#include "util/u_debug.h"
#include "util/u_math.h"
#include "util/u_memory.h"
+#include "tgsi/tgsi_info.h"
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_util.h"
#include "tgsi/tgsi_exec.h"
#include "lp_bld_const.h"
#include "lp_bld_intr.h"
#include "lp_bld_arit.h"
+#include "lp_bld_logic.h"
#include "lp_bld_swizzle.h"
#include "lp_bld_tgsi.h"
+#include "lp_bld_debug.h"
#define LP_MAX_TEMPS 256
for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)
#define IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
- ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+ ((INST)->FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
#define IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
if (IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
{
struct lp_build_context base;
- LLVMValueRef (*inputs)[4];
+ LLVMValueRef x, y, w;
+ LLVMValueRef a0_ptr;
+ LLVMValueRef dadx_ptr;
+ LLVMValueRef dady_ptr;
+
LLVMValueRef consts_ptr;
- LLVMValueRef (*outputs)[4];
+ LLVMValueRef (*outputs)[NUM_CHANNELS];
LLVMValueRef samplers_ptr;
- LLVMValueRef immediates[LP_MAX_IMMEDIATES][4];
- LLVMValueRef temps[LP_MAX_TEMPS][4];
+ LLVMValueRef oow;
+
+ LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
+
+ LLVMValueRef immediates[LP_MAX_IMMEDIATES][NUM_CHANNELS];
+ LLVMValueRef temps[LP_MAX_TEMPS][NUM_CHANNELS];
+
+ LLVMValueRef mask;
/** Coords/texels store */
LLVMValueRef store_ptr;
/**
* Register fetch.
*/
-
static LLVMValueRef
emit_fetch(
struct lp_build_tgsi_soa_context *bld,
- const struct tgsi_full_src_register *reg,
+ const struct tgsi_full_instruction *inst,
+ unsigned index,
const unsigned chan_index )
{
+ const struct tgsi_full_src_register *reg = &inst->FullSrcRegisters[index];
unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
LLVMValueRef res;
default:
assert( 0 );
+ return bld->base.undef;
}
break;
default:
assert( 0 );
+ return bld->base.undef;
}
switch( tgsi_util_get_full_src_register_sign_mode( reg, chan_index ) ) {
return res;
}
-#define FETCH( FUNC, INST, INDEX, CHAN )\
- emit_fetch( FUNC, &(INST).FullSrcRegisters[INDEX], CHAN )
/**
* Register store.
*/
-
static void
emit_store(
struct lp_build_tgsi_soa_context *bld,
- const struct tgsi_full_dst_register *reg,
const struct tgsi_full_instruction *inst,
+ unsigned index,
unsigned chan_index,
LLVMValueRef value)
{
+ const struct tgsi_full_dst_register *reg = &inst->FullDstRegisters[index];
+
switch( inst->Instruction.Saturate ) {
case TGSI_SAT_NONE:
break;
}
}
-#define STORE( FUNC, INST, INDEX, CHAN, VAL )\
- emit_store( FUNC, &(INST).FullDstRegisters[INDEX], &(INST), CHAN, VAL )
-
void PIPE_CDECL
lp_build_tgsi_fetch_texel_soa( struct tgsi_sampler **samplers,
}
if(apply_lodbias)
- lodbias = FETCH( bld, *inst, 0, 3 );
+ lodbias = emit_fetch( bld, inst, 0, 3 );
else
lodbias = bld->base.zero;
"store");
if (projected) {
- oow = FETCH( bld, *inst, 0, 3 );
+ oow = emit_fetch( bld, inst, 0, 3 );
oow = lp_build_rcp(&bld->base, oow);
}
LLVMValueRef coord_ptr = LLVMBuildGEP(bld->base.builder, bld->store_ptr, &index, 1, "");
LLVMValueRef coord;
- coord = FETCH( bld, *inst, 0, i );
+ coord = emit_fetch( bld, inst, 0, i );
if (projected)
coord = lp_build_mul(&bld->base, coord, oow);
lp_build_intrinsic(bld->base.builder, "fetch_texel", LLVMVoidType(), args, 3);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, i ) {
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, i ) {
LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
LLVMValueRef res_ptr = LLVMBuildGEP(bld->base.builder, bld->store_ptr, &index, 1, "");
LLVMValueRef res = LLVMBuildLoad(bld->base.builder, res_ptr, "");
- STORE( bld, *inst, 0, i, res );
+ emit_store( bld, inst, 0, i, res );
}
}
static void
emit_kil(
struct lp_build_tgsi_soa_context *bld,
- const struct tgsi_full_src_register *reg )
+ const struct tgsi_full_instruction *inst )
{
-#if 0
- unsigned uniquemask;
- unsigned unique_count = 0;
+ const struct tgsi_full_src_register *reg = &inst->FullSrcRegisters[0];
+ LLVMValueRef terms[NUM_CHANNELS];
unsigned chan_index;
- unsigned i;
- /* This mask stores component bits that were already tested. Note that
- * we test if the value is less than zero, so 1.0 and 0.0 need not to be
- * tested. */
- uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE);
+ memset(&terms, 0, sizeof terms);
FOR_EACH_CHANNEL( chan_index ) {
unsigned swizzle;
- /* unswizzle channel */
- swizzle = tgsi_util_get_full_src_register_extswizzle(
- reg,
- chan_index );
+ /* Unswizzle channel */
+ swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
- /* check if the component has not been already tested */
- if( !(uniquemask & (1 << swizzle)) ) {
- uniquemask |= 1 << swizzle;
+ /* Note that we test if the value is less than zero, so 1.0 and 0.0 need
+ * not to be tested. */
+ if(swizzle == TGSI_EXTSWIZZLE_ZERO || swizzle == TGSI_EXTSWIZZLE_ONE)
+ continue;
- /* allocate register */
- emit_fetch(
- bld,
- unique_count++,
- reg,
- chan_index );
- }
+ /* Check if the component has not been already tested. */
+ assert(swizzle < NUM_CHANNELS);
+ if( !terms[swizzle] )
+ /* TODO: change the comparison operator instead of setting the sign */
+ terms[swizzle] = emit_fetch(bld, inst, 0, chan_index );
}
- x86_push(
- bld,
- x86_make_reg( file_REG32, reg_AX ) );
- x86_push(
- bld,
- x86_make_reg( file_REG32, reg_DX ) );
-
- for (i = 0 ; i < unique_count; i++ ) {
- LLVMValueRef dataXMM = make_xmm(i);
-
- sse_cmpps(
- bld,
- dataXMM,
- get_temp(
- TGSI_EXEC_TEMP_00000000_I,
- TGSI_EXEC_TEMP_00000000_C ),
- cc_LessThan );
-
- if( i == 0 ) {
- sse_movmskps(
- bld,
- x86_make_reg( file_REG32, reg_AX ),
- dataXMM );
- }
- else {
- sse_movmskps(
- bld,
- x86_make_reg( file_REG32, reg_DX ),
- dataXMM );
- x86_or(
- bld,
- x86_make_reg( file_REG32, reg_AX ),
- x86_make_reg( file_REG32, reg_DX ) );
+ FOR_EACH_CHANNEL( chan_index ) {
+ if(terms[chan_index]) {
+ LLVMValueRef mask;
+
+ mask = lp_build_cmp(&bld->base, PIPE_FUNC_GEQUAL, terms[chan_index], bld->base.zero);
+
+ if(bld->mask)
+ bld->mask = LLVMBuildAnd(bld->base.builder, bld->mask, mask, "");
+ else
+ bld->mask = mask;
}
}
-
- x86_or(
- bld,
- get_temp(
- TGSI_EXEC_TEMP_KILMASK_I,
- TGSI_EXEC_TEMP_KILMASK_C ),
- x86_make_reg( file_REG32, reg_AX ) );
-
- x86_pop(
- bld,
- x86_make_reg( file_REG32, reg_DX ) );
- x86_pop(
- bld,
- x86_make_reg( file_REG32, reg_AX ) );
-#endif
}
switch (inst->Instruction.Opcode) {
#if 0
case TGSI_OPCODE_ARL:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_flr(bld, 0, 0);
emit_f2it( bld, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
#endif
case TGSI_OPCODE_MOV:
case TGSI_OPCODE_SWZ:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_LIT:
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ) {
- STORE( bld, *inst, 0, CHAN_X, bld->base.one);
- }
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
- STORE( bld, *inst, 0, CHAN_W, bld->base.one);
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ) {
+ emit_store( bld, inst, 0, CHAN_X, bld->base.one);
}
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
- tmp0 = lp_build_max( &bld->base, tmp0, bld->base.one);
- STORE( bld, *inst, 0, CHAN_Y, tmp0);
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ) {
+ src0 = emit_fetch( bld, inst, 0, CHAN_X );
+ dst0 = lp_build_max( &bld->base, src0, bld->base.zero);
+ emit_store( bld, inst, 0, CHAN_Y, dst0);
}
-#if 0
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) {
/* XMM[1] = SrcReg[0].yyyy */
- tmp1 = FETCH( bld, *inst, 0, CHAN_Y );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Y );
/* XMM[1] = max(XMM[1], 0) */
tmp1 = lp_build_max( &bld->base, tmp1, bld->base.zero);
/* XMM[2] = SrcReg[0].wwww */
- tmp2 = FETCH( bld, *inst, 0, CHAN_W );
+ tmp2 = emit_fetch( bld, inst, 0, CHAN_W );
tmp1 = lp_build_pow( &bld->base, tmp1, tmp2);
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
- tmp2 = bld->base.zero;
- sse_cmpps(
- bld,
- make_xmm( 2 ),
- make_xmm( 0 ),
- cc_LessThan );
- sse_andps(
- bld,
- make_xmm( 2 ),
- make_xmm( 1 ) );
- STORE( bld, *inst, 0, CHAN_Z, tmp2);
- }
- break;
-#else
- return 0;
-#endif
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
+ tmp2 = lp_build_cmp(&bld->base, PIPE_FUNC_GREATER, tmp0, bld->base.zero);
+ dst0 = lp_build_select(&bld->base, tmp2, tmp1, bld->base.zero);
+ emit_store( bld, inst, 0, CHAN_Z, dst0);
+ }
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) ) {
+ emit_store( bld, inst, 0, CHAN_W, bld->base.one);
+ }
+ break;
case TGSI_OPCODE_RCP:
/* TGSI_OPCODE_RECIP */
- src0 = FETCH( bld, *inst, 0, CHAN_X );
+ src0 = emit_fetch( bld, inst, 0, CHAN_X );
dst0 = lp_build_rcp(&bld->base, src0);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, dst0 );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, dst0 );
}
break;
case TGSI_OPCODE_RSQ:
/* TGSI_OPCODE_RECIPSQRT */
- src0 = FETCH( bld, *inst, 0, CHAN_X );
+ src0 = emit_fetch( bld, inst, 0, CHAN_X );
src0 = lp_build_abs(&bld->base, src0);
dst0 = lp_build_rsqrt(&bld->base, src0);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, dst0 );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, dst0 );
