summaryrefslogtreecommitdiff
path: root/src/mesa
diff options
context:
space:
mode:
authorEric Anholt <eric@anholt.net>2009-08-12 09:52:44 -0700
committerEric Anholt <eric@anholt.net>2009-11-06 21:08:54 -0800
commit1e5400c575b72fbde16ef0d55fd3ba577fc1b6a5 (patch)
tree697344a3527c71654b2839bb60259ae1f0fbf854 /src/mesa
parent7059aa0eff9ff6ec361e584b413f63b25762a89c (diff)
i965: Share math functions between brw_wm_glsl.c and brw_wm_emit.c.
Diffstat (limited to 'src/mesa')
-rw-r--r--src/mesa/drivers/dri/i965/brw_wm.h16
-rw-r--r--src/mesa/drivers/dri/i965/brw_wm_emit.c139
-rw-r--r--src/mesa/drivers/dri/i965/brw_wm_glsl.c177
3 files changed, 111 insertions, 221 deletions
diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h
index df8e467aaf..7d470e8dfe 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.h
+++ b/src/mesa/drivers/dri/i965/brw_wm.h
@@ -357,11 +357,27 @@ void emit_mad(struct brw_compile *p,
const struct brw_reg *arg0,
const struct brw_reg *arg1,
const struct brw_reg *arg2);
+void emit_math1(struct brw_wm_compile *c,
+ GLuint function,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0);
+void emit_math2(struct brw_wm_compile *c,
+ GLuint function,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1);
void emit_sop(struct brw_compile *p,
const struct brw_reg *dst,
GLuint mask,
GLuint cond,
const struct brw_reg *arg0,
const struct brw_reg *arg1);
+void emit_xpd(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1);
#endif
diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c
index 41ebadb553..763fe4b4d6 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_emit.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c
@@ -665,11 +665,11 @@ void emit_dph(struct brw_compile *p,
}
-static void emit_xpd( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *arg1 )
+void emit_xpd(struct brw_compile *p,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1)
{
GLuint i;
@@ -690,41 +690,68 @@ static void emit_xpd( struct brw_compile *p,
}
-static void emit_math1( struct brw_compile *p,
- GLuint function,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0 )
+void emit_math1(struct brw_wm_compile *c,
+ GLuint function,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0)
{
+ struct brw_compile *p = &c->func;
int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1;
+ GLuint saturate = ((mask & SATURATE) ?
+ BRW_MATH_SATURATE_SATURATE :
+ BRW_MATH_SATURATE_NONE);
if (!(mask & WRITEMASK_XYZW))
return; /* Do not emit dead code */
assert(is_power_of_two(mask & WRITEMASK_XYZW));
+ /* If compressed, this will write message reg 2,3 from arg0.x's 16
+ * channels.
+ */
brw_MOV(p, brw_message_reg(2), arg0[0]);
/* Send two messages to perform all 16 operations:
*/
- brw_math_16(p,
- dst[dst_chan],
+ brw_push_insn_state(p);
+ brw_set_compression_control(p, BRW_COMPRESSION_NONE);
+ brw_math(p,
+ dst[dst_chan],
+ function,
+ saturate,
+ 2,
+ brw_null_reg(),
+ BRW_MATH_DATA_VECTOR,
+ BRW_MATH_PRECISION_FULL);
+
+ if (c->dispatch_width == 16) {
+ brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+ brw_math(p,
+ offset(dst[dst_chan],1),
function,
- (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 2,
+ saturate,
+ 3,
brw_null_reg(),
+ BRW_MATH_DATA_VECTOR,
BRW_MATH_PRECISION_FULL);
+ }
+ brw_pop_insn_state(p);
}
-static void emit_math2( struct brw_compile *p,
- GLuint function,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0,
- const struct brw_reg *arg1)
+void emit_math2(struct brw_wm_compile *c,
+ GLuint function,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0,
+ const struct brw_reg *arg1)
{
+ struct brw_compile *p = &c->func;
int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1;
+ GLuint saturate = ((mask & SATURATE) ?
