diff options
Diffstat (limited to 'src/gallium/auxiliary')
45 files changed, 794 insertions, 409 deletions
diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c index 8f6ca15dfa..1c07ab1365 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_aaline.c +++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c @@ -422,9 +422,9 @@ aaline_create_texture(struct aaline_stage *aaline) /* This texture is new, no need to flush. */ - transfer = screen->get_tex_transfer(screen, aaline->texture, 0, level, 0, + transfer = pipe->get_tex_transfer(pipe, aaline->texture, 0, level, 0, PIPE_TRANSFER_WRITE, 0, 0, size, size); - data = screen->transfer_map(screen, transfer); + data = pipe->transfer_map(pipe, transfer); if (data == NULL) return FALSE; @@ -448,8 +448,8 @@ aaline_create_texture(struct aaline_stage *aaline) } /* unmap */ - screen->transfer_unmap(screen, transfer); - screen->tex_transfer_destroy(transfer); + pipe->transfer_unmap(pipe, transfer); + pipe->tex_transfer_destroy(pipe, transfer); } return TRUE; } diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c index d0d99aa331..38c22bf4e9 100644 --- a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c +++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c @@ -374,19 +374,21 @@ pstip_update_texture(struct pstip_stage *pstip) { static const uint bit31 = 1 << 31; struct pipe_context *pipe = pstip->pipe; - struct pipe_screen *screen = pipe->screen; struct pipe_transfer *transfer; const uint *stipple = pstip->state.stipple->stipple; uint i, j; ubyte *data; /* XXX: want to avoid flushing just because we use stipple: + * + * Flush should no longer be necessary if driver is properly + * interleaving drawing and transfers on a given context: */ pipe->flush( pipe, PIPE_FLUSH_TEXTURE_CACHE, NULL ); - transfer = screen->get_tex_transfer(screen, pstip->texture, 0, 0, 0, - PIPE_TRANSFER_WRITE, 0, 0, 32, 32); - data = screen->transfer_map(screen, transfer); + transfer = pipe->get_tex_transfer(pipe, pstip->texture, 0, 0, 0, + PIPE_TRANSFER_WRITE, 0, 0, 32, 32); + data = pipe->transfer_map(pipe, transfer); /* * Load alpha texture. @@ -408,8 +410,8 @@ pstip_update_texture(struct pstip_stage *pstip) } /* unmap */ - screen->transfer_unmap(screen, transfer); - screen->tex_transfer_destroy(transfer); + pipe->transfer_unmap(pipe, transfer); + pipe->tex_transfer_destroy(pipe, transfer); } diff --git a/src/gallium/auxiliary/os/os_llvm.h b/src/gallium/auxiliary/gallivm/lp_bld.h index d5edfbfe92..70a4960f91 100644 --- a/src/gallium/auxiliary/os/os_llvm.h +++ b/src/gallium/auxiliary/gallivm/lp_bld.h @@ -31,8 +31,8 @@ */ -#ifndef OS_LLVM_H -#define OS_LLVM_H +#ifndef LP_BLD_H +#define LP_BLD_H #include <llvm-c/Core.h> @@ -40,8 +40,8 @@ /** Set version to 0 if missing to avoid #ifdef HAVE_LLVM everywhere */ #ifndef HAVE_LLVM -#define HAVE_LLVM 0x0 +#define HAVE_LLVM 0x0207 #endif -#endif /* OS_LLVM_H */ +#endif /* LP_BLD_H */ diff --git a/src/gallium/auxiliary/gallivm/lp_bld_alpha.h b/src/gallium/auxiliary/gallivm/lp_bld_alpha.h index fe3cedcc48..0f99fec65e 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_alpha.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_alpha.h @@ -35,7 +35,7 @@ #define LP_BLD_ALPHA_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct pipe_alpha_state; struct lp_type; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.c b/src/gallium/auxiliary/gallivm/lp_bld_arit.c index 90c84c6a4e..8e8fcccf56 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.c @@ -361,12 +361,12 @@ lp_build_mul_u8n(LLVMBuilderRef builder, LLVMValueRef c8; LLVMValueRef ab; - c8 = lp_build_int_const_scalar(i16_type, 8); + c8 = lp_build_const_int_vec(i16_type, 8); #if 0 /* a*b/255 ~= (a*(b + 1)) >> 256 */ - b = LLVMBuildAdd(builder, b, lp_build_int_const_scalar(i16_type, 1), ""); + b = LLVMBuildAdd(builder, b, lp_build_const_int_vec(i16_type, 1), ""); ab = LLVMBuildMul(builder, a, b, ""); #else @@ -374,7 +374,7 @@ lp_build_mul_u8n(LLVMBuilderRef builder, /* ab/255 ~= (ab + (ab >> 8) + 0x80) >> 8 */ ab = LLVMBuildMul(builder, a, b, ""); ab = LLVMBuildAdd(builder, ab, LLVMBuildLShr(builder, ab, c8, ""), ""); - ab = LLVMBuildAdd(builder, ab, lp_build_int_const_scalar(i16_type, 0x80), ""); + ab = LLVMBuildAdd(builder, ab, lp_build_const_int_vec(i16_type, 0x80), ""); #endif @@ -429,7 +429,7 @@ lp_build_mul(struct lp_build_context *bld, } if(type.fixed) - shift = lp_build_int_const_scalar(type, type.width/2); + shift = lp_build_const_int_vec(type, type.width/2); else shift = NULL; @@ -491,7 +491,7 @@ lp_build_mul_imm(struct lp_build_context *bld, * for Inf and NaN. */ unsigned mantissa = lp_mantissa(bld->type); - factor = lp_build_int_const_scalar(bld->type, (unsigned long long)shift << mantissa); + factor = lp_build_const_int_vec(bld->type, (unsigned long long)shift << mantissa); a = LLVMBuildBitCast(bld->builder, a, lp_build_int_vec_type(bld->type), ""); a = LLVMBuildAdd(bld->builder, a, factor, ""); a = LLVMBuildBitCast(bld->builder, a, lp_build_vec_type(bld->type), ""); @@ -499,12 +499,12 @@ lp_build_mul_imm(struct lp_build_context *bld, #endif } else { - factor = lp_build_const_scalar(bld->type, shift); + factor = lp_build_const_vec(bld->type, shift); return LLVMBuildShl(bld->builder, a, factor, ""); } } - factor = lp_build_const_scalar(bld->type, (double)b); + factor = lp_build_const_vec(bld->type, (double)b); return lp_build_mul(bld, a, factor); } @@ -567,7 +567,7 @@ lp_build_lerp(struct lp_build_context *bld, * but it will be wrong for other uses. Basically we need a more * powerful lp_type, capable of further distinguishing the values * interpretation from the value storage. */ - res = LLVMBuildAnd(bld->builder, res, lp_build_int_const_scalar(bld->type, (1 << bld->type.width/2) - 1), ""); + res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(bld->type, (1 << bld->type.width/2) - 1), ""); return res; } @@ -689,7 +689,7 @@ lp_build_abs(struct lp_build_context *bld, /* vector of floats */ LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); unsigned long long absMask = ~(1ULL << (type.width - 1)); - LLVMValueRef mask = lp_build_int_const_scalar(type, ((unsigned long long) absMask)); + LLVMValueRef mask = lp_build_const_int_vec(type, ((unsigned long long) absMask)); a = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); a = LLVMBuildAnd(bld->builder, a, mask, ""); a = LLVMBuildBitCast(bld->builder, a, vec_type, ""); @@ -720,12 +720,12 @@ lp_build_negate(struct lp_build_context *bld, } +/** Return -1, 0 or +1 depending on the sign of a */ LLVMValueRef lp_build_sgn(struct lp_build_context *bld, LLVMValueRef a) { const struct lp_type type = bld->type; - LLVMTypeRef vec_type = lp_build_vec_type(type); LLVMValueRef cond; LLVMValueRef res; @@ -735,27 +735,42 @@ lp_build_sgn(struct lp_build_context *bld, res = bld->one; } else if(type.floating) { - /* Take the sign bit and add it to 1 constant */ - LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); - LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1)); + LLVMTypeRef vec_type; + LLVMTypeRef int_type; + LLVMValueRef mask; LLVMValueRef sign; LLVMValueRef one; - sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); + unsigned long long maskBit = (unsigned long long)1 << (type.width - 1); + + if (type.length == 1) { + int_type = lp_build_int_elem_type(type); + vec_type = lp_build_elem_type(type); + mask = LLVMConstInt(int_type, maskBit, 0); + } + else { + /* vector */ + int_type = lp_build_int_vec_type(type); + vec_type = lp_build_vec_type(type); + mask = lp_build_const_int_vec(type, maskBit); + } + + /* Take the sign bit and add it to 1 constant */ + sign = LLVMBuildBitCast(bld->builder, a, int_type, ""); sign = LLVMBuildAnd(bld->builder, sign, mask, ""); - one = LLVMConstBitCast(bld->one, int_vec_type); + one = LLVMConstBitCast(bld->one, int_type); res = LLVMBuildOr(bld->builder, sign, one, ""); res = LLVMBuildBitCast(bld->builder, res, vec_type, ""); } else { - LLVMValueRef minus_one = lp_build_const_scalar(type, -1.0); + LLVMValueRef minus_one = lp_build_const_vec(type, -1.0); cond = lp_build_cmp(bld, PIPE_FUNC_GREATER, a, bld->zero); res = lp_build_select(bld, cond, bld->one, minus_one); } /* Handle zero */ cond = lp_build_cmp(bld, PIPE_FUNC_EQUAL, a, bld->zero); - res = lp_build_select(bld, cond, bld->zero, bld->one); + res = lp_build_select(bld, cond, bld->zero, res); return res; } @@ -774,8 +789,8 @@ lp_build_set_sign(struct lp_build_context *bld, const struct lp_type type = bld->type; LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); LLVMTypeRef vec_type = lp_build_vec_type(type); - LLVMValueRef shift = lp_build_int_const_scalar(type, type.width - 1); - LLVMValueRef mask = lp_build_int_const_scalar(type, + LLVMValueRef shift = lp_build_const_int_vec(type, type.width - 1); + LLVMValueRef mask = lp_build_const_int_vec(type, ~((unsigned long long) 1 << (type.width - 1))); LLVMValueRef val, res; @@ -915,7 +930,10 @@ lp_build_floor(struct lp_build_context *bld, assert(type.floating); if (type.length == 1) { - return LLVMBuildFPTrunc(bld->builder, a, LLVMFloatType(), ""); + LLVMValueRef res; + res = lp_build_ifloor(bld, a); + res = LLVMBuildSIToFP(bld->builder, res, LLVMFloatType(), ""); + return res; } if(util_cpu_caps.has_sse4_1) @@ -978,7 +996,7 @@ lp_build_itrunc(struct lp_build_context *bld, if (type.length == 1) { LLVMTypeRef int_type = LLVMIntType(type.width); - return LLVMBuildFPTrunc(bld->builder, a, int_type, ""); + return LLVMBuildFPToSI(bld->builder, a, int_type, ""); } else { LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); @@ -1016,7 +1034,7 @@ lp_build_iround(struct lp_build_context *bld, } else { LLVMTypeRef vec_type = lp_build_vec_type(type); - LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); LLVMValueRef sign; LLVMValueRef half; @@ -1025,7 +1043,7 @@ lp_build_iround(struct lp_build_context *bld, sign = LLVMBuildAnd(bld->builder, sign, mask, ""); /* sign * 0.5 */ - half = lp_build_const_scalar(type, 0.5); + half = lp_build_const_vec(type, 0.5); half = LLVMBuildBitCast(bld->builder, half, int_vec_type, ""); half = LLVMBuildOr(bld->builder, sign, half, ""); half = LLVMBuildBitCast(bld->builder, half, vec_type, ""); @@ -1068,18 +1086,18 @@ lp_build_ifloor(struct lp_build_context *bld, /* Take the sign bit and add it to 1 constant */ LLVMTypeRef vec_type = lp_build_vec_type(type); unsigned mantissa = lp_mantissa(type); - LLVMValueRef mask = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); LLVMValueRef sign; LLVMValueRef offset; /* sign = a < 0 ? ~0 : 0 */ sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, ""); sign = LLVMBuildAnd(bld->builder, sign, mask, ""); - sign = LLVMBuildAShr(bld->builder, sign, lp_build_int_const_scalar(type, type.width - 1), ""); + sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), ""); lp_build_name(sign, "floor.sign"); /* offset = -0.99999(9)f */ - offset = lp_build_const_scalar(type, -(double)(((unsigned long long)1 << mantissa) - 1)/((unsigned long long)1 << mantissa)); + offset = lp_build_const_vec(type, -(double)(((unsigned long long)1 << mantissa) - 1)/((unsigned long long)1 << mantissa)); offset = LLVMConstBitCast(offset, int_vec_type); /* offset = a < 0 ? -0.99999(9)f : 0.0f */ @@ -1250,7 +1268,7 @@ lp_build_exp(struct lp_build_context *bld, LLVMValueRef x) { /* log2(e) = 1/log(2) */ - LLVMValueRef log2e = lp_build_const_scalar(bld->type, 1.4426950408889634); + LLVMValueRef log2e = lp_build_const_vec(bld->type, 1.4426950408889634); return lp_build_mul(bld, log2e, lp_build_exp2(bld, x)); } @@ -1264,7 +1282,7 @@ lp_build_log(struct lp_build_context *bld, LLVMValueRef x) { /* log(2) */ - LLVMValueRef log2 = lp_build_const_scalar(bld->type, 0.69314718055994529); + LLVMValueRef log2 = lp_build_const_vec(bld->type, 0.69314718055994529); return lp_build_mul(bld, log2, lp_build_exp2(bld, x)); } @@ -1300,7 +1318,7 @@ lp_build_polynomial(struct lp_build_context *bld, if (type.length == 1) coeff = LLVMConstReal(float_type, coeffs[i]); else - coeff = lp_build_const_scalar(type, coeffs[i]); + coeff = lp_build_const_vec(type, coeffs[i]); if(res) res = lp_build_add(bld, coeff, lp_build_mul(bld, x, res)); @@ -1357,11 +1375,11 @@ lp_build_exp2_approx(struct lp_build_context *bld, assert(type.floating && type.width == 32); - x = lp_build_min(bld, x, lp_build_const_scalar(type, 129.0)); - x = lp_build_max(bld, x, lp_build_const_scalar(type, -126.99999)); + x = lp_build_min(bld, x, lp_build_const_vec(type, 129.0)); + x = lp_build_max(bld, x, lp_build_const_vec(type, -126.99999)); /* ipart = int(x - 0.5) */ - ipart = LLVMBuildSub(bld->builder, x, lp_build_const_scalar(type, 0.5f), ""); + ipart = LLVMBuildSub(bld->builder, x, lp_build_const_vec(type, 0.5f), ""); ipart = LLVMBuildFPToSI(bld->builder, ipart, int_vec_type, ""); /* fpart = x - ipart */ @@ -1371,8 +1389,8 @@ lp_build_exp2_approx(struct lp_build_context *bld, if(p_exp2_int_part || p_exp2) { /* expipart = (float) (1 << ipart) */ - expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_int_const_scalar(type, 127), ""); - expipart = LLVMBuildShl(bld->builder, expipart, lp_build_int_const_scalar(type, 23), ""); + expipart = LLVMBuildAdd(bld->builder, ipart, lp_build_const_int_vec(type, 127), ""); + expipart = LLVMBuildShl(bld->builder, expipart, lp_build_const_int_vec(type, 23), ""); expipart = LLVMBuildBitCast(bld->builder, expipart, vec_type, ""); } @@ -1438,8 +1456,8 @@ lp_build_log2_approx(struct lp_build_context *bld, LLVMTypeRef vec_type = lp_build_vec_type(type); LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); - LLVMValueRef expmask = lp_build_int_const_scalar(type, 0x7f800000); - LLVMValueRef mantmask = lp_build_int_const_scalar(type, 0x007fffff); + LLVMValueRef expmask = lp_build_const_int_vec(type, 0x7f800000); + LLVMValueRef mantmask = lp_build_const_int_vec(type, 0x007fffff); LLVMValueRef one = LLVMConstBitCast(bld->one, int_vec_type); LLVMValueRef i = NULL; @@ -1464,8 +1482,8 @@ lp_build_log2_approx(struct lp_build_context *bld, } if(p_floor_log2 || p_log2) { - logexp = LLVMBuildLShr(bld->builder, exp, lp_build_int_const_scalar(type, 23), ""); - logexp = LLVMBuildSub(bld->builder, logexp, lp_build_int_const_scalar(type, 127), ""); + logexp = LLVMBuildLShr(bld->builder, exp, lp_build_const_int_vec(type, 23), ""); + logexp = LLVMBuildSub(bld->builder, logexp, lp_build_const_int_vec(type, 127), ""); logexp = LLVMBuildSIToFP(bld->builder, logexp, vec_type, ""); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_arit.h b/src/gallium/auxiliary/gallivm/lp_bld_arit.h index 7a10fe1220..31efa9921c 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_arit.