#include "radeon_program_tex.h"
-#include "../r300_reg.h"
-
/* Series of transformations to be done on textures. */
-static struct rc_src_register shadow_ambient(struct radeon_compiler * c, int tmu)
+static struct rc_src_register shadow_ambient(struct r300_fragment_program_compiler *compiler,
+ int tmu)
{
struct rc_src_register reg = { 0, };
- reg.File = RC_FILE_CONSTANT;
- reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu);
- reg.Swizzle = RC_SWIZZLE_WWWW;
+ if (compiler->enable_shadow_ambient) {
+ reg.File = RC_FILE_CONSTANT;
+ reg.Index = rc_constants_add_state(&compiler->Base.Program.Constants,
+ RC_STATE_SHADOW_AMBIENT, tmu);
+ reg.Swizzle = RC_SWIZZLE_WWWW;
+ } else {
+ reg.File = RC_FILE_NONE;
+ reg.Swizzle = RC_SWIZZLE_0000;
+ }
return reg;
}
+static void lower_texture_rect(struct r300_fragment_program_compiler *compiler,
+ struct rc_instruction *inst)
+{
+ struct rc_instruction *inst_rect;
+ unsigned temp = rc_find_free_temporary(&compiler->Base);
+
+ if (inst->U.I.TexSrcTarget == RC_TEXTURE_RECT ||
+ compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords) {
+ inst_rect = rc_insert_new_instruction(&compiler->Base, inst->Prev);
+
+ inst_rect->U.I.Opcode = RC_OPCODE_MUL;
+ inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_rect->U.I.DstReg.Index = temp;
+ inst_rect->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+ inst_rect->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
+ inst_rect->U.I.SrcReg[1].Index =
+ rc_constants_add_state(&compiler->Base.Program.Constants,
+ RC_STATE_R300_TEXRECT_FACTOR, inst->U.I.TexSrcUnit);
+
+ reset_srcreg(&inst->U.I.SrcReg[0]);
+ inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+ inst->U.I.SrcReg[0].Index = temp;
+
+ inst->U.I.TexSrcTarget = RC_TEXTURE_2D;
+ }
+}
+
/**
* Transform TEX, TXP, TXB, and KIL instructions in the following ways:
* - implement texture compare (shadow extensions)
/* ARB_shadow & EXT_shadow_funcs */
if (inst->U.I.Opcode != RC_OPCODE_KIL &&
- c->Program.ShadowSamplers & (1 << inst->U.I.TexSrcUnit)) {
+ ((c->Program.ShadowSamplers & (1 << inst->U.I.TexSrcUnit)) ||
+ (compiler->state.unit[inst->U.I.TexSrcUnit].compare_mode_enabled))) {
rc_compare_func comparefunc = compiler->state.unit[inst->U.I.TexSrcUnit].texture_compare_func;
if (comparefunc == RC_COMPARE_FUNC_NEVER || comparefunc == RC_COMPARE_FUNC_ALWAYS) {
inst->U.I.SrcReg[0].File = RC_FILE_NONE;
inst->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111;
} else {
- inst->U.I.SrcReg[0] = shadow_ambient(c, inst->U.I.TexSrcUnit);
+ inst->U.I.SrcReg[0] = shadow_ambient(compiler, inst->U.I.TexSrcUnit);
}
return 1;
} else {
rc_compare_func comparefunc = compiler->state.unit[inst->U.I.TexSrcUnit].texture_compare_func;
- unsigned int depthmode = compiler->state.unit[inst->U.I.TexSrcUnit].depth_texture_mode;
- struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst);
- struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp);
- struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad);
- int pass, fail;
-
- inst_rcp->U.I.Opcode = RC_OPCODE_RCP;
- inst_rcp->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rcp->U.I.DstReg.Index = rc_find_free_temporary(c);
- 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_SWIZZLE_WWWW;
-
- inst_cmp->U.I.DstReg = inst->U.I.DstReg;
+ struct rc_instruction * inst_rcp = NULL;
+ struct rc_instruction * inst_mad;
+ struct rc_instruction * inst_cmp;
+ unsigned tmp_texsample = rc_find_free_temporary(c);
+ unsigned tmp_sum = rc_find_free_temporary(c);
+ unsigned tmp_recip_w = 0;
+ int pass, fail, tex;
+
+ /* Save the output register. */
+ struct rc_dst_register output_reg = inst->U.I.DstReg;
