/* Writemasks */
#define R500_WRITEMASK_G 0x2
+#define R500_WRITEMASK_RGB 0x7
#define R500_WRITEMASK_A 0x8
#define R500_WRITEMASK_AR 0x9
#define R500_WRITEMASK_AG 0xA
return swiz;
}
+static inline GLuint make_rgba_swizzle(GLuint src) {
+ GLuint swiz = 0x0;
+ GLuint temp;
+ int i;
+ for (i = 0; i < 4; i++) {
+ temp = GET_SWZ(src, i);
+ /* Fix SWIZZLE_ONE */
+ if (temp == 5) temp++;
+ swiz |= temp << i*3;
+ }
+ return swiz;
+}
+
static inline GLuint make_alpha_swizzle(struct prog_src_register src) {
GLuint swiz = GET_SWZ(src.Swizzle, 3);
COMPILE_STATE;
- int r = cs->temp_in_use + 1 + slot;
+ int r = fp->temp_reg_offset + cs->temp_in_use + slot;
if (r > R500_US_NUM_TEMP_REGS) {
ERROR("Too many temporary registers requested, can't compile!\n");
fp->inst[counter].inst5 = 0x0;
}
-static void emit_mov(struct r500_fragment_program *fp, int counter, struct prog_src_register src, GLuint dest) {
+static void emit_mov(struct r500_fragment_program *fp, int counter, struct prog_instruction *fpi, GLuint src_reg, GLuint swizzle, GLuint dest) {
/* The r3xx shader uses MAD to implement MOV. We are using CMP, since
* it is technically more accurate and recommended by ATI/AMD. */
- GLuint src_reg = make_src(fp, src);
+ emit_alu(fp, counter, fpi);
fp->inst[counter].inst1 = R500_RGB_ADDR0(src_reg);
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src_reg);
+ /* (De)mangle the swizzle from Mesa to R500. */
+ swizzle = make_rgba_swizzle(swizzle);
+ /* 0x1FF is 9 bits, size of an RGB swizzle. */
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
- | MAKE_SWIZ_RGB_A(make_rgb_swizzle(src))
+ | MAKE_SWIZ_RGB_A((swizzle & 0x1ff))
| R500_ALU_RGB_SEL_B_SRC0
- | MAKE_SWIZ_RGB_B(make_rgb_swizzle(src))
+ | MAKE_SWIZ_RGB_B((swizzle & 0x1ff))
| R500_ALU_RGB_OMOD_DISABLE;
fp->inst[counter].inst4 |= R500_ALPHA_OP_CMP
| R500_ALPHA_ADDRD(dest)
- | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(src))
- | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(src))
+ | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(GET_SWZ(swizzle, 3))
+ | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(GET_SWZ(swizzle, 3))
| R500_ALPHA_OMOD_DISABLE;
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
| R500_ALU_RGBA_ADDRD(dest)
switch (fpi->Opcode) {
case OPCODE_ABS:
- emit_alu(fp, counter, fpi);
- emit_mov(fp, counter, fpi->SrcReg[0], dest);
+ emit_mov(fp, counter, fpi, make_src(fp, fpi->SrcReg[0]), fpi->SrcReg[0].Swizzle, dest);
fp->inst[counter].inst3 |= R500_ALU_RGB_MOD_A_ABS
| R500_ALU_RGB_MOD_B_ABS;
fp->inst[counter].inst4 |= R500_ALPHA_MOD_A_ABS
counter++;
/* Second instruction */
fp->inst[counter].inst0 = R500_INST_TYPE_ALU | (R500_WRITEMASK_AB << 11);
- fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 0));
- fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 0)) | R500_ALPHA_ADDR1(src[1]);
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 0)) | R500_RGB_ADDR1(src[1]);
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 0));