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_arit.h @@ -37,7 +37,7 @@ #define LP_BLD_ARIT_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct lp_type; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_blend.h b/src/gallium/auxiliary/gallivm/lp_bld_blend.h index 5a9e1c1fb2..ebbdb1a604 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_blend.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_blend.h @@ -40,7 +40,7 @@ * for a standalone example. */ -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" #include "pipe/p_format.h" diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.c b/src/gallium/auxiliary/gallivm/lp_bld_const.c index 8a275fa72f..57843e9a60 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_const.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_const.c @@ -263,7 +263,7 @@ lp_build_one(struct lp_type type) if(type.sign) /* TODO: Unfortunately this caused "Tried to create a shift operation * on a non-integer type!" */ - vec = LLVMConstLShr(vec, lp_build_int_const_scalar(type, 1)); + vec = LLVMConstLShr(vec, lp_build_const_int_vec(type, 1)); #endif return vec; @@ -283,8 +283,8 @@ lp_build_one(struct lp_type type) * Build constant-valued vector from a scalar value. */ LLVMValueRef -lp_build_const_scalar(struct lp_type type, - double val) +lp_build_const_vec(struct lp_type type, + double val) { LLVMTypeRef elem_type = lp_build_elem_type(type); LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; @@ -309,7 +309,7 @@ lp_build_const_scalar(struct lp_type type, LLVMValueRef -lp_build_int_const_scalar(struct lp_type type, +lp_build_const_int_vec(struct lp_type type, long long val) { LLVMTypeRef elem_type = lp_build_int_elem_type(type); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_const.h b/src/gallium/auxiliary/gallivm/lp_bld_const.h index 4078636103..9ca2f0664e 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_const.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_const.h @@ -37,9 +37,9 @@ #define LP_BLD_CONST_H -#include "os/os_llvm.h" +#include "pipe/p_compiler.h" +#include "gallivm/lp_bld.h" -#include <pipe/p_compiler.h> struct lp_type; @@ -85,13 +85,11 @@ lp_build_one(struct lp_type type); LLVMValueRef -lp_build_const_scalar(struct lp_type type, - double val); +lp_build_const_vec(struct lp_type type, double val); LLVMValueRef -lp_build_int_const_scalar(struct lp_type type, - long long val); +lp_build_const_int_vec(struct lp_type type, long long val); LLVMValueRef diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.c b/src/gallium/auxiliary/gallivm/lp_bld_conv.c index f77cf78721..3f7f2ebde9 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_conv.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.c @@ -114,13 +114,13 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, scale = (double)mask/ubound; bias = (double)((unsigned long long)1 << (mantissa - n)); - res = LLVMBuildMul(builder, src, lp_build_const_scalar(src_type, scale), ""); - res = LLVMBuildAdd(builder, res, lp_build_const_scalar(src_type, bias), ""); + res = LLVMBuildMul(builder, src, lp_build_const_vec(src_type, scale), ""); + res = LLVMBuildAdd(builder, res, lp_build_const_vec(src_type, bias), ""); res = LLVMBuildBitCast(builder, res, int_vec_type, ""); if(dst_width > n) { int shift = dst_width - n; - res = LLVMBuildShl(builder, res, lp_build_int_const_scalar(src_type, shift), ""); + res = LLVMBuildShl(builder, res, lp_build_const_int_vec(src_type, shift), ""); /* TODO: Fill in the empty lower bits for additional precision? */ /* YES: this fixes progs/trivial/tri-z-eq.c. @@ -130,21 +130,21 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, #if 0 { LLVMValueRef msb; - msb = LLVMBuildLShr(builder, res, lp_build_int_const_scalar(src_type, dst_width - 1), ""); - msb = LLVMBuildShl(builder, msb, lp_build_int_const_scalar(src_type, shift), ""); - msb = LLVMBuildSub(builder, msb, lp_build_int_const_scalar(src_type, 1), ""); + msb = LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, dst_width - 1), ""); + msb = LLVMBuildShl(builder, msb, lp_build_const_int_vec(src_type, shift), ""); + msb = LLVMBuildSub(builder, msb, lp_build_const_int_vec(src_type, 1), ""); res = LLVMBuildOr(builder, res, msb, ""); } #elif 0 while(shift > 0) { - res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_int_const_scalar(src_type, n), ""), ""); + res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, n), ""), ""); shift -= n; n *= 2; } #endif } else - res = LLVMBuildAnd(builder, res, lp_build_int_const_scalar(src_type, mask), ""); + res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(src_type, mask), ""); return