/*
* Copyright (C) 2005 Ben Skeggs.
*
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Adaptation and modification for ATI/AMD Radeon R500 GPU chipsets.
+ *
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining
#define R500_SWIZ_RGB_ZERO ((4 << 0) | (4 << 3) | (4 << 6))
#define R500_SWIZ_RGB_ONE ((6 << 0) | (6 << 3) | (6 << 6))
#define R500_SWIZ_RGB_RGB ((0 << 0) | (1 << 3) | (2 << 6))
+#define R500_SWIZ_MOD_NEG 1
+#define R500_SWIZ_MOD_ABS 2
+#define R500_SWIZ_MOD_NEG_ABS 3
/* Swizzles for inst2 */
#define MAKE_SWIZ_TEX_STRQ(x) (x << 8)
#define MAKE_SWIZ_TEX_RGBA(x) (x << 24)
#define MAKE_SWIZ_ALPHA_C(x) (x << 27)
/* 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
+#define R500_WRITEMASK_ARG 0xB
+#define R500_WRITEMASK_AB 0xC
#define R500_WRITEMASK_ARGB 0xF
/* 1/(2pi), needed for quick modulus in trig insts
0.15915494309189535,
0.15915494309189535};
+static const GLfloat LIT[] = {127.999999,
+ 127.999999,
+ 127.999999,
+ -127.999999};
+
static void dump_program(struct r500_fragment_program *fp);
static inline GLuint make_rgb_swizzle(struct prog_src_register src) {
temp = GET_SWZ(src.Swizzle, i);
/* Fix SWIZZLE_ONE */
if (temp == 5) temp++;
- swiz += temp << i*3;
+ swiz |= temp << i*3;
+ }
+ if (src.NegateBase)
+ swiz |= (R500_SWIZ_MOD_NEG << 9);
+ 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;
}
GLuint swiz = GET_SWZ(src.Swizzle, 3);
if (swiz == 5) swiz++;
+
+ if (src.NegateBase)
+ swiz |= (R500_SWIZ_MOD_NEG << 3);
+
return swiz;
}
}
static inline GLuint make_strq_swizzle(struct prog_src_register src) {
- GLuint swiz = 0x0;
- GLuint temp = src.Swizzle;
+ GLuint swiz = 0x0, temp = 0x0;
int i;
for (i = 0; i < 4; i++) {
- swiz += (temp & 0x3) << i*2;
- temp >>= 3;
+ temp = GET_SWZ(src.Swizzle, i) & 0x3;
+ swiz |= temp << i*2;
}
return swiz;
}
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");
}
static void emit_tex(struct r500_fragment_program *fp,
- struct prog_instruction *fpi, int opcode, int dest, int counter)
+ struct prog_instruction *fpi, int dest, int counter)
{
int hwsrc, hwdest;
GLuint mask;
if (fpi->TexSrcTarget == TEXTURE_RECT_INDEX)
fp->inst[counter].inst1 |= R500_TEX_UNSCALED;
- switch (opcode) {
+ switch (fpi->Opcode) {
case OPCODE_KIL:
fp->inst[counter].inst1 |= R500_TEX_INST_TEXKILL;
break;
fp->inst[counter].inst1 |= R500_TEX_INST_PROJ;
break;
default:
- ERROR("emit_tex can't handle opcode %x\n", opcode);
+ ERROR("emit_tex can't handle opcode %x\n", fpi->Opcode);
}
fp->inst[counter].inst2 = R500_TEX_SRC_ADDR(hwsrc)
- /* | MAKE_SWIZ_TEX_STRQ(make_strq_swizzle(fpi->SrcReg[0])) */
- | R500_TEX_SRC_S_SWIZ_R | R500_TEX_SRC_T_SWIZ_G
- | R500_TEX_SRC_R_SWIZ_B | R500_TEX_SRC_Q_SWIZ_A
+ | MAKE_SWIZ_TEX_STRQ(make_strq_swizzle(fpi->SrcReg[0]))
+ /* | R500_TEX_SRC_S_SWIZ_R | R500_TEX_SRC_T_SWIZ_G
+ | R500_TEX_SRC_R_SWIZ_B | R500_TEX_SRC_Q_SWIZ_A */
| R500_TEX_DST_ADDR(hwdest)
| R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G
| R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A;
static void emit_alu(struct r500_fragment_program *fp, int counter, struct prog_instruction *fpi) {
if (fpi->DstReg.File == PROGRAM_OUTPUT) {
- fp->inst[counter].inst0 = R500_INST_TYPE_OUT
- /* output_mask */
- | (fpi->DstReg.WriteMask << 15);
+ fp->inst[counter].inst0 = R500_INST_TYPE_OUT;
+
+ if (fpi->DstReg.Index == FRAG_RESULT_COLR)
+ fp->inst[counter].inst0 |= (fpi->DstReg.WriteMask << 15);
+
+ if (fpi->DstReg.Index == FRAG_RESULT_DEPR)
+ fp->inst[counter].inst4 = R500_ALPHA_W_OMASK;
} else {
fp->inst[counter].inst0 = R500_INST_TYPE_ALU
/* pixel_mask */
}
fp->inst[counter].inst0 |= R500_INST_TEX_SEM_WAIT;
+
+ /* Ideally, we shouldn't have to explicitly clear memory here! */
+ fp->inst[counter].inst1 = 0x0;
+ fp->inst[counter].inst2 = 0x0;
+ fp->inst[counter].inst3 = 0x0;
+ 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
