Mesa (7.9): r300/compiler: saturate Z before the shadow comparison
Marek Olšák
mareko at kemper.freedesktop.org
Sat Mar 12 21:40:28 UTC 2011
Module: Mesa
Branch: 7.9
Commit: 20c7067257e0db51fab8188168e3c5e2656c76b8
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=20c7067257e0db51fab8188168e3c5e2656c76b8
Author: Marek Olšák <maraeo at gmail.com>
Date: Tue Mar 8 04:00:07 2011 +0100
r300/compiler: saturate Z before the shadow comparison
This fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=31159
NOTE: This is a candidate for the 7.9 and 7.10 branches.
(cherry picked from commit eb1acd16130a2945122d373357ee72585e520755)
---
.../drivers/dri/r300/compiler/radeon_program_tex.c | 81 +++++++++++---------
1 files changed, 44 insertions(+), 37 deletions(-)
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c
index 37f6def..df88134 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c
@@ -134,6 +134,13 @@ int radeonTransformTEX(
(compiler->state.unit[inst->U.I.TexSrcUnit].compare_mode_enabled))) {
rc_compare_func comparefunc = compiler->state.unit[inst->U.I.TexSrcUnit].texture_compare_func;
+ /* Fake EQUAL/NOTEQUAL, they are equal to NEVER/ALWAYS due to precision issues anyway. */
+ if (comparefunc == RC_COMPARE_FUNC_EQUAL) {
+ comparefunc = RC_COMPARE_FUNC_NEVER;
+ } else if (comparefunc == RC_COMPARE_FUNC_NOTEQUAL) {
+ comparefunc = RC_COMPARE_FUNC_ALWAYS;
+ }
+
if (comparefunc == RC_COMPARE_FUNC_NEVER || comparefunc == RC_COMPARE_FUNC_ALWAYS) {
inst->U.I.Opcode = RC_OPCODE_MOV;
@@ -146,14 +153,11 @@ int radeonTransformTEX(
return 1;
} else {
- rc_compare_func comparefunc = compiler->state.unit[inst->U.I.TexSrcUnit].texture_compare_func;
struct rc_instruction * inst_rcp = NULL;
- struct rc_instruction * inst_mad;
- struct rc_instruction * inst_cmp;
+ struct rc_instruction *inst_mul, *inst_cmp, *inst_add;
unsigned tmp_texsample;
unsigned tmp_sum;
- unsigned tmp_recip_w = 0;
- int pass, fail, tex;
+ int pass, fail;
/* Save the output register. */
struct rc_dst_register output_reg = inst->U.I.DstReg;
@@ -164,49 +168,49 @@ int radeonTransformTEX(
inst->U.I.DstReg.Index = tmp_texsample;
inst->U.I.DstReg.WriteMask = RC_MASK_XYZW;
- if (inst->U.I.Opcode == RC_OPCODE_TXP) {
- tmp_recip_w = rc_find_free_temporary(c);
+ tmp_sum = rc_find_free_temporary(c);
+ if (inst->U.I.Opcode == RC_OPCODE_TXP) {
/* Compute 1/W. */
inst_rcp = rc_insert_new_instruction(c, inst);
inst_rcp->U.I.Opcode = RC_OPCODE_RCP;
inst_rcp->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rcp->U.I.DstReg.Index = tmp_recip_w;
+ inst_rcp->U.I.DstReg.Index = tmp_sum;
inst_rcp->U.I.DstReg.WriteMask = RC_MASK_W;
inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
inst_rcp->U.I.SrcReg[0].Swizzle =
RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 3));
}
- /* Perspective-divide Z by W (if it's TXP) and add the texture sample (see below). */
- tmp_sum = rc_find_free_temporary(c);
- inst_mad = rc_insert_new_instruction(c, inst_rcp ? inst_rcp : inst);
- inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_mad->U.I.DstReg.Index = tmp_sum;
- inst_mad->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
- inst_mad->U.I.SrcReg[0].Swizzle =
+ /* Divide Z by W (if it's TXP) and saturate. */
+ inst_mul = rc_insert_new_instruction(c, inst_rcp ? inst_rcp : inst);
+ inst_mul->U.I.Opcode = inst->U.I.Opcode == RC_OPCODE_TXP ? RC_OPCODE_MUL : RC_OPCODE_MOV;
+ inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_mul->U.I.DstReg.Index = tmp_sum;