res; } @@ -183,10 +183,10 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, if(src_width > mantissa) { int shift = src_width - mantissa; - res = LLVMBuildLShr(builder, res, lp_build_int_const_scalar(dst_type, shift), ""); + res = LLVMBuildLShr(builder, res, lp_build_const_int_vec(dst_type, shift), ""); } - bias_ = lp_build_const_scalar(dst_type, bias); + bias_ = lp_build_const_vec(dst_type, bias); res = LLVMBuildOr(builder, res, @@ -195,7 +195,7 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, res = LLVMBuildBitCast(builder, res, vec_type, ""); res = LLVMBuildSub(builder, res, bias_, ""); - res = LLVMBuildMul(builder, res, lp_build_const_scalar(dst_type, scale), ""); + res = LLVMBuildMul(builder, res, lp_build_const_vec(dst_type, scale), ""); return res; } @@ -251,7 +251,7 @@ lp_build_conv(LLVMBuilderRef builder, if(dst_min == 0.0) thres = bld.zero; else - thres = lp_build_const_scalar(src_type, dst_min); + thres = lp_build_const_vec(src_type, dst_min); for(i = 0; i < num_tmps; ++i) tmp[i] = lp_build_max(&bld, tmp[i], thres); } @@ -260,7 +260,7 @@ lp_build_conv(LLVMBuilderRef builder, if(dst_max == 1.0) thres = bld.one; else - thres = lp_build_const_scalar(src_type, dst_max); + thres = lp_build_const_vec(src_type, dst_max); for(i = 0; i < num_tmps; ++i) tmp[i] = lp_build_min(&bld, tmp[i], thres); } @@ -288,7 +288,7 @@ lp_build_conv(LLVMBuilderRef builder, LLVMTypeRef tmp_vec_type; if (dst_scale != 1.0) { - LLVMValueRef scale = lp_build_const_scalar(tmp_type, dst_scale); + LLVMValueRef scale = lp_build_const_vec(tmp_type, dst_scale); for(i = 0; i < num_tmps; ++i) tmp[i] = LLVMBuildMul(builder, tmp[i], scale, ""); } @@ -315,7 +315,7 @@ lp_build_conv(LLVMBuilderRef builder, /* FIXME: compensate different offsets too */ if(src_shift > dst_shift) { - LLVMValueRef shift = lp_build_int_const_scalar(tmp_type, src_shift - dst_shift); + LLVMValueRef shift = lp_build_const_int_vec(tmp_type, src_shift - dst_shift); for(i = 0; i < num_tmps; ++i) if(src_type.sign) tmp[i] = LLVMBuildAShr(builder, tmp[i], shift, ""); @@ -388,7 +388,7 @@ lp_build_conv(LLVMBuilderRef builder, } if (src_scale != 1.0) { - LLVMValueRef scale = lp_build_const_scalar(tmp_type, 1.0/src_scale); + LLVMValueRef scale = lp_build_const_vec(tmp_type, 1.0/src_scale); for(i = 0; i < num_tmps; ++i) tmp[i] = LLVMBuildMul(builder, tmp[i], scale, ""); } @@ -400,7 +400,7 @@ lp_build_conv(LLVMBuilderRef builder, /* FIXME: compensate different offsets too */ if(src_shift < dst_shift) { - LLVMValueRef shift = lp_build_int_const_scalar(tmp_type, dst_shift - src_shift); + LLVMValueRef shift = lp_build_const_int_vec(tmp_type, dst_shift - src_shift); for(i = 0; i < num_tmps; ++i) tmp[i] = LLVMBuildShl(builder, tmp[i], shift, ""); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_conv.h b/src/gallium/auxiliary/gallivm/lp_bld_conv.h index 78e8155ff7..628831c3ad 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_conv.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_conv.h @@ -37,7 +37,7 @@ #define LP_BLD_CONV_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct lp_type; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_debug.h b/src/gallium/auxiliary/gallivm/lp_bld_debug.h index 441ad94786..7b010cbdb0 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_debug.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_debug.h @@ -30,7 +30,7 @@ #define LP_BLD_DEBUG_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" #include "pipe/p_compiler.h" #include "util/u_string.h" diff --git a/src/gallium/auxiliary/gallivm/lp_bld_depth.c b/src/gallium/auxiliary/gallivm/lp_bld_depth.c index f08f8eb6d8..cbc48f9865 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_depth.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_depth.c @@ -61,11 +61,104 @@ #include "util/u_format.h" #include "lp_bld_type.h" +#include "lp_bld_arit.h" #include "lp_bld_const.h" #include "lp_bld_logic.h" #include "lp_bld_flow.h" #include "lp_bld_debug.h" #include "lp_bld_depth.h" +#include "lp_bld_swizzle.h" + + + +/** + * Do the stencil test comparison (compare fb Z values against ref value. + * \param stencilVals vector of stencil values from framebuffer + * \param stencilRef the stencil reference value, replicated as a vector + * \return mask of pass/fail values + */ +static LLVMValueRef +lp_build_stencil_test(struct lp_build_context *bld, + const struct pipe_stencil_state *stencil, + LLVMValueRef stencilVals, + LLVMValueRef stencilRef) +{ + const unsigned stencilMax = 255; /* XXX fix */ + struct lp_type type = bld->type; + LLVMValueRef res; + + assert(stencil->enabled); + + if (stencil->valuemask != stencilMax) { + /* compute stencilRef = stencilRef & valuemask */ + LLVMValueRef valuemask = lp_build_const_int_vec(type, stencil->valuemask); + stencilRef = LLVMBuildAnd(bld->builder, stencilRef, valuemask, ""); + /* compute stencilVals = stencilVals & valuemask */ + stencilVals = LLVMBuildAnd(bld->builder, stencilVals, valuemask, ""); + } + + res = lp_build_compare(bld->builder, bld->type, stencil->func, + stencilVals, stencilRef); + + return res; +} + + +/** + * Apply the stencil operator (add/sub/keep/etc) to the given vector + * of stencil values. + * \return new stencil values vector + */ +static LLVMValueRef +lp_build_stencil_op(struct lp_build_context *bld, + unsigned stencil_op, + LLVMValueRef stencilRef, + const struct pipe_stencil_state *stencil, + LLVMValueRef stencilVals) + +{ + const unsigned stencilMax = 255; /* XXX fix */ + struct lp_type type = bld->type; + LLVMValueRef res; + LLVMValueRef max = lp_build_const_int_vec(type, stencilMax); + + switch (stencil_op) { + case PIPE_STENCIL_OP_KEEP: + res = stencilVals; + case PIPE_STENCIL_OP_ZERO: + res = bld->zero; + case PIPE_STENCIL_OP_REPLACE: + res = lp_build_broadcast_scalar(bld, stencilRef); + case PIPE_STENCIL_OP_INCR: + res = lp_build_add(bld, stencilVals, bld->one); + res = lp_build_min(bld, res, max); + case PIPE_STENCIL_OP_DECR: + res = lp_build_sub(bld, stencilVals, bld->one); + res = lp_build_max(bld, res, bld->zero); + case PIPE_STENCIL_OP_INCR_WRAP: + res = lp_build_add(bld, stencilVals, bld->one); + res = LLVMBuildAnd(bld->builder, res, max, ""); + case PIPE_STENCIL_OP_DECR_WRAP: + res = lp_build_sub(bld, stencilVals, bld->one); + res = LLVMBuildAnd(bld->builder, res, max, ""); + case PIPE_STENCIL_OP_INVERT: + res = LLVMBuildNot(bld->builder, stencilVals, ""); + default: + assert(0 && "bad stencil op mode"); + res = NULL; + } + + if (stencil->writemask != stencilMax) { + /* compute res = (res & mask) | (stencilVals & ~mask) */ + LLVMValueRef mask = lp_build_const_int_vec(type, stencil->writemask); + LLVMValueRef cmask = LLVMBuildNot(bld->builder, mask, "notWritemask"); + LLVMValueRef t1 = LLVMBuildAnd(bld->builder, res, mask, "t1"); + LLVMValueRef t2 = LLVMBuildAnd(bld->builder, stencilVals, cmask, "t2"); + res = LLVMBuildOr(bld->builder, t1, t2, "t1_or_t2"); + } + + return res; +} /** @@ -109,7 +202,14 @@ lp_depth_type(const struct util_format_description *format_desc, /** - * Depth test. + * Generate code for performing depth and/or stencil tests. + * We operate on a vector of values (typically a 2x2 quad). + * + * \param type the data type of the fragment depth/stencil values + * \param format_desc description of the depth/stencil surface + * \param mask the alive/dead pixel mask for the quad + * \param src the incoming depth/stencil values (a 2x2 quad) + * \param dst_ptr the outgoing/updated depth/stencil values */ void lp_build_depth_test(LLVMBuilderRef builder, @@ -126,6 +226,9 @@ lp_build_depth_test(LLVMBuilderRef builder, LLVMValueRef z_bitmask = NULL; LLVMValueRef test; + (void) lp_build_stencil_test; + (void) lp_build_stencil_op; + if(!state->enabled) return; @@ -185,11 +288,11 @@ lp_build_depth_test(LLVMBuilderRef builder, if(padding_left || padding_right) { const unsigned long long mask_left = ((unsigned long long)1 << (format_desc->block.bits - padding_left)) - 1; const unsigned long long mask_right = ((unsigned long long)1 << (padding_right)) - 1; - z_bitmask = lp_build_int_const_scalar(type, mask_left ^ mask_right); + z_bitmask = lp_build_const_int_vec(type, mask_left ^ mask_right); } if(padding_left) - src = LLVMBuildLShr(builder, src, lp_build_int_const_scalar(type, padding_left), ""); + src = LLVMBuildLShr(builder, src, lp_build_const_int_vec(type, padding_left), ""); if(padding_right) src = LLVMBuildAnd(builder, src, z_bitmask, ""); if(padding_left || padding_right) @@ -198,6 +301,7 @@ lp_build_depth_test(LLVMBuilderRef builder, lp_build_name(dst, "zsbuf.z"); + /* compare src Z to dst Z, returning 'pass' mask */ test = lp_build_cmp(&bld, state->func, src, dst); lp_build_mask_update(mask, test); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_depth.h b/src/gallium/auxiliary/gallivm/lp_bld_depth.h index 8be80024ae..8375824cbf 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_depth.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_depth.h @@ -36,7 +36,7 @@ #define LP_BLD_DEPTH_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct pipe_depth_state; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.c b/src/gallium/auxiliary/gallivm/lp_bld_flow.c index c2f35419ec..106fc03e46 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_flow.