}
break;
case TGSI_OPCODE_EXP:
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) {
LLVMValueRef *p_exp2_int_part = NULL;
LLVMValueRef *p_frac_part = NULL;
LLVMValueRef *p_exp2 = NULL;
- src0 = FETCH( bld, *inst, 0, CHAN_X );
+ src0 = emit_fetch( bld, inst, 0, CHAN_X );
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ))
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ))
p_exp2_int_part = &tmp0;
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ))
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ))
p_frac_part = &tmp1;
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ))
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ))
p_exp2 = &tmp2;
lp_build_exp2_approx(&bld->base, src0, p_exp2_int_part, p_frac_part, p_exp2);
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ))
- STORE( bld, *inst, 0, CHAN_X, tmp0);
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ))
- STORE( bld, *inst, 0, CHAN_Y, tmp1);
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ))
- STORE( bld, *inst, 0, CHAN_Z, tmp2);
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ))
+ emit_store( bld, inst, 0, CHAN_X, tmp0);
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ))
+ emit_store( bld, inst, 0, CHAN_Y, tmp1);
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ))
+ emit_store( bld, inst, 0, CHAN_Z, tmp2);
}
/* dst.w = 1.0 */
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W )) {
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_W )) {
tmp0 = bld->base.one;
- STORE( bld, *inst, 0, CHAN_W, tmp0);
+ emit_store( bld, inst, 0, CHAN_W, tmp0);
}
break;
case TGSI_OPCODE_LOG:
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z )) {
LLVMValueRef *p_floor_log2;
LLVMValueRef *p_exp;
LLVMValueRef *p_log2;
- src0 = FETCH( bld, *inst, 0, CHAN_X );
+ src0 = emit_fetch( bld, inst, 0, CHAN_X );
src0 = lp_build_abs( &bld->base, src0 );
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ))
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ))
p_floor_log2 = &tmp0;
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ))
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ))
p_exp = &tmp1;
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ))
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ))
p_log2 = &tmp2;
lp_build_log2_approx(&bld->base, src0, p_exp, p_floor_log2, p_log2);
/* dst.x = floor(lg2(abs(src.x))) */
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ))
- STORE( bld, *inst, 0, CHAN_X, tmp0);
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ))
+ emit_store( bld, inst, 0, CHAN_X, tmp0);
/* dst.y = abs(src)/ex2(floor(lg2(abs(src.x)))) */
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y )) {
tmp1 = lp_build_div( &bld->base, src0, tmp1);
- STORE( bld, *inst, 0, CHAN_Y, tmp1);
+ emit_store( bld, inst, 0, CHAN_Y, tmp1);
}
/* dst.z = lg2(abs(src.x)) */
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ))
- STORE( bld, *inst, 0, CHAN_Z, tmp2);
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ))
+ emit_store( bld, inst, 0, CHAN_Z, tmp2);
}
/* dst.w = 1.0 */
- if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W )) {
+ if (IS_DST0_CHANNEL_ENABLED( inst, CHAN_W )) {
tmp0 = bld->base.one;
- STORE( bld, *inst, 0, CHAN_W, tmp0);
+ emit_store( bld, inst, 0, CHAN_W, tmp0);
}
break;
case TGSI_OPCODE_MUL:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- src0 = FETCH( bld, *inst, 0, chan_index );
- src1 = FETCH( bld, *inst, 1, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
dst0 = lp_build_mul(&bld->base, src0, src1);
- STORE( bld, *inst, 0, chan_index, dst0);
+ emit_store( bld, inst, 0, chan_index, dst0);
}
break;
case TGSI_OPCODE_ADD:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- src0 = FETCH( bld, *inst, 0, chan_index );
- src1 = FETCH( bld, *inst, 1, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
dst0 = lp_build_add(&bld->base, src0, src1);
- STORE( bld, *inst, 0, chan_index, dst0);
+ emit_store( bld, inst, 0, chan_index, dst0);
}
break;
case TGSI_OPCODE_DP3:
/* TGSI_OPCODE_DOT3 */
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
- tmp1 = FETCH( bld, *inst, 1, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_X );
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_Y );
- tmp2 = FETCH( bld, *inst, 1, CHAN_Y );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Y );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Y );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_Z );