/* Select [z, z, z, y] */
temp_swiz = 2 | (2 << 3) | (2 << 6);
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ONE);
fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
| R500_ALPHA_ADDRD(get_temp(fp, 1))
- | R500_ALPHA_SEL_A_SRC0 | R500_ALPHA_SWIZ_A_B
- | R500_ALPHA_SEL_B_SRC0 | R500_ALPHA_SWIZ_B_A;
+ | R500_ALPHA_SEL_A_SRC0 | R500_ALPHA_SWIZ_A_A
+ | R500_ALPHA_SEL_B_SRC0 | R500_ALPHA_SWIZ_B_B;
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
| R500_ALU_RGBA_ADDRD(get_temp(fp, 1))
| MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
| MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
counter++;
/* Final instruction */
- emit_alu(fp, counter, fpi);
- fpi->SrcReg[0].Index = get_temp(fp, 0);
- fpi->SrcReg[0].Swizzle = 1672;
- emit_mov(fp, counter, fpi->SrcReg[0], dest);
+ emit_mov(fp, counter, fpi, get_temp(fp, 0), SWIZZLE_NOOP, dest);
break;
case OPCODE_LRP:
/* src0 * src1 + INV(src0) * src2
| R500_ALU_RGBA_ADDRD(dest);
break;
case OPCODE_MOV:
- emit_alu(fp, counter, fpi);
- emit_mov(fp, counter, fpi->SrcReg[0], dest);
+ emit_mov(fp, counter, fpi, make_src(fp, fpi->SrcReg[0]), fpi->SrcReg[0].Swizzle, dest);
break;
case OPCODE_MUL:
/* Variation on MAD: src0*src1+0 */
break;
case OPCODE_SWZ:
/* TODO: The rarer negation masks! */
+ emit_mov(fp, counter, fpi, make_src(fp, fpi->SrcReg[0]), fpi->SrcReg[0].Swizzle, dest);
+ break;
+ case OPCODE_XPD:
+ /* src0 * src1 - src1 * src0
+ * 1) MUL temp.xyz, src0.yzx, src1.zxy
+ * 2) MAD src0.zxy, src1.yzx, -temp.xyz */
+ src[0] = make_src(fp, fpi->SrcReg[0]);
+ src[1] = make_src(fp, fpi->SrcReg[1]);
+ fp->inst[counter].inst0 = R500_INST_TYPE_ALU | R500_INST_TEX_SEM_WAIT
+ | (R500_WRITEMASK_RGB << 11);
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
+ | R500_RGB_ADDR1(src[1]);
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
+ | R500_ALPHA_ADDR1(src[1]);
+ /* Select [y, z, x] */
+ temp_swiz = make_rgb_swizzle(fpi->SrcReg[0]);
+ temp_swiz = (GET_SWZ(temp_swiz, 1) << 0) | (GET_SWZ(temp_swiz, 2) << 3) | (GET_SWZ(temp_swiz, 0) << 6);
+ fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(temp_swiz);
+ /* Select [z, x, y] */
+ temp_swiz = make_rgb_swizzle(fpi->SrcReg[1]);
+ temp_swiz = (GET_SWZ(temp_swiz, 2) << 0) | (GET_SWZ(temp_swiz, 0) << 3) | (GET_SWZ(temp_swiz, 1) << 6);
+ fp->inst[counter].inst3 |= R500_ALU_RGB_SEL_B_SRC1
+ | MAKE_SWIZ_RGB_B(temp_swiz);
+ fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
+ | R500_ALPHA_ADDRD(get_temp(fp, 0))
+ | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
+ | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
+ | R500_ALU_RGBA_ADDRD(get_temp(fp, 0))
+ | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
+ | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
+ counter++;
emit_alu(fp, counter, fpi);
- emit_mov(fp, counter, fpi->SrcReg[0], dest);
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
+ | R500_RGB_ADDR1(src[1])
+ | R500_RGB_ADDR2(get_temp(fp, 0));