+ inst_mul->U.I.DstReg.WriteMask = RC_MASK_W;
+ inst_mul->U.I.SaturateMode = RC_SATURATE_ZERO_ONE;
+ inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+ inst_mul->U.I.SrcReg[0].Swizzle =
RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 2));
if (inst->U.I.Opcode == RC_OPCODE_TXP) {
- inst_mad->U.I.Opcode = RC_OPCODE_MAD;
- inst_mad->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
- inst_mad->U.I.SrcReg[1].Index = tmp_recip_w;
- inst_mad->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW;
- tex = 2;
- } else {
- inst_mad->U.I.Opcode = RC_OPCODE_ADD;
- tex = 1;
- }
- inst_mad->U.I.SrcReg[tex].File = RC_FILE_TEMPORARY;
- inst_mad->U.I.SrcReg[tex].Index = tmp_texsample;
- inst_mad->U.I.SrcReg[tex].Swizzle = compiler->state.unit[inst->U.I.TexSrcUnit].depth_texture_swizzle;
-
- /* Fake EQUAL/NOTEQUAL, it seems to pass some tests suprisingly. */
- if (comparefunc == RC_COMPARE_FUNC_EQUAL) {
- comparefunc = RC_COMPARE_FUNC_GEQUAL;
- } else if (comparefunc == RC_COMPARE_FUNC_NOTEQUAL) {
- comparefunc = RC_COMPARE_FUNC_LESS;
+ inst_mul->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
+ inst_mul->U.I.SrcReg[1].Index = tmp_sum;
+ inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW;
}
+ /* Add the depth texture value. */
+ inst_add = rc_insert_new_instruction(c, inst_mul);
+ inst_add->U.I.Opcode = RC_OPCODE_ADD;
+ inst_add->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_add->U.I.DstReg.Index = tmp_sum;
+ inst_add->U.I.DstReg.WriteMask = RC_MASK_W;
+ inst_add->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+ inst_add->U.I.SrcReg[0].Index = tmp_sum;
+ inst_add->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW;
+ inst_add->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
+ inst_add->U.I.SrcReg[1].Index = tmp_texsample;
+ inst_add->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW;
+
/* Recall that SrcReg[0] is r, SrcReg[tex] is tex and:
* LESS: r < tex <=> -tex+r < 0
* GEQUAL: r >= tex <=> not (-tex+r < 0)
@@ -215,9 +219,9 @@ int radeonTransformTEX(
*
* This negates either r or tex: */
if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GEQUAL)
- inst_mad->U.I.SrcReg[tex].Negate = inst_mad->U.I.SrcReg[tex].Negate ^ RC_MASK_XYZW;
+ inst_add->U.I.SrcReg[1].Negate = inst_add->U.I.SrcReg[1].Negate ^ RC_MASK_XYZW;
else
- inst_mad->U.I.SrcReg[0].Negate = inst_mad->U.I.SrcReg[0].Negate ^ RC_MASK_XYZW;
+ inst_add->U.I.SrcReg[0].Negate = inst_add->U.I.SrcReg[0].Negate ^ RC_MASK_XYZW;
/* This negates the whole expresion: */
if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GREATER) {
@@ -228,16 +232,19 @@ int radeonTransformTEX(
fail = 1;
}
- inst_cmp = rc_insert_new_instruction(c, inst_mad);
+ inst_cmp = rc_insert_new_instruction(c, inst_add);
inst_cmp->U.I.Opcode = RC_OPCODE_CMP;
inst_cmp->U.I.DstReg = output_reg;
inst_cmp->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
inst_cmp->U.I.SrcReg[0].Index = tmp_sum;
+ inst_cmp->U.I.SrcReg[0].Swizzle =
+ combine_swizzles(RC_SWIZZLE_WWWW,
+ compiler->state.unit[inst->U.I.TexSrcUnit].depth_texture_swizzle);
inst_cmp->U.I.SrcReg[pass].File = RC_FILE_NONE;
inst_cmp->U.I.SrcReg[pass].Swizzle = RC_SWIZZLE_1111;
inst_cmp->U.I.SrcReg[fail] = shadow_ambient(compiler, inst->U.I.TexSrcUnit);
- assert(tmp_texsample != tmp_sum && tmp_sum != tmp_recip_w);
+ assert(tmp_texsample != tmp_sum);
}
}
More information about the mesa-commit
mailing list