+
+ /* Redirect TEX to a new temp. */
inst->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst->U.I.DstReg.Index = rc_find_free_temporary(c);
+ inst->U.I.DstReg.Index = tmp_texsample;
inst->U.I.DstReg.WriteMask = RC_MASK_XYZW;
- inst_mad->U.I.Opcode = RC_OPCODE_MAD;
+ if (inst->U.I.Opcode == RC_OPCODE_TXP) {
+ tmp_recip_w = rc_find_free_temporary(c);
+
+ /* 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.WriteMask = RC_MASK_W;
+ inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+ inst_rcp->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW;
+ }
+
+ /* Perspective-divide r by W (if it's TXP) and add the texture sample (see below). */
+ 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 = rc_find_free_temporary(c);
+ 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 = RC_SWIZZLE_ZZZZ;
- inst_mad->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
- inst_mad->U.I.SrcReg[1].Index = inst_rcp->U.I.DstReg.Index;
- inst_mad->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW;
- inst_mad->U.I.SrcReg[2].File = RC_FILE_TEMPORARY;
- inst_mad->U.I.SrcReg[2].Index = inst->U.I.DstReg.Index;
- if (depthmode == 0) /* GL_LUMINANCE */
- inst_mad->U.I.SrcReg[2].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_Z);
- else if (depthmode == 2) /* GL_ALPHA */
- inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_WWWW;
-
- /* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
- * r < tex <=> -tex+r < 0
- * r >= tex <=> not (-tex+r < 0 */
+ 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;
+ }
+
+ /* Recall that SrcReg[0] is r, SrcReg[tex] is tex and:
+ * LESS: r < tex <=> -tex+r < 0
+ * GEQUAL: r >= tex <=> not (-tex+r < 0)
+ * GREATER: r > tex <=> tex-r < 0
+ * LEQUAL: r <= tex <=> not ( tex-r < 0)
+ *
+ * This negates either r or tex: */
if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GEQUAL)
- inst_mad->U.I.SrcReg[2].Negate = inst_mad->U.I.SrcReg[2].Negate ^ RC_MASK_XYZW;
+ inst_mad->U.I.SrcReg[tex].Negate = inst_mad->U.I.SrcReg[tex].Negate ^ RC_MASK_XYZW;
else
inst_mad->U.I.SrcReg[0].Negate = inst_mad->U.I.SrcReg[0].Negate ^ RC_MASK_XYZW;
- inst_cmp->U.I.Opcode = RC_OPCODE_CMP;
- /* DstReg has been filled out above */
- inst_cmp->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
- inst_cmp->U.I.SrcReg[0].Index = inst_mad->U.I.DstReg.Index;
-
+ /* This negates the whole expresion: */
if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GREATER) {
pass = 1;
fail = 2;
fail = 1;
}
+ inst_cmp = rc_insert_new_instruction(c, inst_mad);
+ 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[pass].File = RC_FILE_NONE;
inst_cmp->U.I.SrcReg[pass].Swizzle = RC_SWIZZLE_1111;
- inst_cmp->U.I.SrcReg[fail] = shadow_ambient(c, inst->U.I.TexSrcUnit);
+ inst_cmp->U.I.SrcReg[fail] = shadow_ambient(compiler, inst->U.I.TexSrcUnit);
}
}
*
* Mirroring is trickier. We're going to start out like repeat:
*
- * MUL temp0, texcoord, <scaling factor constant> ; De-mirror across axes
- * MUL temp0, abs(temp0), 0.5 ; Pattern repeats in [0, 2]
+ * MUL temp, texcoord, <scaling factor constant> ; De-mirror across axes
+ * MUL temp, temp, 0.5 ; Pattern repeats in [0, 2]
* ; so scale to [0, 1]
- * FRC temp0, temp0 ; Make the pattern repeat
- * SGE temp1, temp0, 0.5 ; Select components that need to be "reflected"
- * ; across the mirror
- * MAD temp0, neg(0.5), temp1, temp0 ; Add -0.5 to the
- * ; selected components
- * ADD temp0, temp0, temp0 ; Poor man's 2x to undo earlier MUL
+ * FRC temp, temp ; Make the pattern repeat
+ * MAD temp, temp, 2, -1 ; Move the pattern to [-1, 1]
+ * ADD temp, 1, -abs(temp) ; Now comes a neat trick: use abs to mirror the pattern.