+ 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)
| MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
}
+static void emit_mad(struct r500_fragment_program *fp, int counter, struct prog_instruction *fpi, int one, int two, int three) {
+ /* Note: This code was all Corbin's. Corbin is a rather hackish coder.
+ * If you can make it pretty or fast, please do so! */
+ emit_alu(fp, counter, fpi);
+ /* Common MAD stuff */
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MAD
+ | R500_ALPHA_ADDRD(make_dest(fp, fpi->DstReg));
+ fp->inst[counter].inst5 |= R500_ALU_RGBA_OP_MAD
+ | R500_ALU_RGBA_ADDRD(make_dest(fp, fpi->DstReg));
+ switch (one) {
+ case 0:
+ case 1:
+ case 2:
+ fp->inst[counter].inst1 |= R500_RGB_ADDR0(make_src(fp, fpi->SrcReg[one]));
+ fp->inst[counter].inst2 |= R500_ALPHA_ADDR0(make_src(fp, fpi->SrcReg[one]));
+ fp->inst[counter].inst3 |= R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[one]));
+ fp->inst[counter].inst4 |= R500_ALPHA_SEL_A_SRC0
+ | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[one]));
+ break;
+ case R500_SWIZZLE_ZERO:
+ fp->inst[counter].inst3 |= MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ZERO);
+ fp->inst[counter].inst4 |= MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ZERO);
+ break;
+ case R500_SWIZZLE_ONE:
+ fp->inst[counter].inst3 |= MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE);
+ fp->inst[counter].inst4 |= MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE);
+ break;
+ default:
+ ERROR("Bad src index in emit_mad: %d\n", one);
+ break;
+ }
+ switch (two) {
+ case 0:
+ case 1:
+ case 2:
+ fp->inst[counter].inst1 |= R500_RGB_ADDR1(make_src(fp, fpi->SrcReg[two]));
+ fp->inst[counter].inst2 |= R500_ALPHA_ADDR1(make_src(fp, fpi->SrcReg[two]));
+ fp->inst[counter].inst3 |= R500_ALU_RGB_SEL_B_SRC1
+ | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[two]));
+ fp->inst[counter].inst4 |= R500_ALPHA_SEL_B_SRC1
+ | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[two]));
+ break;
+ case R500_SWIZZLE_ZERO:
+ fp->inst[counter].inst3 |= MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ZERO);
+ fp->inst[counter].inst4 |= MAKE_SWIZ_ALPHA_B(R500_SWIZZLE_ZERO);
+ break;
+ case R500_SWIZZLE_ONE:
+ fp->inst[counter].inst3 |= MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ONE);
+ fp->inst[counter].inst4 |= MAKE_SWIZ_ALPHA_B(R500_SWIZZLE_ONE);
+ break;
+ default:
+ ERROR("Bad src index in emit_mad: %d\n", two);
+ break;
+ }
+ switch (three) {
+ case 0:
+ case 1:
+ case 2:
+ fp->inst[counter].inst1 |= R500_RGB_ADDR2(make_src(fp, fpi->SrcReg[three]));
+ fp->inst[counter].inst2 |= R500_ALPHA_ADDR2(make_src(fp, fpi->SrcReg[three]));
+ fp->inst[counter].inst5 |= R500_ALU_RGBA_SEL_C_SRC2
+ | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[three]))
+ | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
+ | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[three]));
+ break;
+ case R500_SWIZZLE_ZERO:
+ fp->inst[counter].inst5 |= MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
+ | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
+ break;
+ case R500_SWIZZLE_ONE:
+ fp->inst[counter].inst5 |= MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ONE)
+ | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ONE);
+ break;
+ default:
+ ERROR("Bad src index in emit_mad: %d\n", three);
+ break;
+ }
+}
+
static GLboolean parse_program(struct r500_fragment_program *fp)
{
struct gl_fragment_program *mp = &fp->mesa_program;
const struct prog_instruction *inst = mp->Base.Instructions;
struct prog_instruction *fpi;
- GLuint src[3], dest, temp[2];
- int temp_swiz, pixel_mask = 0, output_mask = 0, counter = 0;
+ GLuint src[3], dest = 0;
+ int temp_swiz, counter = 0;
if (!inst || inst[0].Opcode == OPCODE_END) {
ERROR("The program is empty!\n");