- tmp2 = FETCH( bld, *inst, 1, CHAN_Z );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Z );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Z );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_DP4:
/* TGSI_OPCODE_DOT4 */
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
- tmp1 = FETCH( bld, *inst, 1, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_X );
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_Y );
- tmp2 = FETCH( bld, *inst, 1, CHAN_Y );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Y );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Y );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_Z );
- tmp2 = FETCH( bld, *inst, 1, CHAN_Z );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Z );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Z );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_W );
- tmp2 = FETCH( bld, *inst, 1, CHAN_W );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_W );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_W );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_DST:
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) {
tmp0 = bld->base.one;
- STORE( bld, *inst, 0, CHAN_X, tmp0);
+ emit_store( bld, inst, 0, CHAN_X, tmp0);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
- tmp0 = FETCH( bld, *inst, 0, CHAN_Y );
- tmp1 = FETCH( bld, *inst, 1, CHAN_Y );
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) {
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_Y );
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_Y );
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
- STORE( bld, *inst, 0, CHAN_Y, tmp0);
+ emit_store( bld, inst, 0, CHAN_Y, tmp0);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
- tmp0 = FETCH( bld, *inst, 0, CHAN_Z );
- STORE( bld, *inst, 0, CHAN_Z, tmp0);
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) {
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_Z );
+ emit_store( bld, inst, 0, CHAN_Z, tmp0);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
- tmp0 = FETCH( bld, *inst, 1, CHAN_W );
- STORE( bld, *inst, 0, CHAN_W, tmp0);
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) {
+ tmp0 = emit_fetch( bld, inst, 1, CHAN_W );
+ emit_store( bld, inst, 0, CHAN_W, tmp0);
}
break;
case TGSI_OPCODE_MIN:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- src0 = FETCH( bld, *inst, 0, chan_index );
- src1 = FETCH( bld, *inst, 1, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
dst0 = lp_build_min( &bld->base, src0, src1 );
- STORE( bld, *inst, 0, chan_index, dst0);
+ emit_store( bld, inst, 0, chan_index, dst0);
}
break;
case TGSI_OPCODE_MAX:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- src0 = FETCH( bld, *inst, 0, chan_index );
- src1 = FETCH( bld, *inst, 1, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
dst0 = lp_build_max( &bld->base, src0, src1 );
- STORE( bld, *inst, 0, chan_index, dst0);
+ emit_store( bld, inst, 0, chan_index, dst0);
}
break;
-#if 0
case TGSI_OPCODE_SLT:
/* TGSI_OPCODE_SETLT */
- emit_setcc( bld, inst, cc_LessThan );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LESS, src0, src1 );
+ dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
case TGSI_OPCODE_SGE:
/* TGSI_OPCODE_SETGE */
- emit_setcc( bld, inst, cc_NotLessThan );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GEQUAL, src0, src1 );
+ dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
-#endif
case TGSI_OPCODE_MAD:
/* TGSI_OPCODE_MADD */
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
- tmp1 = FETCH( bld, *inst, 1, chan_index );
- tmp2 = FETCH( bld, *inst, 2, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
+ tmp1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp2 = emit_fetch( bld, inst, 2, chan_index );
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
tmp0 = lp_build_add( &bld->base, tmp0, tmp2);
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_SUB:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
- tmp1 = FETCH( bld, *inst, 1, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
+ tmp1 = emit_fetch( bld, inst, 1, chan_index );
tmp0 = lp_build_sub( &bld->base, tmp0, tmp1);
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_LRP:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- src0 = FETCH( bld, *inst, 0, chan_index );
- src1 = FETCH( bld, *inst, 1, chan_index );
- src2 = FETCH( bld, *inst, 2, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ src2 = emit_fetch( bld, inst, 2, chan_index );
tmp0 = lp_build_sub( &bld->base, src1, src2 );