+ BRW_MATH_SATURATE_SATURATE :
+ BRW_MATH_SATURATE_NONE);
if (!(mask & WRITEMASK_XYZW))
return; /* Do not emit dead code */
@@ -735,37 +762,41 @@ static void emit_math2( struct brw_compile *p,
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, brw_message_reg(2), arg0[0]);
- brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
- brw_MOV(p, brw_message_reg(4), sechalf(arg0[0]));
+ if (c->dispatch_width == 16) {
+ brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+ brw_MOV(p, brw_message_reg(4), sechalf(arg0[0]));
+ }
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_MOV(p, brw_message_reg(3), arg1[0]);
- brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
- brw_MOV(p, brw_message_reg(5), sechalf(arg1[0]));
+ if (c->dispatch_width == 16) {
+ brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+ brw_MOV(p, brw_message_reg(5), sechalf(arg1[0]));
+ }
-
- /* Send two messages to perform all 16 operations:
- */
brw_set_compression_control(p, BRW_COMPRESSION_NONE);
brw_math(p,
dst[dst_chan],
function,
- (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
+ saturate,
2,
brw_null_reg(),
BRW_MATH_DATA_VECTOR,
BRW_MATH_PRECISION_FULL);
- brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
- brw_math(p,
- offset(dst[dst_chan],1),
- function,
- (mask & SATURATE) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 4,
- brw_null_reg(),
- BRW_MATH_DATA_VECTOR,
- BRW_MATH_PRECISION_FULL);
-
+ /* Send two messages to perform all 16 operations:
+ */
+ if (c->dispatch_width == 16) {
+ brw_set_compression_control(p, BRW_COMPRESSION_2NDHALF);
+ brw_math(p,
+ offset(dst[dst_chan],1),
+ function,
+ saturate,
+ 4,
+ brw_null_reg(),
+ BRW_MATH_DATA_VECTOR,
+ BRW_MATH_PRECISION_FULL);
+ }
brw_pop_insn_state(p);
}
@@ -909,11 +940,13 @@ static void emit_txb( struct brw_wm_compile *c,
}
-static void emit_lit( struct brw_compile *p,
- const struct brw_reg *dst,
- GLuint mask,
- const struct brw_reg *arg0 )
+static void emit_lit(struct brw_wm_compile *c,
+ const struct brw_reg *dst,
+ GLuint mask,
+ const struct brw_reg *arg0)
{
+ struct brw_compile *p = &c->func;
+
assert((mask & WRITEMASK_XW) == 0);
if (mask & WRITEMASK_Y) {
@@ -923,7 +956,7 @@ static void emit_lit( struct brw_compile *p,
}
if (mask & WRITEMASK_Z) {
- emit_math2(p, BRW_MATH_FUNCTION_POW,
+ emit_math2(c, BRW_MATH_FUNCTION_POW,
&dst[2],
WRITEMASK_X | (mask & SATURATE),
&arg0[1],
@@ -1380,27 +1413,27 @@ void brw_wm_emit( struct brw_wm_compile *c )
/* Higher math functions:
*/
case OPCODE_RCP:
- emit_math1(p, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]);
break;
case OPCODE_RSQ:
- emit_math1(p, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]);
break;
case OPCODE_SIN:
- emit_math1(p, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]);
break;
case OPCODE_COS:
- emit_math1(p, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]);
break;
case OPCODE_EX2:
- emit_math1(p, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]);
break;
case OPCODE_LG2:
- emit_math1(p, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]);
break;
case OPCODE_SCS:
@@ -1408,13 +1441,13 @@ void brw_wm_emit( struct brw_wm_compile *c )
* fixup for 16-element execution.