if (fpi->Opcode != OPCODE_KIL) {
dest = make_dest(fp, fpi->DstReg);
-
- pixel_mask = fpi->DstReg.WriteMask << 11;
- output_mask = fpi->DstReg.WriteMask << 15;
}
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
| R500_ALPHA_MOD_B_ABS;
break;
case OPCODE_ADD:
- src[0] = make_src(fp, fpi->SrcReg[0]);
- src[1] = make_src(fp, fpi->SrcReg[1]);
/* Variation on MAD: 1*src0+src1 */
- emit_alu(fp, counter, fpi);
- fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
- | R500_RGB_ADDR1(src[1]) | R500_RGB_ADDR2(0);
- fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
- | R500_ALPHA_ADDR1(src[1]) | R500_ALPHA_ADDR2(0);
- fp->inst[counter].inst3 = /* 1 */
- MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
- | R500_ALU_RGB_SEL_B_SRC0 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
- | R500_ALPHA_ADDRD(dest)
- | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
- | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
- | R500_ALU_RGBA_ADDRD(dest)
- | R500_ALU_RGBA_SEL_C_SRC1
- | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[1]))
- | R500_ALU_RGBA_ALPHA_SEL_C_SRC1
- | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[1]));
+ emit_mad(fp, counter, fpi, R500_SWIZZLE_ONE, 0, 1);
break;
case OPCODE_CMP:
/* This inst's selects need to be swapped as follows:
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[2]))
| R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_CMP
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_CMP
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[2]))
| R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0;
- fp->inst[counter].inst4 = R500_ALPHA_OP_COS
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_COS
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
| R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_DP
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_DP
| R500_ALPHA_ADDRD(dest)
| 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].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
| R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_DP
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_DP
| R500_ALPHA_ADDRD(dest)
| 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].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
| R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_DP
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_DP
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
| R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
/* Select [1, y, 1, w] */
temp_swiz = (make_rgb_swizzle(fpi->SrcReg[0]) & ~0x1c7) | R500_SWIZZLE_ONE | (R500_SWIZZLE_ONE << 6);
fp->inst[counter].inst3 |= MAKE_SWIZ_RGB_B(temp_swiz);
- fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MAD
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
| R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[1]));
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]);
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_EX2
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_EX2
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]);
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_FRC
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_FRC
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_FRC
| R500_ALU_RGBA_ADDRD(dest);
break;
- case OPCODE_KIL:
- emit_tex(fp, fpi, OPCODE_KIL, dest, counter);
- break;
case OPCODE_LG2:
src[0] = make_src(fp, fpi->SrcReg[0]);
emit_alu(fp, counter, fpi);
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]);
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_LN2
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_LN2
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
| R500_ALU_RGBA_ADDRD(dest);
break;
+ case OPCODE_LIT:
+ /* To be honest, I have no idea how I came up with the following.
