIPE_ALIGN_VAR(16, uint8_t dst[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH]);
double fref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
uint8_t ref[LP_MAX_VECTOR_LENGTH*LP_MAX_VECTOR_LENGTH];
int64_t start_counter = 0;
+/* Macros for data alignment. */
#if defined(__GNUC__)
-#define ALIGN16_DECL(TYPE, NAME, SIZE) TYPE NAME##___aligned[SIZE] __attribute__(( aligned( 16 ) ))
-#define ALIGN16_ASSIGN(NAME) NAME##___aligned
-#define ALIGN16_ATTRIB __attribute__(( aligned( 16 ) ))
-#define ALIGN8_ATTRIB __attribute__(( aligned( 8 ) ))
+
+/* See http://gcc.gnu.org/onlinedocs/gcc-4.4.2/gcc/Type-Attributes.html */
+#define PIPE_ALIGN_TYPE(_alignment, _type) _type __attribute__((aligned(_alignment)))
+
+/* See http://gcc.gnu.org/onlinedocs/gcc-4.4.2/gcc/Variable-Attributes.html */
+#define PIPE_ALIGN_VAR(_alignment, _decl) _decl __attribute__((aligned(_alignment)))
+
#if (__GNUC__ > 4 || (__GNUC__ == 4 &&__GNUC_MINOR__>1)) && !defined(PIPE_ARCH_X86_64)
#define ALIGN_STACK __attribute__((force_align_arg_pointer))
#else
#define ALIGN_STACK
#endif
-#else
-#define ALIGN16_DECL(TYPE, NAME, SIZE) TYPE NAME##___unaligned[SIZE + 1]
-#define ALIGN16_ASSIGN(NAME) align16(NAME##___unaligned)
-#define ALIGN16_ATTRIB
-#define ALIGN8_ATTRIB
+
+#elif defined(_MSC_VER)
+
+/* See http://msdn.microsoft.com/en-us/library/83ythb65.aspx */
+#define PIPE_ALIGN_TYPE(_alignment, _type) __declspec(align(_alignment)) _type
+#define PIPE_ALIGN_VAR(_alignment, _decl) __declspec(align(_alignment)) _decl
+
#define ALIGN_STACK
+
+#else
+
+#error "Unsupported compiler"
+
#endif
projects / mesa.git / commitdiff