From 3369cd9a6f943365242d7832e69788d4aede9a8f Mon Sep 17 00:00:00 2001 From: "Xiang, Haihao" Date: Mon, 7 Jan 2008 14:08:36 +0800 Subject: [PATCH] i915: Keith Whitwell's swizzling TEX patch. fix #8283 --- src/mesa/drivers/dri/i915/i915_context.h | 6 +++ src/mesa/drivers/dri/i915/i915_fragprog.c | 51 ++++++++++++++++++++++- src/mesa/drivers/dri/i915/i915_program.c | 38 ++++++++++++----- src/mesa/drivers/dri/i915/i915_program.h | 1 + 4 files changed, 83 insertions(+), 13 deletions(-) diff --git a/src/mesa/drivers/dri/i915/i915_context.h b/src/mesa/drivers/dri/i915/i915_context.h index 1070de1a540..c6958dd8d4b 100644 --- a/src/mesa/drivers/dri/i915/i915_context.h +++ b/src/mesa/drivers/dri/i915/i915_context.h @@ -29,6 +29,7 @@ #define I915CONTEXT_INC #include "intel_context.h" +#include "i915_reg.h" #define I915_FALLBACK_TEXTURE 0x1000 #define I915_FALLBACK_COLORMASK 0x2000 @@ -110,6 +111,7 @@ #define I915_PROGRAM_SIZE 192 +#define I915_MAX_INSN (I915_MAX_TEX_INSN+I915_MAX_ALU_INSN) /* Hardware version of a parsed fragment program. "Derived" from the * mesa fragment_program struct. @@ -161,6 +163,10 @@ struct i915_fragment_program */ + /* Track which R registers are "live" for each instruction. + * A register is live between the time it's written to and the last time + * it's read. */ + GLuint usedRegs[I915_MAX_INSN]; /* Helpers for i915_fragprog.c: */ diff --git a/src/mesa/drivers/dri/i915/i915_fragprog.c b/src/mesa/drivers/dri/i915/i915_fragprog.c index 4c3f2236e5e..bafc8f02b84 100644 --- a/src/mesa/drivers/dri/i915/i915_fragprog.c +++ b/src/mesa/drivers/dri/i915/i915_fragprog.c @@ -231,7 +231,7 @@ do { \ GLuint coord = src_vector( p, &inst->SrcReg[0], program); \ /* Texel lookup */ \ \ - i915_emit_texld( p, \ + i915_emit_texld( p, get_live_regs(p, inst), \ get_result_vector( p, inst ), \ get_result_flags( inst ), \ sampler, \ @@ -254,6 +254,43 @@ do { \ #define EMIT_2ARG_ARITH( OP ) EMIT_ARITH( OP, 2 ) #define EMIT_3ARG_ARITH( OP ) EMIT_ARITH( OP, 3 ) +/* + * TODO: consider moving this into core + */ +static void calc_live_regs( struct i915_fragment_program *p ) +{ + const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current; + GLuint regsUsed = 0xffff0000; + GLint i; + + for (i = program->Base.NumInstructions - 1; i >= 0; i--) { + struct prog_instruction *inst = &program->Base.Instructions[i]; + int opArgs = _mesa_num_inst_src_regs(inst->Opcode); + int a; + + /* Register is written to: unmark as live for this and preceeding ops */ + if (inst->DstReg.File == PROGRAM_TEMPORARY) + regsUsed &= ~(1 << inst->DstReg.Index); + + for (a = 0; a < opArgs; a++) { + /* Register is read from: mark as live for this and preceeding ops */ + if (inst->SrcReg[a].File == PROGRAM_TEMPORARY) + regsUsed |= 1 << inst->SrcReg[a].Index; + } + + p->usedRegs[i] = regsUsed; + } +} + +static GLuint get_live_regs( struct i915_fragment_program *p, + const struct prog_instruction *inst ) +{ + const struct gl_fragment_program *program = p->ctx->FragmentProgram._Current; + GLuint nr = inst - program->Base.Instructions; + + return p->usedRegs[nr]; +} + /* Possible concerns: * @@ -289,6 +326,15 @@ upload_program(struct i915_fragment_program *p) return; } + if (program->Base.NumInstructions > I915_MAX_INSN) { + i915_program_error( p, "Exceeded max instructions" ); + return; + } + + /* Not always needed: + */ + calc_live_regs(p); + while (1) { GLuint src0, src1, src2, flags; GLuint tmp = 0; @@ -423,7 +469,8 @@ upload_program(struct i915_fragment_program *p) src0 = src_vector(p, &inst->SrcReg[0], program); tmp = i915_get_utemp(p); - i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */ + i915_emit_texld(p, get_live_regs(p, inst), + tmp, A0_DEST_CHANNEL_ALL, /* use a dummy dest reg */ 0, src0, T0_TEXKILL); break; diff --git a/src/mesa/drivers/dri/i915/i915_program.c b/src/mesa/drivers/dri/i915/i915_program.c index c6c64340232..f79d00ddcf3 100644 --- a/src/mesa/drivers/dri/i915/i915_program.c +++ b/src/mesa/drivers/dri/i915/i915_program.c @@ -194,27 +194,43 @@ i915_emit_arith(struct i915_fragment_program * p, return dest; } +static GLuint get_free_rreg (struct i915_fragment_program *p, + GLuint live_regs) +{ + int bit = ffs(~live_regs); + if (!bit) { + i915_program_error(p, "Can't find free R reg"); + return UREG_BAD; + } + return UREG(REG_TYPE_R, bit - 1); +} + GLuint i915_emit_texld( struct i915_fragment_program *p, + GLuint live_regs, GLuint dest, GLuint destmask, GLuint sampler, GLuint coord, GLuint op ) { - if (coord != UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord))) { - /* No real way to work around this in the general case - need to - * allocate and declare a new temporary register (a utemp won't - * do). Will fallback for now. - */ - i915_program_error(p, "Can't (yet) swizzle TEX arguments"); - return 0; - } - - /* Don't worry about saturate as we only support + if (coord != UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord))) { + /* With the help of the "needed registers" table created earlier, pick + * a register we can MOV the swizzled TC to (since TEX doesn't support + * swizzled sources) */ + GLuint swizCoord = get_free_rreg(p, live_regs); + if (swizCoord == UREG_BAD) + return 0; + + i915_emit_arith( p, A0_MOV, swizCoord, A0_DEST_CHANNEL_ALL, 0, coord, 0, 0 ); + coord = swizCoord; + } + + /* Don't worry about saturate as we only support texture formats + * that are always in the 0..1 range. */ if (destmask != A0_DEST_CHANNEL_ALL) { GLuint tmp = i915_get_utemp(p); - i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op ); + i915_emit_texld( p, 0, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op ); i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 ); return dest; } diff --git a/src/mesa/drivers/dri/i915/i915_program.h b/src/mesa/drivers/dri/i915/i915_program.h index 3c12b34f163..14a3f08801f 100644 --- a/src/mesa/drivers/dri/i915/i915_program.h +++ b/src/mesa/drivers/dri/i915/i915_program.h @@ -112,6 +112,7 @@ extern void i915_release_utemps(struct i915_fragment_program *p); extern GLuint i915_emit_texld(struct i915_fragment_program *p, + GLuint live_regs, GLuint dest, GLuint destmask, GLuint sampler, GLuint coord, GLuint op); -- 2.30.2