+ * All I know is that it's based on the r3xx stuff, and was
+ * concieved with the help of NyQuil. Mmm, MyQuil. */
+
+ /* First instruction */
+ src[0] = make_src(fp, fpi->SrcReg[0]);
+ src[1] = emit_const4fv(fp, LIT);
+ fp->inst[counter].inst0 = R500_INST_TYPE_ALU | R500_INST_TEX_SEM_WAIT
+ | (R500_WRITEMASK_ARG << 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]);
+ fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
+ | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ZERO);
+ fp->inst[counter].inst4 = R500_ALPHA_OP_MAX
+ | 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 | R500_ALPHA_SWIZ_B_A;
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAX
+ | R500_ALU_RGBA_ADDRD(get_temp(fp, 0));
+ 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)) | 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_A(temp_swiz)
+ | R500_ALU_RGB_SEL_B_SRC0
+ | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_RGB);
+ fp->inst[counter].inst4 = R500_ALPHA_OP_LN2
+ | R500_ALPHA_ADDRD(get_temp(fp, 0))
+ | R500_ALPHA_SEL_A_SRC0 | R500_ALPHA_SWIZ_A_G;
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MIN
+ | R500_ALU_RGBA_ADDRD(get_temp(fp, 0));
+ counter++;
+ /* Third instruction */
+ fp->inst[counter].inst0 = R500_INST_TYPE_ALU | (R500_WRITEMASK_AG << 11);
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 0));
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 0));
+ /* Select [x, x, x, z] */
+ temp_swiz = 0;
+ fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(temp_swiz)
+ | R500_ALU_RGB_SEL_B_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_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)
+ | R500_ALU_RGBA_A_SWIZ_0;
+ counter++;
+ /* Fourth instruction */
+ fp->inst[counter].inst0 = R500_INST_TYPE_ALU | (R500_WRITEMASK_AR << 11);
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 0));
+ fp->inst[counter].inst3 = MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
+ | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ONE);
+ fp->inst[counter].inst4 = R500_ALPHA_OP_EX2
+ | R500_ALPHA_ADDRD(get_temp(fp, 0))
+ | R500_ALPHA_SEL_A_SRC0 | R500_ALPHA_SWIZ_A_A;
+ 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++;
+ /* Fifth instruction */
+ fp->inst[counter].inst0 = R500_INST_TYPE_ALU | (R500_WRITEMASK_AB << 11);
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 0));
+ /* Select [w, w, w] */
+ temp_swiz = 3 | (3 << 3) | (3 << 6);
+ fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ZERO)
+ | R500_ALU_RGB_SEL_B_SRC0
+ | MAKE_SWIZ_RGB_B(temp_swiz);
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MAD
+ | R500_ALPHA_ADDRD(get_temp(fp, 0))
+ | R500_ALPHA_SWIZ_A_1
+ | R500_ALPHA_SWIZ_B_1;
+ /* Select [-y, -y, -y] */
+ temp_swiz = 1 | (1 << 3) | (1 << 6);
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
+ | R500_ALU_RGBA_ADDRD(get_temp(fp, 0))
+ | MAKE_SWIZ_RGBA_C(temp_swiz)
+ | R500_ALU_RGBA_MOD_C_NEG
+ | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
+ counter++;
+ /* Final instruction */
+ emit_mov(fp, counter, fpi, get_temp(fp, 0), SWIZZLE_NOOP, dest);
+ break;
case OPCODE_LRP:
/* src0 * src1 + INV(src0) * src2
* 1) MUL src0, src1, temp
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRCP
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
| R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_RGB);
- fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MAD
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRCP | MAKE_SWIZ_ALPHA_A(make_alpha_swizzle(fpi->SrcReg[0]))
| R500_ALPHA_SEL_B_SRC1 | R500_ALPHA_SWIZ_B_A;
| MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[2]));
break;
case OPCODE_MAD:
- src[0] = make_src(fp, fpi->SrcReg[0]);
- src[1] = make_src(fp, fpi->SrcReg[1]);
- src[2] = make_src(fp, fpi->SrcReg[2]);
- emit_alu(fp, counter, fpi);
- fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
- | R500_RGB_ADDR1(src[1]) | R500_RGB_ADDR2(src[2]);
- fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
- | R500_ALPHA_ADDR1(src[1]) | R500_ALPHA_ADDR2(src[2]);
- fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
- | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
- | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
- | R500_ALPHA_ADDRD(dest)
- | 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(dest)
- | R500_ALU_RGBA_SEL_C_SRC2
- | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[2]))
- | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
- | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[2]));
+ emit_mad(fp, counter, fpi, 0, 1, 2);
break;
case OPCODE_MAX:
src[0] = make_src(fp, fpi->SrcReg[0]);
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
| R500_ALU_RGB_SEL_B_SRC1
| MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_MAX
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MAX
| R500_ALPHA_ADDRD(dest)
| 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]));
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
| R500_ALU_RGB_SEL_B_SRC1
| MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_MIN
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_MIN
| R500_ALPHA_ADDRD(dest)
| 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]));
| 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:
- src[0] = make_src(fp, fpi->SrcReg[0]);
- src[1] = make_src(fp, fpi->SrcReg[1]);
/* Variation on MAD: src0*src1+0 */
- emit_alu(fp, counter, fpi);
- 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]);
- fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
- | MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]))
- | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[1]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
- | R500_ALPHA_ADDRD(dest)
- | 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(dest)
- // | R500_ALU_RGBA_SEL_C_SRC2
- | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_ZERO)
- // | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
- | MAKE_SWIZ_ALPHA_C(R500_SWIZZLE_ZERO);
+ emit_mad(fp, counter, fpi, 0, 1, R500_SWIZZLE_ZERO);
break;
case OPCODE_POW:
/* POW(a,b) = EX2(LN2(a)*b) */
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_EX2
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_EX2
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]);
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_RCP
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_RCP
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0]);
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_RSQ
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_RSQ
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_COS
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_COS
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(make_rgb_swizzle(fpi->SrcReg[0]));
- fp->inst[counter].inst4 = R500_ALPHA_OP_SIN
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_SIN
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
| R500_ALU_RGBA_ADDRD(dest);
break;
case OPCODE_SGE:
- /* We use SRCP, so as a precaution we're
- * going to set NOP in previous inst, if possible. */
- /* This inst's selects need to be swapped as follows:
- * 0 -> C ; 1 -> B ; 2 -> A */
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_ARGB << 11);
+ fp->inst[counter].inst1 = R500_RGB_ADDR1(src[0])
+ | R500_RGB_ADDR2(src[1]);
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR1(src[0])
+ | R500_ALPHA_ADDR2(src[1]);
+ fp->inst[counter].inst3 = /* 1 */
+ MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
+ | R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[0]));
+ fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
+ | R500_ALPHA_ADDRD(get_temp(fp, 0))
+ | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
+ | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[0]));
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
+ | R500_ALU_RGBA_ADDRD(get_temp(fp, 0))
+ | R500_ALU_RGBA_SEL_C_SRC2
+ | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[1]))
+ | R500_ALU_RGBA_MOD_C_NEG
+ | R500_ALU_RGBA_ALPHA_SEL_C_SRC2
+ | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[1]))
+ | R500_ALU_RGBA_ALPHA_MOD_C_NEG;
+ counter++;
+ /* This inst's selects need to be swapped as follows:
+ * 0 -> C ; 1 -> B ; 2 -> A */
emit_alu(fp, counter, fpi);
- fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
- | R500_RGB_ADDR1(src[1])
- | R500_RGB_SRCP_OP_RGB1_MINUS_RGB0;
- fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
- | R500_ALPHA_ADDR1(src[1])
- | R500_ALPHA_SRCP_OP_A1_MINUS_A0;
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 0));
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 0));
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
| MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
- | R500_ALU_RGB_SEL_B_SRC1
+ | R500_ALU_RGB_SEL_B_SRC0
| MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ZERO);
- fp->inst[counter].inst4 = R500_ALPHA_OP_CMP
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_CMP
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
- | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(R500_SWIZZLE_ZERO);
+ | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(R500_SWIZZLE_ZERO);
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
| R500_ALU_RGBA_ADDRD(dest)
- | R500_ALU_RGBA_SEL_C_SRCP
- | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[0]))
- | R500_ALU_RGBA_ALPHA_SEL_C_SRCP
- | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[0]));
+ | R500_ALU_RGBA_SEL_C_SRC0
+ | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_RGB)
+ | R500_ALU_RGBA_ALPHA_SEL_C_SRC0
+ | R500_ALU_RGBA_A_SWIZ_A;
break;
case OPCODE_SIN:
src[0] = make_src(fp, fpi->SrcReg[0]);
fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 1));
fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0;
- fp->inst[counter].inst4 = R500_ALPHA_OP_SIN
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_SIN
| R500_ALPHA_ADDRD(dest)
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(make_sop_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_SOP
| R500_ALU_RGBA_ADDRD(dest);
break;
case OPCODE_SLT:
- /* We use SRCP, so as a precaution we're
- * going to set NOP in previous inst, if possible. */
- /* This inst's selects need to be swapped as follows:
- * 0 -> C ; 1 -> B ; 2 -> A */
- src[0] = make_src(fp, fpi->SrcReg[0]);
- src[1] = make_src(fp, fpi->SrcReg[1]);
- emit_alu(fp, counter, fpi);
- fp->inst[counter].inst1 = R500_RGB_ADDR0(src[0])
- | R500_RGB_ADDR1(src[1])
- | R500_RGB_SRCP_OP_RGB1_MINUS_RGB0;
- fp->inst[counter].inst2 = R500_ALPHA_ADDR0(src[0])
- | R500_ALPHA_ADDR1(src[1])
- | R500_ALPHA_SRCP_OP_A1_MINUS_A0;
- fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
- | MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ZERO)
- | R500_ALU_RGB_SEL_B_SRC1
- | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ONE);
- fp->inst[counter].inst4 = R500_ALPHA_OP_CMP
- | R500_ALPHA_ADDRD(dest)
- | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ZERO)
- | R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(R500_SWIZZLE_ONE);
- fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
- | R500_ALU_RGBA_ADDRD(dest)
- | R500_ALU_RGBA_SEL_C_SRCP
- | MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[0]))
- | R500_ALU_RGBA_ALPHA_SEL_C_SRCP
- | MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[0]));
- break;
- case OPCODE_SUB:
src[0] = make_src(fp, fpi->SrcReg[0]);
src[1] = make_src(fp, fpi->SrcReg[1]);
- /* Variation on MAD: 1*src0-src1 */
- emit_alu(fp, counter, fpi);
