From 2b58c31257f8900067276b6d6537bb2ce54b1b10 Mon Sep 17 00:00:00 2001 From: Eric Anholt Date: Wed, 12 Aug 2009 10:19:31 -0700 Subject: i965: Share most of the WM functions between brw_wm_glsl.c and brw_wm_emit.c The PINTERP code should be faster for brw_wm_glsl.c now since brw_wm_emit.c's had been improved, and pixel_w should no longer stomp on a neighbor to dst. --- src/mesa/drivers/dri/i965/brw_wm.h | 34 +++++ src/mesa/drivers/dri/i965/brw_wm_emit.c | 115 ++++++++------ src/mesa/drivers/dri/i965/brw_wm_glsl.c | 257 +------------------------------- 3 files changed, 109 insertions(+), 297 deletions(-) diff --git a/src/mesa/drivers/dri/i965/brw_wm.h b/src/mesa/drivers/dri/i965/brw_wm.h index 7d470e8dfe..eba828f6e3 100644 --- a/src/mesa/drivers/dri/i965/brw_wm.h +++ b/src/mesa/drivers/dri/i965/brw_wm.h @@ -325,11 +325,19 @@ void emit_alu2(struct brw_compile *p, GLuint mask, const struct brw_reg *arg0, const struct brw_reg *arg1); +void emit_cinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); void emit_ddxy(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, GLboolean is_ddx, const struct brw_reg *arg0); +void emit_delta_xy(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); void emit_dp3(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, @@ -345,6 +353,14 @@ void emit_dph(struct brw_compile *p, GLuint mask, const struct brw_reg *arg0, const struct brw_reg *arg1); +void emit_frontfacing(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask); +void emit_linterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas); void emit_lrp(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, @@ -368,12 +384,30 @@ void emit_math2(struct brw_wm_compile *c, GLuint mask, const struct brw_reg *arg0, const struct brw_reg *arg1); +void emit_pinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas, + const struct brw_reg *w); +void emit_pixel_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask); +void emit_pixel_w(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas); void emit_sop(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, GLuint cond, const struct brw_reg *arg0, const struct brw_reg *arg1); +void emit_wpos_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0); void emit_xpd(struct brw_compile *p, const struct brw_reg *dst, GLuint mask, diff --git a/src/mesa/drivers/dri/i965/brw_wm_emit.c b/src/mesa/drivers/dri/i965/brw_wm_emit.c index 763fe4b4d6..6eaa2792be 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_emit.c +++ b/src/mesa/drivers/dri/i965/brw_wm_emit.c @@ -44,6 +44,7 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) return reg; } + /* Payload R0: * * R0.0 -- pixel mask, one bit for each of 4 pixels in 4 tiles, @@ -60,42 +61,50 @@ static INLINE struct brw_reg sechalf( struct brw_reg reg ) * R1.8 -- ? */ - -static void emit_pixel_xy(struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask) +void emit_pixel_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask) { + struct brw_compile *p = &c->func; struct brw_reg r1 = brw_vec1_grf(1, 0); struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); + struct brw_reg dst0_uw, dst1_uw; + brw_push_insn_state(p); brw_set_compression_control(p, BRW_COMPRESSION_NONE); + if (c->dispatch_width == 16) { + dst0_uw = vec16(retype(dst[0], BRW_REGISTER_TYPE_UW)); + dst1_uw = vec16(retype(dst[1], BRW_REGISTER_TYPE_UW)); + } else { + dst0_uw = vec8(retype(dst[0], BRW_REGISTER_TYPE_UW)); + dst1_uw = vec8(retype(dst[1], BRW_REGISTER_TYPE_UW)); + } + /* Calculate pixel centers by adding 1 or 0 to each of the * micro-tile coordinates passed in r1. */ if (mask & WRITEMASK_X) { brw_ADD(p, - vec16(retype(dst[0], BRW_REGISTER_TYPE_UW)), + dst0_uw, stride(suboffset(r1_uw, 4), 2, 4, 0), brw_imm_v(0x10101010)); } if (mask & WRITEMASK_Y) { brw_ADD(p, - vec16(retype(dst[1], BRW_REGISTER_TYPE_UW)), + dst1_uw, stride(suboffset(r1_uw,5), 2, 4, 0), brw_imm_v(0x11001100)); } - - brw_set_compression_control(p, BRW_COMPRESSION_COMPRESSED); + brw_pop_insn_state(p); } - -static void emit_delta_xy(struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) +void emit_delta_xy(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { struct brw_reg r1 = brw_vec1_grf(1, 0); @@ -118,10 +127,10 @@ static void emit_delta_xy(struct brw_compile *p, } } -static void emit_wpos_xy(struct brw_wm_compile *c, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0) +void emit_wpos_xy(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { struct brw_compile *p = &c->func; @@ -146,12 +155,14 @@ static void emit_wpos_xy(struct brw_wm_compile *c, } -static void emit_pixel_w( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *deltas) +void emit_pixel_w(struct brw_wm_compile *c, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas) { + struct brw_compile *p = &c->func; + /* Don't need this if all you are doing is interpolating color, for * instance. */ @@ -165,21 +176,29 @@ static void emit_pixel_w( struct brw_compile *p, brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), deltas[1]); /* Calc w */ - brw_math_16( p, dst[3], - BRW_MATH_FUNCTION_INV, - BRW_MATH_SATURATE_NONE, - 2, brw_null_reg(), - BRW_MATH_PRECISION_FULL); + if (c->dispatch_width == 16) { + brw_math_16(p, dst[3], + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 2, brw_null_reg(), + BRW_MATH_PRECISION_FULL); + } else { + brw_math(p, dst[3], + BRW_MATH_FUNCTION_INV, + BRW_MATH_SATURATE_NONE, + 2, brw_null_reg(), + BRW_MATH_DATA_VECTOR, + BRW_MATH_PRECISION_FULL); + } } } - -static void emit_linterp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *deltas ) +void emit_linterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas) { struct brw_reg interp[4]; GLuint nr = arg0[0].nr; @@ -199,12 +218,12 @@ static void emit_linterp( struct brw_compile *p, } -static void emit_pinterp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0, - const struct brw_reg *deltas, - const struct brw_reg *w) +void emit_pinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0, + const struct brw_reg *deltas, + const struct brw_reg *w) { struct brw_reg interp[4]; GLuint nr = arg0[0].nr; @@ -229,10 +248,10 @@ static void emit_pinterp( struct brw_compile *p, } -static void emit_cinterp( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask, - const struct brw_reg *arg0 ) +void emit_cinterp(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask, + const struct brw_reg *arg0) { struct brw_reg interp[4]; GLuint nr = arg0[0].nr; @@ -251,9 +270,9 @@ static void emit_cinterp( struct brw_compile *p, } /* Sets the destination channels to 1.0 or 0.0 according to glFrontFacing. */ -static void emit_frontfacing( struct brw_compile *p, - const struct brw_reg *dst, - GLuint mask ) +void emit_frontfacing(struct