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.c @@ -648,7 +648,9 @@ lp_build_if(struct lp_build_if_state *ctx, ifthen->phi[i] = LLVMBuildPhi(builder, LLVMTypeOf(*flow->variables[i]), ""); /* add add the initial value of the var from the entry block */ - LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &ifthen->entry_block, 1); + if (!LLVMIsUndef(*flow->variables[i])) + LLVMAddIncoming(ifthen->phi[i], flow->variables[i], + &ifthen->entry_block, 1); } /* create/insert true_block before merge_block */ @@ -695,18 +697,21 @@ lp_build_endif(struct lp_build_if_state *ctx) { struct lp_build_flow_context *flow = ctx->flow; struct lp_build_flow_if *ifthen; + LLVMBasicBlockRef curBlock = LLVMGetInsertBlock(ctx->builder); unsigned i; ifthen = &lp_build_flow_pop(flow, LP_BUILD_FLOW_IF)->ifthen; assert(ifthen); + /* Insert branch to the merge block from current block */ + LLVMBuildBr(ctx->builder, ifthen->merge_block); + if (ifthen->false_block) { LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block); /* for each variable, update the Phi node with a (variable, block) pair */ for (i = 0; i < flow->num_variables; i++) { assert(*flow->variables[i]); - LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &ifthen->false_block, 1); - + LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &curBlock, 1); /* replace the variable ref with the phi function */ *flow->variables[i] = ifthen->phi[i]; } @@ -742,15 +747,18 @@ lp_build_endif(struct lp_build_if_state *ctx) ifthen->true_block, ifthen->merge_block); } - /* Append an unconditional Br(anch) instruction on the true_block */ - LLVMPositionBuilderAtEnd(ctx->builder, ifthen->true_block); - LLVMBuildBr(ctx->builder, ifthen->merge_block); + /* Insert branch from end of true_block to merge_block */ if (ifthen->false_block) { - /* Append an unconditional Br(anch) instruction on the false_block */ - LLVMPositionBuilderAtEnd(ctx->builder, ifthen->false_block); + /* Append an unconditional Br(anch) instruction on the true_block */ + LLVMPositionBuilderAtEnd(ctx->builder, ifthen->true_block); LLVMBuildBr(ctx->builder, ifthen->merge_block); } - + else { + /* No else clause. + * Note that we've already inserted the branch at the end of + * true_block. See the very first LLVMBuildBr() call in this function. + */ + } /* Resume building code at end of the ifthen->merge_block */ LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_flow.h b/src/gallium/auxiliary/gallivm/lp_bld_flow.h index e158836549..c2b50e1b60 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_flow.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_flow.h @@ -35,7 +35,7 @@ #define LP_BLD_FLOW_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct lp_type; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format.h b/src/gallium/auxiliary/gallivm/lp_bld_format.h index 8972c0dc17..73ab6de3f2 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_format.h @@ -34,7 +34,7 @@ * Pixel format helpers. */ -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" #include "pipe/p_format.h" diff --git a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c index abb27e4c32..45ee4b12ce 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_format_soa.c @@ -114,10 +114,10 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, case UTIL_FORMAT_TYPE_UNSIGNED: if(type.floating) { if(start) - input = LLVMBuildLShr(builder, input, lp_build_int_const_scalar(type, start), ""); + input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(type, start), ""); if(stop < format_desc->block.bits) { unsigned mask = ((unsigned long long)1 << width) - 1; - input = LLVMBuildAnd(builder, input, lp_build_int_const_scalar(type, mask), ""); + input = LLVMBuildAnd(builder, input, lp_build_const_int_vec(type, mask), ""); } if(format_desc->channel[chan].normalized) diff --git a/src/gallium/auxiliary/gallivm/lp_bld_interp.c b/src/gallium/auxiliary/gallivm/lp_bld_interp.c index 2fc894017d..09efb16121 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_interp.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_interp.c @@ -289,17 +289,17 @@ pos_update(struct lp_build_interp_soa_context *bld, int quad_index) /* top-right or bottom-right quad in block */ /* build x += xstep */ x = lp_build_add(&bld->base, x, - lp_build_const_scalar(bld->base.type, xstep)); + lp_build_const_vec(bld->base.type, xstep)); } if (quad_index == 2) { /* bottom-left quad in block */ /* build y += ystep */ y = lp_build_add(&bld->base, y, - lp_build_const_scalar(bld->base.type, ystep)); + lp_build_const_vec(bld->base.type, ystep)); /* build x -= xstep */ x = lp_build_sub(&bld->base, x, - lp_build_const_scalar(bld->base.type, xstep)); + lp_build_const_vec(bld->base.type, xstep)); } lp_build_name(x, "pos.x"); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_interp.h b/src/gallium/auxiliary/gallivm/lp_bld_interp.h index 177b5e943e..a4937bbb04 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_interp.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_interp.h @@ -41,7 +41,7 @@ #define LP_BLD_INTERP_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" #include "tgsi/tgsi_exec.h" diff --git a/src/gallium/auxiliary/gallivm/lp_bld_intr.h b/src/gallium/auxiliary/gallivm/lp_bld_intr.h index 7d5506c733..977f767322 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_intr.