*/
if (dst_flags & WRITEMASK_X)
- emit_math1(p, BRW_MATH_FUNCTION_COS, dst, (dst_flags&SATURATE)|WRITEMASK_X, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_COS, dst, (dst_flags&SATURATE)|WRITEMASK_X, args[0]);
if (dst_flags & WRITEMASK_Y)
- emit_math1(p, BRW_MATH_FUNCTION_SIN, dst+1, (dst_flags&SATURATE)|WRITEMASK_X, args[0]);
+ emit_math1(c, BRW_MATH_FUNCTION_SIN, dst+1, (dst_flags&SATURATE)|WRITEMASK_X, args[0]);
break;
case OPCODE_POW:
- emit_math2(p, BRW_MATH_FUNCTION_POW, dst, dst_flags, args[0], args[1]);
+ emit_math2(c, BRW_MATH_FUNCTION_POW, dst, dst_flags, args[0], args[1]);
break;
/* Comparisons:
@@ -1452,7 +1485,7 @@ void brw_wm_emit( struct brw_wm_compile *c )
break;
case OPCODE_LIT:
- emit_lit(p, dst, dst_flags, args[0]);
+ emit_lit(c, dst, dst_flags, args[0]);
break;
/* Texturing operations:
diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
index 2df47344d5..02b119154f 100644
--- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c
+++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c
@@ -550,42 +550,6 @@ static struct brw_reg get_src_reg(struct brw_wm_compile *c,
}
}
-
-/**
- * Same as \sa get_src_reg() but if the register is a literal, emit
- * a brw_reg encoding the literal.
- * Note that a brw instruction only allows one src operand to be a literal.
- * For instructions with more than one operand, only the second can be a
- * literal. This means that we treat some literals as constants/uniforms
- * (which why PROGRAM_CONSTANT is checked in fetch_constants()).
- *
- */
-static struct brw_reg get_src_reg_imm(struct brw_wm_compile *c,
- const struct prog_instruction *inst,
- GLuint srcRegIndex, GLuint channel)
-{
- const struct prog_src_register *src = &inst->SrcReg[srcRegIndex];
- if (src->File == PROGRAM_CONSTANT) {
- /* a literal */
- const int component = GET_SWZ(src->Swizzle, channel);
- const GLfloat *param =
- c->fp->program.Base.Parameters->ParameterValues[src->Index];
- GLfloat value = param[component];
- if (src->Negate & (1 << channel))
- value = -value;
- if (src->Abs)
- value = FABSF(value);
-#if 0
- printf(" form immed value %f for chan %d\n", value, channel);
-#endif
- return brw_imm_f(value);
- }
- else {
- return get_src_reg(c, inst, srcRegIndex, channel);
- }
-}
-
-
/**
* Subroutines are minimal support for resusable instruction sequences.
* They are implemented as simply as possible to minimise overhead: there
@@ -1013,100 +977,6 @@ static void emit_frontfacing(struct brw_wm_compile *c,
brw_set_predicate_control_flag_value(p, 0xff);
}
-static void emit_xpd(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- int i;
- struct brw_compile *p = &c->func;
- GLuint mask = inst->DstReg.WriteMask;
- for (i = 0; i < 4; i++) {
- GLuint i2 = (i+2)%3;
- GLuint i1 = (i+1)%3;
- if (mask & (1<<i)) {
- struct brw_reg src0, src1, dst;
- dst = get_dst_reg(c, inst, i);
- src0 = negate(get_src_reg(c, inst, 0, i2));
- src1 = get_src_reg_imm(c, inst, 1, i1);
- brw_MUL(p, brw_null_reg(), src0, src1);
- src0 = get_src_reg(c, inst, 0, i1);
- src1 = get_src_reg_imm(c, inst, 1, i2);
- brw_set_saturate(p, inst->SaturateMode != SATURATE_OFF);
- brw_MAC(p, dst, src0, src1);
- brw_set_saturate(p, 0);
- }
- }
- brw_set_saturate(p, 0);
-}
-
-/**
- * Emit a scalar instruction, like RCP, RSQ, LOG, EXP.
- * Note that the result of the function is smeared across the dest
- * register's X, Y, Z and W channels (subject to writemasking of course).