tmp0 = lp_build_mul( &bld->base, src0, tmp0 );
dst0 = lp_build_add( &bld->base, tmp0, src2 );
- STORE( bld, *inst, 0, chan_index, dst0 );
+ emit_store( bld, inst, 0, chan_index, dst0 );
}
break;
break;
case TGSI_OPCODE_DP2A:
- tmp0 = FETCH( bld, *inst, 0, CHAN_X ); /* xmm0 = src[0].x */
- tmp1 = FETCH( bld, *inst, 1, CHAN_X ); /* xmm1 = src[1].x */
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); /* xmm0 = src[0].x */
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); /* xmm1 = src[1].x */
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 * xmm1 */
- tmp1 = FETCH( bld, *inst, 0, CHAN_Y ); /* xmm1 = src[0].y */
- tmp2 = FETCH( bld, *inst, 1, CHAN_Y ); /* xmm2 = src[1].y */
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); /* xmm1 = src[0].y */
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); /* xmm2 = src[1].y */
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); /* xmm1 = xmm1 * xmm2 */
tmp0 = lp_build_add( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 + xmm1 */
- tmp1 = FETCH( bld, *inst, 2, CHAN_X ); /* xmm1 = src[2].x */
+ tmp1 = emit_fetch( bld, inst, 2, CHAN_X ); /* xmm1 = src[2].x */
tmp0 = lp_build_add( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 + xmm1 */
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0); /* dest[ch] = xmm0 */
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0); /* dest[ch] = xmm0 */
}
break;
#if 0
case TGSI_OPCODE_FRC:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_frc( bld, 0, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
break;
case TGSI_OPCODE_FLR:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_flr( bld, 0, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_ROUND:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_rnd( bld, 0, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
#endif
case TGSI_OPCODE_EX2: {
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
tmp0 = lp_build_exp2( &bld->base, tmp0);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
}
case TGSI_OPCODE_LG2:
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
tmp0 = lp_build_log2( &bld->base, tmp0);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_POW:
- src0 = FETCH( bld, *inst, 0, CHAN_X );
- src1 = FETCH( bld, *inst, 1, CHAN_X );
+ src0 = emit_fetch( bld, inst, 0, CHAN_X );
+ src1 = emit_fetch( bld, inst, 1, CHAN_X );
dst0 = lp_build_pow( &bld->base, src0, src1 );
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, dst0 );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, dst0 );
}
break;
case TGSI_OPCODE_XPD:
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
- tmp1 = FETCH( bld, *inst, 1, CHAN_Z );
- tmp3 = FETCH( bld, *inst, 0, CHAN_Z );
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ) {
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_Z );
+ tmp3 = emit_fetch( bld, inst, 0, CHAN_Z );
}
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
- tmp0 = FETCH( bld, *inst, 0, CHAN_Y );
- tmp4 = FETCH( bld, *inst, 1, CHAN_Y );
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) {
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_Y );
+ tmp4 = emit_fetch( bld, inst, 1, CHAN_Y );
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) {
tmp2 = tmp0;
tmp2 = lp_build_mul( &bld->base, tmp2, tmp1);
tmp5 = tmp3;
tmp5 = lp_build_mul( &bld->base, tmp5, tmp4);
tmp2 = lp_build_sub( &bld->base, tmp2, tmp5);
- STORE( bld, *inst, 0, CHAN_X, tmp2);
+ emit_store( bld, inst, 0, CHAN_X, tmp2);
}
- if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
- IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
- tmp2 = FETCH( bld, *inst, 1, CHAN_X );
- tmp5 = FETCH( bld, *inst, 0, CHAN_X );
+ if( IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) ||
+ IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) ) {
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_X );
+ tmp5 = emit_fetch( bld, inst, 0, CHAN_X );
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) {
tmp3 = lp_build_mul( &bld->base, tmp3, tmp2);
tmp1 = lp_build_mul( &bld->base, tmp1, tmp5);
tmp3 = lp_build_sub( &bld->base, tmp3, tmp1);
- STORE( bld, *inst, 0, CHAN_Y, tmp3);
+ emit_store( bld, inst, 0, CHAN_Y, tmp3);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) {
tmp5 = lp_build_mul( &bld->base, tmp5, tmp4);
tmp0 = lp_build_mul( &bld->base, tmp0, tmp2);
tmp5 = lp_build_sub( &bld->base, tmp5, tmp0);
- STORE( bld, *inst, 0, CHAN_Z, tmp5);
+ emit_store( bld, inst, 0, CHAN_Z, tmp5);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) {
tmp0 = bld->base.one;