+ * ; The pattern is backwards, so reverse it (1-x).
*
* This gives us coords in [0, 1].
*
- * ~ C.
+ * ~ C & M. ;)
*/
if (inst->U.I.Opcode != RC_OPCODE_KIL &&
(inst->U.I.TexSrcTarget == RC_TEXTURE_RECT ||
compiler->state.unit[inst->U.I.TexSrcUnit].fake_npot ||
compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords)) {
rc_wrap_mode wrapmode = compiler->state.unit[inst->U.I.TexSrcUnit].wrap_mode;
- struct rc_instruction *inst_rect = NULL;
- unsigned temp = rc_find_free_temporary(c);
-
- if (inst->U.I.TexSrcTarget == RC_TEXTURE_RECT ||
- compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords) {
- inst_rect = rc_insert_new_instruction(c, inst->Prev);
-
- inst_rect->U.I.Opcode = RC_OPCODE_MUL;
- inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rect->U.I.DstReg.Index = temp;
- inst_rect->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
- inst_rect->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
- inst_rect->U.I.SrcReg[1].Index =
- rc_constants_add_state(&c->Program.Constants,
- RC_STATE_R300_TEXRECT_FACTOR, inst->U.I.TexSrcUnit);
-
- reset_srcreg(&inst->U.I.SrcReg[0]);
- inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
- inst->U.I.SrcReg[0].Index = temp;
- inst->U.I.TexSrcTarget = RC_TEXTURE_2D;
+ /* R300 cannot sample from rectangles. */
+ if (!c->is_r500) {
+ lower_texture_rect(compiler, inst);
}
if (compiler->state.unit[inst->U.I.TexSrcUnit].fake_npot &&
wrapmode != RC_WRAP_NONE) {
struct rc_instruction *inst_mov;
+ unsigned temp = rc_find_free_temporary(c);
+
+ /* For NPOT fallback, we need normalized coordinates anyway. */
+ if (c->is_r500) {
+ lower_texture_rect(compiler, inst);
+ }
if (wrapmode == RC_WRAP_REPEAT) {
/* Both instructions will be paired up. */
inst_frc->U.I.DstReg.Index = temp;
inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ;
inst_frc->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
- } else if (wrapmode == RC_WRAP_MIRROR) {
- unsigned temp1;
+ } else if (wrapmode == RC_WRAP_MIRRORED_REPEAT) {
/*
- * MUL temp0, abs(temp0), 0.5
- * FRC temp0, temp0
- * SGE temp1, temp0, 0.5
- * MAD temp0, neg(0.5), temp1, temp0
- * ADD temp0, temp0, temp0
+ * Function:
+ * f(v) = 1 - abs(frac(v * 0.5) * 2 - 1)
+ *
+ * Code:
+ * MUL temp, src0, 0.5
+ * FRC temp, temp
+ * MAD temp, temp, 2, -1
+ * ADD temp, 1, -abs(temp)
*/
- inst_rect = rc_insert_new_instruction(c, inst->Prev);
-
- inst_rect->U.I.Opcode = RC_OPCODE_MUL;
- inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rect->U.I.DstReg.Index = temp;
- inst_rect->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
- inst_rect->U.I.SrcReg[1].Swizzle = RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_HALF);
-
- inst_rect = rc_insert_new_instruction(c, inst->Prev);
-
- inst_rect->U.I.Opcode = RC_OPCODE_FRC;
- inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rect->U.I.DstReg.Index = temp;
- inst_rect->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
- inst_rect->U.I.SrcReg[0].Index = temp;
-
- temp1 = rc_find_free_temporary(c);
- inst_rect = rc_insert_new_instruction(c, inst->Prev);
-
- inst_rect->U.I.Opcode = RC_OPCODE_SGE;
- inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rect->U.I.DstReg.Index = temp1;
- inst_rect->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
- inst_rect->U.I.SrcReg[0].Index = temp;
- inst_rect->U.I.SrcReg[1].Swizzle = RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_HALF);
-
- inst_rect = rc_insert_new_instruction(c, inst->Prev);
-
- inst_rect->U.I.Opcode = RC_OPCODE_MAD;
- inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rect->U.I.DstReg.Index = temp;
- inst_rect->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
- inst_rect->U.I.SrcReg[0].Index = temp1;
- inst_rect->U.I.SrcReg[1].Swizzle = RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_HALF);
- inst_rect->U.I.SrcReg[1].Negate = 1;
- inst_rect->U.I.SrcReg[2].File = RC_FILE_TEMPORARY;