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
+ | R500_ALPHA_ADDR1(src[1])
+ | R500_ALPHA_ADDR2(get_temp(fp, 0));
+ /* Select [z, x, y] */
+ temp_swiz = make_rgb_swizzle(fpi->SrcReg[0]);
+ temp_swiz = (GET_SWZ(temp_swiz, 2) << 0) | (GET_SWZ(temp_swiz, 0) << 3) | (GET_SWZ(temp_swiz, 1) << 6);
+ fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(temp_swiz);
+ /* Select [y, z, x] */
+ temp_swiz = make_rgb_swizzle(fpi->SrcReg[1]);
+ temp_swiz = (GET_SWZ(temp_swiz, 1) << 0) | (GET_SWZ(temp_swiz, 2) << 3) | (GET_SWZ(temp_swiz, 0) << 6);
+ fp->inst[counter].inst3 |= R500_ALU_RGB_SEL_B_SRC1
+ | MAKE_SWIZ_RGB_B(temp_swiz);
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MAD
+ | R500_ALPHA_ADDRD(dest)
+ | R500_ALPHA_SWIZ_A_1
+ | R500_ALPHA_SWIZ_B_1;
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
+ | R500_ALU_RGBA_ADDRD(dest)
+ | R500_ALU_RGBA_SEL_C_SRC2
+ | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_RGB)
+ | R500_ALU_RGBA_MOD_C_NEG
+ | R500_ALU_RGBA_A_SWIZ_0;
break;
case OPCODE_KIL:
case OPCODE_TEX:
fp->translated = GL_FALSE;
fp->error = GL_FALSE;
fp->cs = cs = &(R300_CONTEXT(fp->ctx)->state.pfs_compile);
- fp->cur_node = 0;
- fp->first_node_has_tex = 0;
fp->const_nr = 0;
/* Size of pixel stack, plus 1. */
fp->max_temp_idx = 1;
/* Temp register offset. */
fp->temp_reg_offset = 0;
- fp->node[0].alu_end = -1;
- fp->node[0].tex_end = -1;
_mesa_memset(cs, 0, sizeof(*fp->cs));
for (i = 0; i < PFS_MAX_ALU_INST; i++) {
for (fpi = mp->Base.Instructions; fpi->Opcode != OPCODE_END; fpi++) {
for (i = 0; i < 3; i++) {
if (fpi->SrcReg[i].File == PROGRAM_TEMPORARY) {
- if (fpi->SrcReg[i].Index > temps_used)
- temps_used = fpi->SrcReg[i].Index;
+ if (fpi->SrcReg[i].Index >= temps_used)
+ temps_used = fpi->SrcReg[i].Index + 1;
}
}
}
- cs->temp_in_use = temps_used;
+ cs->temp_in_use = temps_used + 1;
- fp->max_temp_idx = fp->temp_reg_offset + cs->temp_in_use + 1;
+ fp->max_temp_idx = fp->temp_reg_offset + cs->temp_in_use;
+
+ if (RADEON_DEBUG & DEBUG_PIXEL)
+ fprintf(stderr, "FP temp indices: fp->max_temp_idx: %d cs->temp_in_use: %d\n", fp->max_temp_idx, cs->temp_in_use);
}
static void update_params(struct r500_fragment_program *fp)
static char *toop(int op_val)
{
- char *str;
+ char *str = NULL;
switch (op_val) {
case 0: str = "MAD"; break;
case 1: str = "DP3"; break;
static void dump_program(struct r500_fragment_program *fp)
{
- int pc = 0;
+
+ fprintf(stderr, "R500 Fragment Program:\n--------\n");
+
int n;
uint32_t inst;
uint32_t inst0;
char *str = NULL;
+ if (fp->const_nr) {
+ fprintf(stderr, "--------\nConstants:\n");
+ for (n = 0; n < fp->const_nr; n++) {
+ fprintf(stderr, "Constant %d: %f %f\n\t %f %f\n", n,
+ fp->constant[n][0], fp->constant[n][1], fp->constant[n][2],
+ fp->constant[n][3]);
+ }
+ fprintf(stderr, "--------\n");
+ }
+
for (n = 0; n < fp->inst_end+1; n++) {
inst0 = inst = fp->inst[n].inst0;
fprintf(stderr,"%d\t0:CMN_INST 0x%08x:", n, inst);
projects / mesa.git / blobdiff