- STORE( bld, *inst, 0, CHAN_W, tmp0);
+ emit_store( bld, inst, 0, CHAN_W, tmp0);
}
break;
case TGSI_OPCODE_ABS:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
tmp0 = lp_build_abs( &bld->base, tmp0 ) ;
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
break;
case TGSI_OPCODE_DPH:
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
- tmp1 = FETCH( bld, *inst, 1, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_X );
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_Y );
- tmp2 = FETCH( bld, *inst, 1, CHAN_Y );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Y );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Y );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 0, CHAN_Z );
- tmp2 = FETCH( bld, *inst, 1, CHAN_Z );
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Z );
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Z );
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2);
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- tmp1 = FETCH( bld, *inst, 1, CHAN_W );
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_W );
tmp0 = lp_build_add( &bld->base, tmp0, tmp1);
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_COS:
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
tmp0 = lp_build_cos( &bld->base, tmp0 );
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
emit_kilp( bld );
return 0; /* XXX fix me */
break;
+#endif
case TGSI_OPCODE_KIL:
/* conditional kill */
- emit_kil( bld, &inst->FullSrcRegisters[0] );
+ emit_kil( bld, inst );
break;
-#endif
case TGSI_OPCODE_PK2H:
return 0;
break;
case TGSI_OPCODE_SEQ:
- return 0;
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_EQUAL, src0, src1 );
+ dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
case TGSI_OPCODE_SFL:
break;
case TGSI_OPCODE_SGT:
- return 0;
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src0, src1 );
+ dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
case TGSI_OPCODE_SIN:
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
tmp0 = lp_build_sin( &bld->base, tmp0 );
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
case TGSI_OPCODE_SLE:
- return 0;
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LEQUAL, src0, src1 );
+ dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
case TGSI_OPCODE_SNE:
- return 0;
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_NOTEQUAL, src0, src1 );
+ dst0 = lp_build_select( &bld->base, tmp0, bld->base.one, bld->base.zero );
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
case TGSI_OPCODE_STR:
#if 0
case TGSI_OPCODE_ARR:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_rnd( bld, 0, 0 );
emit_f2it( bld, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
#endif
#if 0
case TGSI_OPCODE_SSG:
/* TGSI_OPCODE_SGN */
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_sgn( bld, 0, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
+#endif
case TGSI_OPCODE_CMP:
- emit_cmp (bld, inst);
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ src0 = emit_fetch( bld, inst, 0, chan_index );
+ src1 = emit_fetch( bld, inst, 1, chan_index );
+ src2 = emit_fetch( bld, inst, 2, chan_index );
+ tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_LESS, src0, bld->base.zero );
+ dst0 = lp_build_select( &bld->base, tmp0, src1, src2);
+ emit_store( bld, inst, 0, chan_index, dst0);
+ }
break;
-#endif
case TGSI_OPCODE_SCS:
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_X ) {
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
tmp0 = lp_build_cos( &bld->base, tmp0 );
- STORE( bld, *inst, 0, CHAN_X, tmp0);
+ emit_store( bld, inst, 0, CHAN_X, tmp0);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
- tmp0 = FETCH( bld, *inst, 0, CHAN_X );
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Y ) {
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X );
tmp0 = lp_build_sin( &bld->base, tmp0 );
- STORE( bld, *inst, 0, CHAN_Y, tmp0);
+ emit_store( bld, inst, 0, CHAN_Y, tmp0);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_Z ) {
tmp0 = bld->base.zero;
- STORE( bld, *inst, 0, CHAN_Z, tmp0);
+ emit_store( bld, inst, 0, CHAN_Z, tmp0);
}
- IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+ IF_IS_DST0_CHANNEL_ENABLED( inst, CHAN_W ) {
tmp0 = bld->base.one;
- STORE( bld, *inst, 0, CHAN_W, tmp0);
+ emit_store( bld, inst, 0, CHAN_W, tmp0);
}
break;
{
uint dims = (inst->Instruction.Opcode == TGSI_OPCODE_NRM) ? 3 : 4;
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X) ||
- IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y) ||
- IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z) ||
- (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_W) && dims == 4)) {
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X) ||
+ IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y) ||
+ IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z) ||
+ (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W) && dims == 4)) {
/* NOTE: Cannot use xmm regs 2/3 here (see emit_rsqrt() above). */
/* xmm4 = src.x */
/* xmm0 = src.x * src.x */
- tmp0 = FETCH(bld, *inst, 0, CHAN_X);
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X)) {
+ tmp0 = emit_fetch(bld, inst, 0, CHAN_X);
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X)) {
tmp4 = tmp0;
}
tmp0 = lp_build_mul( &bld->base, tmp0, tmp0);
/* xmm5 = src.y */
/* xmm0 = xmm0 + src.y * src.y */
- tmp1 = FETCH(bld, *inst, 0, CHAN_Y);
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+ tmp1 = emit_fetch(bld, inst, 0, CHAN_Y);
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y)) {
tmp5 = tmp1;
}
tmp1 = lp_build_mul( &bld->base, tmp1, tmp1);
/* xmm6 = src.z */
/* xmm0 = xmm0 + src.z * src.z */
- tmp1 = FETCH(bld, *inst, 0, CHAN_Z);
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+ tmp1 = emit_fetch(bld, inst, 0, CHAN_Z);
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z)) {
tmp6 = tmp1;
}
tmp1 = lp_build_mul( &bld->base, tmp1, tmp1);
if (dims == 4) {
/* xmm7 = src.w */
/* xmm0 = xmm0 + src.w * src.w */
- tmp1 = FETCH(bld, *inst, 0, CHAN_W);
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_W)) {
+ tmp1 = emit_fetch(bld, inst, 0, CHAN_W);
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W)) {
tmp7 = tmp1;
}
tmp1 = lp_build_mul( &bld->base, tmp1, tmp1);
tmp1 = lp_build_rsqrt( &bld->base, tmp0);
/* dst.x = xmm1 * src.x */
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X)) {
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X)) {
tmp4 = lp_build_mul( &bld->base, tmp4, tmp1);
- STORE(bld, *inst, 0, CHAN_X, tmp4);
+ emit_store(bld, inst, 0, CHAN_X, tmp4);
}
/* dst.y = xmm1 * src.y */
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Y)) {
tmp5 = lp_build_mul( &bld->base, tmp5, tmp1);
- STORE(bld, *inst, 0, CHAN_Y, tmp5);
+ emit_store(bld, inst, 0, CHAN_Y, tmp5);
}
/* dst.z = xmm1 * src.z */
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_Z)) {
tmp6 = lp_build_mul( &bld->base, tmp6, tmp1);
- STORE(bld, *inst, 0, CHAN_Z, tmp6);
+ emit_store(bld, inst, 0, CHAN_Z, tmp6);
}
/* dst.w = xmm1 * src.w */
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X) && dims == 4) {
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_X) && dims == 4) {
tmp7 = lp_build_mul( &bld->base, tmp7, tmp1);
- STORE(bld, *inst, 0, CHAN_W, tmp7);
+ emit_store(bld, inst, 0, CHAN_W, tmp7);
}
}
/* dst0.w = 1.0 */
- if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_W) && dims == 3) {
+ if (IS_DST0_CHANNEL_ENABLED(inst, CHAN_W) && dims == 3) {
tmp0 = bld->base.one;
- STORE(bld, *inst, 0, CHAN_W, tmp0);
+ emit_store(bld, inst, 0, CHAN_W, tmp0);
}
}
break;
break;
case TGSI_OPCODE_DP2:
- tmp0 = FETCH( bld, *inst, 0, CHAN_X ); /* xmm0 = src[0].x */
- tmp1 = FETCH( bld, *inst, 1, CHAN_X ); /* xmm1 = src[1].x */
+ tmp0 = emit_fetch( bld, inst, 0, CHAN_X ); /* xmm0 = src[0].x */
+ tmp1 = emit_fetch( bld, inst, 1, CHAN_X ); /* xmm1 = src[1].x */
tmp0 = lp_build_mul( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 * xmm1 */
- tmp1 = FETCH( bld, *inst, 0, CHAN_Y ); /* xmm1 = src[0].y */
- tmp2 = FETCH( bld, *inst, 1, CHAN_Y ); /* xmm2 = src[1].y */
+ tmp1 = emit_fetch( bld, inst, 0, CHAN_Y ); /* xmm1 = src[0].y */
+ tmp2 = emit_fetch( bld, inst, 1, CHAN_Y ); /* xmm2 = src[1].y */
tmp1 = lp_build_mul( &bld->base, tmp1, tmp2); /* xmm1 = xmm1 * xmm2 */
tmp0 = lp_build_add( &bld->base, tmp0, tmp1); /* xmm0 = xmm0 + xmm1 */
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- STORE( bld, *inst, 0, chan_index, tmp0); /* dest[ch] = xmm0 */
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ emit_store( bld, inst, 0, chan_index, tmp0); /* dest[ch] = xmm0 */
}
break;
#if 0
case TGSI_OPCODE_TRUNC:
- FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
- tmp0 = FETCH( bld, *inst, 0, chan_index );
+ FOR_EACH_DST0_ENABLED_CHANNEL( inst, chan_index ) {
+ tmp0 = emit_fetch( bld, inst, 0, chan_index );
emit_f2it( bld, 0 );
emit_i2f( bld, 0 );
- STORE( bld, *inst, 0, chan_index, tmp0);
+ emit_store( bld, inst, 0, chan_index, tmp0);
}
break;
#endif
struct lp_build_tgsi_soa_context *bld,
struct tgsi_full_declaration *decl )