brw_compile *p, + const struct brw_reg *dst, + GLuint mask) { struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD); GLuint i; @@ -1316,7 +1335,7 @@ void brw_wm_emit( struct brw_wm_compile *c ) /* Generated instructions for calculating triangle interpolants: */ case WM_PIXELXY: - emit_pixel_xy(p, dst, dst_flags); + emit_pixel_xy(c, dst, dst_flags); break; case WM_DELTAXY: @@ -1328,7 +1347,7 @@ void brw_wm_emit( struct brw_wm_compile *c ) break; case WM_PIXELW: - emit_pixel_w(p, dst, dst_flags, args[0], args[1]); + emit_pixel_w(c, dst, dst_flags, args[0], args[1]); break; case WM_LINTERP: diff --git a/src/mesa/drivers/dri/i965/brw_wm_glsl.c b/src/mesa/drivers/dri/i965/brw_wm_glsl.c index 02b119154f..d0987362a4 100644 --- a/src/mesa/drivers/dri/i965/brw_wm_glsl.c +++ b/src/mesa/drivers/dri/i965/brw_wm_glsl.c @@ -671,67 +671,6 @@ unalias3(struct brw_wm_compile *c, release_tmps(c, mark); } -static void emit_pixel_xy(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg r1 = brw_vec1_grf(1, 0); - struct brw_reg r1_uw = retype(r1, BRW_REGISTER_TYPE_UW); - - struct brw_reg dst0, dst1; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - - dst0 = get_dst_reg(c, inst, 0); - dst1 = get_dst_reg(c, inst, 1); - /* Calculate pixel centers by adding 1 or 0 to each of the - * micro-tile coordinates passed in r1. - */ - if (mask & WRITEMASK_X) { - brw_ADD(p, - vec8(retype(dst0, BRW_REGISTER_TYPE_UW)), - stride(suboffset(r1_uw, 4), 2, 4, 0), - brw_imm_v(0x10101010)); - } - - if (mask & WRITEMASK_Y) { - brw_ADD(p, - vec8(retype(dst1, BRW_REGISTER_TYPE_UW)), - stride(suboffset(r1_uw, 5), 2, 4, 0), - brw_imm_v(0x11001100)); - } -} - -static void emit_delta_xy(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_reg r1 = brw_vec1_grf(1, 0); - struct brw_reg dst0, dst1, src0, src1; - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - - dst0 = get_dst_reg(c, inst, 0); - dst1 = get_dst_reg(c, inst, 1); - src0 = get_src_reg(c, inst, 0, 0); - src1 = get_src_reg(c, inst, 0, 1); - /* Calc delta X,Y by subtracting origin in r1 from the pixel - * centers. - */ - if (mask & WRITEMASK_X) { - brw_ADD(p, - dst0, - retype(src0, BRW_REGISTER_TYPE_UW), - negate(r1)); - } - - if (mask & WRITEMASK_Y) { - brw_ADD(p, - dst1, - retype(src1, BRW_REGISTER_TYPE_UW), - negate(suboffset(r1,1))); - - } -} - static void fire_fb_write( struct brw_wm_compile *c, GLuint base_reg, GLuint nr, @@ -829,154 +768,6 @@ static void emit_fb_write(struct brw_wm_compile *c, fire_fb_write(c, 0, nr, target, eot); } -static void emit_pixel_w( struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - if (mask & WRITEMASK_W) { - struct brw_reg dst, src0, delta0, delta1; - struct brw_reg interp3; - - dst = get_dst_reg(c, inst, 3); - src0 = get_src_reg(c, inst, 0, 0); - delta0 = get_src_reg(c, inst, 1, 0); - delta1 = get_src_reg(c, inst, 1, 1); - - interp3 = brw_vec1_grf(src0.nr+1, 4); - /* Calc 1/w - just linterp wpos[3] optimized by putting the - * result straight into a message reg. - */ - brw_LINE(p, brw_null_reg(), interp3, delta0); - brw_MAC(p, brw_message_reg(2), suboffset(interp3, 1), delta1); - - /* Calc w */ - brw_math_16( p, dst, - BRW_MATH_FUNCTION_INV, - BRW_MATH_SATURATE_NONE, - 2, brw_null_reg(), - BRW_MATH_PRECISION_FULL); - } -} - -static void emit_linterp(struct brw_wm_compile *c, - const struct prog_instruction *inst) -{ - struct brw_compile *p = &c->func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg interp[4]; - struct brw_reg dst, delta0, delta1; - struct brw_reg src0; - GLuint nr, i; - - src0 = get_src_reg(c, inst, 0, 0); - delta0 = get_src_reg(c, inst, 1, 0); - delta1 = get_src_reg(c, inst, 1, 1); - nr = src0.nr; - - interp[0] = brw_vec1_grf(nr, 0); - interp[1] = brw_vec1_grf(nr, 4); - interp[2] = brw_vec1_grf(nr+1, 0); - interp[3] = brw_vec1_grf(nr+1, 4); - - for(i = 0; i < 4; i++ ) { - if (mask & (1<func; - GLuint mask = inst->DstReg.WriteMask; - - struct brw_reg interp[4]; - struct brw_reg dst, src0; - GLuint nr, i; - - src0 = get_src_reg(c, inst, 0, 0); - nr = src0.nr; - - interp[0] = brw_vec1_grf(nr, 0); - interp[1] = brw_vec1_grf(nr, 4); - interp[2] = brw_vec1_grf(nr+1, 0); - interp[3] = brw_vec1_grf(nr+1, 4); - - for(i = 0; i < 4; i++ ) { - if (mask & (1<func; - GLuint mask = inst->DstReg.WriteMask; - - struct brw_reg interp[4]; - struct brw_reg dst, delta0, delta1; - struct brw_reg src0, w; - GLuint nr, i; - - src0 = get_src_reg(c, inst, 0, 0); - delta0 = get_src_reg(c, inst, 1, 0); - delta1 = get_src_reg(c, inst, 1, 1); - w = get_src_reg(c, inst, 2, 3); - nr = src0.nr; - - interp[0] = brw_vec1_grf(nr, 0); - interp[1] = brw_vec1_grf(nr, 4); - interp[2] = brw_vec1_grf(nr+1, 0); - interp[3] = brw_vec1_grf(nr+1, 4); - - for(i = 0; i < 4; i++ ) { - if (mask & (1<func; - struct brw_reg r1_6ud = retype(brw_vec1_grf(1, 6), BRW_REGISTER_TYPE_UD); - struct brw_reg dst; - GLuint mask = inst->DstReg.WriteMask; - int i; - - for (i = 0; i < 4; i++) { - if (mask & (1<func; - GLuint mask = inst->DstReg.WriteMask; - struct brw_reg src0[2], dst[2]; - - dst[0] = get_dst_reg(c, inst, 0); - dst[1] = get_dst_reg(c, inst, 1); - src0[0] = get_src_reg(c, inst, 0, 0); - src0[1] = get_src_reg(c, inst, 0, 1); - - /* Calculate the pixel offset from window bottom left into destination - * X and Y channels. - */ - if (mask & WRITEMASK_X) { - /* X' = X - origin_x */ - brw_ADD(p, - dst[0], - retype(src0[0], BRW_REGISTER_TYPE_W), - brw_imm_d(0 - c->key.origin_x)); - } - - if (mask & WRITEMASK_Y) { - /* Y' = height - (Y - origin_y) = height + origin_y - Y */ - brw_ADD(p, - dst[1], - negate(retype(src0[1], BRW_REGISTER_TYPE_W)), - brw_imm_d(c->key.origin_y + c->key.drawable_height - 1)); - } -} /* TODO BIAS on SIMD8 not working yet... @@ -2378,31 +2137,31 @@ static void brw_wm_emit_glsl(struct brw_context *brw, struct brw_wm_compile *c) switch (inst->Opcode) { case WM_PIXELXY: - emit_pixel_xy(c, inst); + emit_pixel_xy(c, dst, dst_flags); break; case WM_DELTAXY: - emit_delta_xy(c, inst); + emit_delta_xy(p, dst, dst_flags, args[0]); break; case WM_PIXELW: - emit_pixel_w(c, inst); + emit_pixel_w(c, dst, dst_flags, args[0], args[1]); break; case WM_LINTERP: - emit_linterp(c, inst); + emit_linterp(p, dst, dst_flags, args[0], args[1]); break; case WM_PINTERP: - emit_pinterp(c, inst); + emit_pinterp(p, dst, dst_flags, args[0], args[1], args[2]); break; case WM_CINTERP: - emit_cinterp(c, inst); + emit_cinterp(p, dst, dst_flags, args[0]); break; case WM_WPOSXY: - emit_wpos_xy(c, inst); + emit_wpos_xy(c, dst, dst_flags, args[0]); break; case WM_FB_WRITE: emit_fb_write(c, inst); break; case WM_FRONTFACING: - emit_frontfacing(c, inst); + emit_frontfacing(p, dst, dst_flags); break; case OPCODE_ADD: emit_alu2(p, brw_ADD, dst, dst_flags, args[0], args[1]); -- cgit v1.2.3