- */
-static void emit_math1(struct brw_wm_compile *c,
- const struct prog_instruction *inst, GLuint func)
-{
- struct brw_compile *p = &c->func;
- struct brw_reg src0, dst;
- GLuint mask = inst->DstReg.WriteMask;
- int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1;
-
- if (!(mask & WRITEMASK_XYZW))
- return;
-
- assert(is_power_of_two(mask & WRITEMASK_XYZW));
-
- /* Get first component of source register */
- dst = get_dst_reg(c, inst, dst_chan);
- src0 = get_src_reg(c, inst, 0, 0);
-
- brw_MOV(p, brw_message_reg(2), src0);
- brw_math(p,
- dst,
- func,
- (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 2,
- brw_null_reg(),
- BRW_MATH_DATA_VECTOR,
- BRW_MATH_PRECISION_FULL);
-}
-
-static void emit_rcp(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_INV);
-}
-
-static void emit_rsq(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_RSQ);
-}
-
-static void emit_sin(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_SIN);
-}
-
-static void emit_cos(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_COS);
-}
-
-static void emit_ex2(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_EXP);
-}
-
-static void emit_lg2(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- emit_math1(c, inst, BRW_MATH_FUNCTION_LOG);
-}
-
static void emit_arl(struct brw_wm_compile *c,
const struct prog_instruction *inst)
{
@@ -1169,36 +1039,6 @@ static void emit_min_max(struct brw_wm_compile *c,
release_tmps(c, mark);
}
-static void emit_pow(struct brw_wm_compile *c,
- const struct prog_instruction *inst)
-{
- struct brw_compile *p = &c->func;
- struct brw_reg dst, src0, src1;
- GLuint mask = inst->DstReg.WriteMask;
- int dst_chan = _mesa_ffs(mask & WRITEMASK_XYZW) - 1;
-
- if (!(mask & WRITEMASK_XYZW))
- return;
-
- assert(is_power_of_two(mask & WRITEMASK_XYZW));
-
- dst = get_dst_reg(c, inst, dst_chan);
- src0 = get_src_reg_imm(c, inst, 0, 0);
- src1 = get_src_reg_imm(c, inst, 1, 0);
-
- brw_MOV(p, brw_message_reg(2), src0);
- brw_MOV(p, brw_message_reg(3), src1);
-
- brw_math(p,
- dst,
- BRW_MATH_FUNCTION_POW,
- (inst->SaturateMode != SATURATE_OFF) ? BRW_MATH_SATURATE_SATURATE : BRW_MATH_SATURATE_NONE,
- 2,
- brw_null_reg(),
- BRW_MATH_DATA_VECTOR,
- BRW_MATH_PRECISION_FULL);
-}
-
/**
* For GLSL shaders, this KIL will be unconditional.
* It may be contained inside an IF/ENDIF structure of course.
@@ -2594,28 +2434,28 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
emit_dp4(p, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_XPD:
- emit_xpd(c, inst);
+ emit_xpd(p, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_DPH:
emit_dph(p, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_RCP:
- emit_rcp(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_INV, dst, dst_flags, args[0]);
break;
case OPCODE_RSQ:
- emit_rsq(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_RSQ, dst, dst_flags, args[0]);
break;
case OPCODE_SIN:
- emit_sin(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_SIN, dst, dst_flags, args[0]);
break;
case OPCODE_COS:
- emit_cos(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_COS, dst, dst_flags, args[0]);
break;
case OPCODE_EX2:
- emit_ex2(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_EXP, dst, dst_flags, args[0]);
break;
case OPCODE_LG2:
- emit_lg2(c, inst);
+ emit_math1(c, BRW_MATH_FUNCTION_LOG, dst, dst_flags, args[0]);
break;
case OPCODE_MIN:
case OPCODE_MAX:
@@ -2654,7 +2494,8 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c)
emit_alu2(p, brw_MUL, dst, dst_flags, args[0], args[1]);
break;
case OPCODE_POW:
- emit_pow(c, inst);
+ emit_math2(c, BRW_MATH_FUNCTION_POW,
+ dst, dst_flags, args[0], args[1]);
break;
case OPCODE_MAD:
emit_mad(p, dst, dst_flags, args[0], args[1], args[2]);