- inst_rect->U.I.SrcReg[2].Index = temp;
-
- inst_rect = rc_insert_new_instruction(c, inst->Prev);
-
- inst_rect->U.I.Opcode = RC_OPCODE_ADD;
- inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
- inst_rect->U.I.DstReg.Index = temp;
- inst_rect->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
- inst_rect->U.I.SrcReg[0].Index = temp;
- inst_rect->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
- inst_rect->U.I.SrcReg[1].Index = temp;
+ struct rc_instruction *inst_mul, *inst_frc, *inst_mad, *inst_add;
+ unsigned two, two_swizzle;
+
+ inst_mul = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_mul->U.I.Opcode = RC_OPCODE_MUL;
+ inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_mul->U.I.DstReg.Index = temp;
+ inst_mul->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+ inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+ inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_HHHH;
+
+ inst_frc = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_frc->U.I.Opcode = RC_OPCODE_FRC;
+ inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_frc->U.I.DstReg.Index = temp;
+ inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+ inst_frc->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+ inst_frc->U.I.SrcReg[0].Index = temp;
+ inst_frc->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0;
+
+ two = rc_constants_add_immediate_scalar(&c->Program.Constants, 2, &two_swizzle);
+ inst_mad = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_mad->U.I.Opcode = RC_OPCODE_MAD;
+ inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_mad->U.I.DstReg.Index = temp;
+ inst_mad->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+ inst_mad->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+ inst_mad->U.I.SrcReg[0].Index = temp;
+ inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0;
+ inst_mad->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
+ inst_mad->U.I.SrcReg[1].Index = two;
+ inst_mad->U.I.SrcReg[1].Swizzle = two_swizzle;
+ inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_1111;
+ inst_mad->U.I.SrcReg[2].Negate = RC_MASK_XYZ;
+
+ inst_add = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_add->U.I.Opcode = RC_OPCODE_ADD;
+ inst_add->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_add->U.I.DstReg.Index = temp;
+ inst_add->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+ inst_add->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111;
+ inst_add->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
+ inst_add->U.I.SrcReg[1].Index = temp;
+ inst_add->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZ0;
+ inst_add->U.I.SrcReg[1].Abs = 1;
+ inst_add->U.I.SrcReg[1].Negate = RC_MASK_XYZ;
+ } else if (wrapmode == RC_WRAP_MIRRORED_CLAMP) {
+ /*
+ * Mirrored clamp modes are bloody simple, we just use abs
+ * to mirror [0, 1] into [-1, 0]. This works for
+ * all modes i.e. CLAMP, CLAMP_TO_EDGE, and CLAMP_TO_BORDER.
+ */
+ struct rc_instruction *inst_mov;
+
+ inst_mov = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_mov->U.I.Opcode = RC_OPCODE_MOV;
+ inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
+ inst_mov->U.I.DstReg.Index = temp;
+ inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+ inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+ inst_mov->U.I.SrcReg[0].Abs = 1;
}
/* Preserve W for TXP/TXB. */
}
}
- /* Cannot write texture to output registers or with masks */
+ /* Cannot write texture to output registers (all chips) or with masks (non-r500) */
if (inst->U.I.Opcode != RC_OPCODE_KIL &&
- (inst->U.I.DstReg.File != RC_FILE_TEMPORARY || inst->U.I.DstReg.WriteMask != RC_MASK_XYZW)) {
+ (inst->U.I.DstReg.File != RC_FILE_TEMPORARY ||
+ (!c->is_r500 && inst->U.I.DstReg.WriteMask != RC_MASK_XYZW))) {
struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst);
inst_mov->U.I.Opcode = RC_OPCODE_MOV;