{
-#if 0
if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+ LLVMBuilderRef builder = bld->base.builder;
unsigned first, last, mask;
- unsigned i, j;
- LLVMValueRef tmp;
+ unsigned attrib, chan;
first = decl->DeclarationRange.First;
last = decl->DeclarationRange.Last;
mask = decl->Declaration.UsageMask;
- for( i = first; i <= last; i++ ) {
- for( j = 0; j < NUM_CHANNELS; j++ ) {
- if( mask & (1 << j) ) {
- switch( decl->Declaration.Interpolate ) {
- case TGSI_INTERPOLATE_CONSTANT:
- bld->inputs[i][j] = bld->interp_coefs[i].a0[j];
- break;
+ for( attrib = first; attrib <= last; attrib++ ) {
+ for( chan = 0; chan < NUM_CHANNELS; chan++ ) {
+ LLVMValueRef input = bld->base.undef;
- case TGSI_INTERPOLATE_LINEAR:
- tmp = bld->interp_coefs[i].a0[j];
- tmp = lp_build_add(&bld->base, tmp, lp_build_mul(&bld->base, bld->pos[0], bld->interp_coefs[i].dadx[j]));
- tmp = lp_build_add(&bld->base, tmp, lp_build_mul(&bld->base, bld->pos[1], bld->interp_coefs[i].dady[j]));
- bld->inputs[i][j] = tmp;
- break;
+ if( mask & (1 << chan) ) {
+ LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), attrib*NUM_CHANNELS + chan, 0);
+ LLVMValueRef a0;
+ LLVMValueRef dadx;
+ LLVMValueRef dady;
+ switch( decl->Declaration.Interpolate ) {
case TGSI_INTERPOLATE_PERSPECTIVE:
- tmp = bld->interp_coefs[i].a0[j];
- tmp = lp_build_add(&bld->base, tmp, lp_build_mul(&bld->base, bld->pos[0], bld->interp_coefs[i].dadx[j]));
- tmp = lp_build_add(&bld->base, tmp, lp_build_mul(&bld->base, bld->pos[1], bld->interp_coefs[i].dady[j]));
- tmp = lp_build_div(&bld->base, tmp, bld->pos[3]);
- bld->inputs[i][j] = tmp;
+ /* fall-through */
+
+ case TGSI_INTERPOLATE_LINEAR: {
+ LLVMValueRef dadx_ptr = LLVMBuildGEP(builder, bld->dadx_ptr, &index, 1, "");
+ LLVMValueRef dady_ptr = LLVMBuildGEP(builder, bld->dady_ptr, &index, 1, "");
+ dadx = LLVMBuildLoad(builder, dadx_ptr, "");
+ dady = LLVMBuildLoad(builder, dady_ptr, "");
+ dadx = lp_build_broadcast_scalar(&bld->base, dadx);
+ dady = lp_build_broadcast_scalar(&bld->base, dady);
+ lp_build_name(dadx, "dadx_%u.%c", attrib, "xyzw"[chan]);
+ lp_build_name(dady, "dady_%u.%c", attrib, "xyzw"[chan]);
+ /* fall-through */
+ }
+
+ case TGSI_INTERPOLATE_CONSTANT: {
+ LLVMValueRef a0_ptr = LLVMBuildGEP(builder, bld->a0_ptr, &index, 1, "");
+ a0 = LLVMBuildLoad(builder, a0_ptr, "");
+ a0 = lp_build_broadcast_scalar(&bld->base, a0);
+ lp_build_name(a0, "a0_%u.%c", attrib, "xyzw"[chan]);
break;
+ }
default:
- assert( 0 );
- break;
+ assert(0);
+ break;
}
+
+ input = a0;
+
+ if (decl->Declaration.Interpolate != TGSI_INTERPOLATE_CONSTANT) {
+ input = lp_build_add(&bld->base, input, lp_build_mul(&bld->base, bld->x, dadx));
+ input = lp_build_add(&bld->base, input, lp_build_mul(&bld->base, bld->y, dady));
+ }
+
+ if (decl->Declaration.Interpolate == TGSI_INTERPOLATE_PERSPECTIVE) {
+ if(!bld->oow)
+ bld->oow = lp_build_rcp(&bld->base, bld->w);
+ input = lp_build_mul(&bld->base, input, bld->oow);
+ }
+
+ lp_build_name(input, "input%u.%c", attrib, "xyzw"[chan]);
}
+
+ bld->inputs[attrib][chan] = input;
}
}
}
-#endif
}
/**
* \param immediates buffer to place immediates, later passed to SSE bld
* \param return 1 for success, 0 if translation failed
*/
-void
+LLVMValueRef
lp_build_tgsi_soa(LLVMBuilderRef builder,
const struct tgsi_token *tokens,
union lp_type type,
- LLVMValueRef (*inputs)[4],
+ LLVMValueRef *pos,
+ LLVMValueRef a0_ptr,
+ LLVMValueRef dadx_ptr,
+ LLVMValueRef dady_ptr,
LLVMValueRef consts_ptr,
LLVMValueRef (*outputs)[4],
LLVMValueRef samplers_ptr)
/* Setup build context */
memset(&bld, 0, sizeof bld);
lp_build_context_init(&bld.base, builder, type);
- bld.inputs = inputs;
+ bld.x = pos[0];
+ bld.y = pos[1];
+ bld.w = pos[3];
+ bld.a0_ptr = a0_ptr;
+ bld.dadx_ptr = dadx_ptr;
+ bld.dady_ptr = dady_ptr;
bld.outputs = outputs;
bld.consts_ptr = consts_ptr;
bld.samplers_ptr = samplers_ptr;
case TGSI_TOKEN_TYPE_INSTRUCTION:
if (!emit_instruction( &bld, &parse.FullToken.FullInstruction )) {
- debug_printf("failed to translate tgsi opcode %d to SSE (%s)\n",
- parse.FullToken.FullInstruction.Instruction.Opcode,
- parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ?
- "vertex shader" : "fragment shader");
+ unsigned opcode = parse.FullToken.FullInstruction.Instruction.Opcode;
+ const struct tgsi_opcode_info *info = tgsi_get_opcode_info(opcode);
+ _debug_printf("warning: failed to translate tgsi opcode %s to LLVM\n",
+ info ? info->mnemonic : "<invalid>");
}
break;
}
tgsi_parse_free( &parse );
+
+ return bld.mask;
}