+ fp->inst[counter].inst0 = R500_INST_TYPE_ALU | R500_INST_TEX_SEM_WAIT
+ | (R500_WRITEMASK_ARGB << 11);
fp->inst[counter].inst1 = R500_RGB_ADDR1(src[0])
| R500_RGB_ADDR2(src[1]);
fp->inst[counter].inst2 = R500_ALPHA_ADDR1(src[0])
MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ONE)
| R500_ALU_RGB_SEL_B_SRC1 | MAKE_SWIZ_RGB_B(make_rgb_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst4 = R500_ALPHA_OP_MAD
- | R500_ALPHA_ADDRD(dest)
+ | R500_ALPHA_ADDRD(get_temp(fp, 0))
| R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ONE)
| R500_ALPHA_SEL_B_SRC1 | MAKE_SWIZ_ALPHA_B(make_alpha_swizzle(fpi->SrcReg[0]));
fp->inst[counter].inst5 = R500_ALU_RGBA_OP_MAD
- | R500_ALU_RGBA_ADDRD(dest)
+ | R500_ALU_RGBA_ADDRD(get_temp(fp, 0))
| R500_ALU_RGBA_SEL_C_SRC2
| MAKE_SWIZ_RGBA_C(make_rgb_swizzle(fpi->SrcReg[1]))
| R500_ALU_RGBA_MOD_C_NEG
| R500_ALU_RGBA_ALPHA_SEL_C_SRC2
| MAKE_SWIZ_ALPHA_C(make_alpha_swizzle(fpi->SrcReg[1]))
| R500_ALU_RGBA_ALPHA_MOD_C_NEG;
+ counter++;
+ /* This inst's selects need to be swapped as follows:
+ * 0 -> C ; 1 -> B ; 2 -> A */
+ emit_alu(fp, counter, fpi);
+ fp->inst[counter].inst1 = R500_RGB_ADDR0(get_temp(fp, 0));
+ fp->inst[counter].inst2 = R500_ALPHA_ADDR0(get_temp(fp, 0));
+ fp->inst[counter].inst3 = R500_ALU_RGB_SEL_A_SRC0
+ | MAKE_SWIZ_RGB_A(R500_SWIZ_RGB_ZERO)
+ | R500_ALU_RGB_SEL_B_SRC0
+ | MAKE_SWIZ_RGB_B(R500_SWIZ_RGB_ONE);
+ fp->inst[counter].inst4 |= R500_ALPHA_OP_CMP
+ | R500_ALPHA_ADDRD(dest)
+ | R500_ALPHA_SEL_A_SRC0 | MAKE_SWIZ_ALPHA_A(R500_SWIZZLE_ZERO)
+ | R500_ALPHA_SEL_B_SRC0 | MAKE_SWIZ_ALPHA_B(R500_SWIZZLE_ONE);
+ fp->inst[counter].inst5 = R500_ALU_RGBA_OP_CMP
+ | R500_ALU_RGBA_ADDRD(dest)
+ | R500_ALU_RGBA_SEL_C_SRC0
+ | MAKE_SWIZ_RGBA_C(R500_SWIZ_RGB_RGB)
+ | R500_ALU_RGBA_ALPHA_SEL_C_SRC0
+ | R500_ALU_RGBA_A_SWIZ_A;
+ break;
+ case OPCODE_SUB:
+ /* Variation on MAD: 1*src0-src1 */
+ fpi->SrcReg[1].NegateBase = 0xF; /* NEG_XYZW */
+ emit_mad(fp, counter, fpi, R500_SWIZZLE_ONE, 0, 1);
break;
case OPCODE_SWZ:
- /* TODO: Negation masks! */
+ /* 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:
- emit_tex(fp, fpi, OPCODE_TEX, dest, counter);
- if (fpi->DstReg.File == PROGRAM_OUTPUT)
- counter++;
- break;
case OPCODE_TXB:
- emit_tex(fp, fpi, OPCODE_TXB, dest, counter);
- if (fpi->DstReg.File == PROGRAM_OUTPUT)
- counter++;
- break;
case OPCODE_TXP:
- emit_tex(fp, fpi, OPCODE_TXP, dest, counter);
+ emit_tex(fp, fpi, dest, counter);
if (fpi->DstReg.File == PROGRAM_OUTPUT)
counter++;
break;
} else {
/* We still need to put an output inst, right? */
WARN_ONCE("Final FP instruction is not an OUT.\n");
-#if 0
-
-#endif
}
fp->cs->nrslots = counter;
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;
- fp->max_temp_idx = fp->temp_reg_offset + cs->temp_in_use + 1;
+ 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)
fp->inst_end = cs->nrslots - 1;
fp->translated = GL_TRUE;
- if (1 || RADEON_DEBUG & DEBUG_PIXEL) {
+ if (RADEON_DEBUG & DEBUG_PIXEL) {
fprintf(stderr, "Mesa program:\n");
fprintf(stderr, "-------------\n");
_mesa_print_program(&fp->mesa_program.Base);
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);
fprintf(stderr,"\t4 ALPHA_INST:0x%08x:", fp->inst[n].inst4);
inst = fp->inst[n].inst4;
- fprintf(stderr,"%s dest:%d%s alp_A_src:%d %s %d alp_B_src:%d %s %d\n", to_alpha_op(inst & 0xf),
+ fprintf(stderr,"%s dest:%d%s alp_A_src:%d %s %d alp_B_src:%d %s %d w:%d\n", to_alpha_op(inst & 0xf),
(inst >> 4) & 0x7f, inst & (1<<11) ? "(rel)":"",
(inst >> 12) & 0x3, toswiz((inst >> 14) & 0x7), (inst >> 17) & 0x3,
- (inst >> 19) & 0x3, toswiz((inst >> 21) & 0x7), (inst >> 24) & 0x3);
+ (inst >> 19) & 0x3, toswiz((inst >> 21) & 0x7), (inst >> 24) & 0x3,
+ (inst >> 31) & 0x1);
fprintf(stderr,"\t5 RGBA_INST: 0x%08x:", fp->inst[n].inst5);
inst = fp->inst[n].inst5;
projects / mesa.git / blobdiff