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_intr.h @@ -37,7 +37,7 @@ #define LP_BLD_INTR_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" /** diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.c b/src/gallium/auxiliary/gallivm/lp_bld_logic.c index f3df3dd138..e2e533fa7e 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_logic.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.c @@ -193,7 +193,7 @@ lp_build_compare(LLVMBuilderRef builder, if(table[func].gt && ((type.width == 8 && type.sign) || (type.width != 8 && !type.sign))) { - LLVMValueRef msb = lp_build_int_const_scalar(type, (unsigned long long)1 << (type.width - 1)); + LLVMValueRef msb = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1)); a = LLVMBuildXor(builder, a, msb, ""); b = LLVMBuildXor(builder, b, msb, ""); } diff --git a/src/gallium/auxiliary/gallivm/lp_bld_logic.h b/src/gallium/auxiliary/gallivm/lp_bld_logic.h index b54ec13b70..d849d29e95 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_logic.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_logic.h @@ -37,7 +37,7 @@ #define LP_BLD_LOGIC_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" #include "pipe/p_defines.h" /* For PIPE_FUNC_xxx */ diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.c b/src/gallium/auxiliary/gallivm/lp_bld_pack.c index 23398f41f9..2daa8a3b58 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_pack.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.c @@ -164,7 +164,7 @@ lp_build_unpack2(LLVMBuilderRef builder, if(dst_type.sign && src_type.sign) { /* Replicate the sign bit in the most significant bits */ - msb = LLVMBuildAShr(builder, src, lp_build_int_const_scalar(src_type, src_type.width - 1), ""); + msb = LLVMBuildAShr(builder, src, lp_build_const_int_vec(src_type, src_type.width - 1), ""); } else /* Most significant bits always zero */ @@ -361,7 +361,7 @@ lp_build_packs2(LLVMBuilderRef builder, if(clamp) { struct lp_build_context bld; unsigned dst_bits = dst_type.sign ? dst_type.width - 1 : dst_type.width; - LLVMValueRef dst_max = lp_build_int_const_scalar(src_type, ((unsigned long long)1 << dst_bits) - 1); + LLVMValueRef dst_max = lp_build_const_int_vec(src_type, ((unsigned long long)1 << dst_bits) - 1); lp_build_context_init(&bld, builder, src_type); lo = lp_build_min(&bld, lo, dst_max); hi = lp_build_min(&bld, hi, dst_max); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_pack.h b/src/gallium/auxiliary/gallivm/lp_bld_pack.h index 346a17d580..41adeed220 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_pack.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_pack.h @@ -37,7 +37,7 @@ #define LP_BLD_PACK_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct lp_type; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.c b/src/gallium/auxiliary/gallivm/lp_bld_sample.c index 543fd5fea3..bb76ad4c6b 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.c @@ -84,8 +84,12 @@ lp_sampler_static_state(struct lp_sampler_static_state *state, state->wrap_t = sampler->wrap_t; state->wrap_r = sampler->wrap_r; state->min_img_filter = sampler->min_img_filter; - state->min_mip_filter = sampler->min_mip_filter; state->mag_img_filter = sampler->mag_img_filter; + if (texture->last_level) { + state->min_mip_filter = sampler->min_mip_filter; + } else { + state->min_mip_filter = PIPE_TEX_MIPFILTER_NONE; + } state->compare_mode = sampler->compare_mode; if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) { @@ -169,7 +173,7 @@ lp_build_sample_offset(struct lp_build_context *bld, LLVMValueRef x_stride; LLVMValueRef offset; - x_stride = lp_build_const_scalar(bld->type, format_desc->block.bits/8); + x_stride = lp_build_const_vec(bld->type, format_desc->block.bits/8); if(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS) { LLVMValueRef x_lo, x_hi; @@ -191,9 +195,9 @@ lp_build_sample_offset(struct lp_build_context *bld, y_hi = LLVMBuildLShr(bld->builder, y, bld->one, ""); x_stride_lo = x_stride; - y_stride_lo = lp_build_const_scalar(bld->type, 2*format_desc->block.bits/8); + y_stride_lo = lp_build_const_vec(bld->type, 2*format_desc->block.bits/8); - x_stride_hi = lp_build_const_scalar(bld->type, 4*format_desc->block.bits/8); + x_stride_hi = lp_build_const_vec(bld->type, 4*format_desc->block.bits/8); y_stride_hi = LLVMBuildShl(bld->builder, y_stride, bld->one, ""); x_offset_lo = lp_build_mul(bld, x_lo, x_stride_lo); diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample.h b/src/gallium/auxiliary/gallivm/lp_bld_sample.h index 7f08bfaac1..92f3c57435 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample.h @@ -36,7 +36,7 @@ #define LP_BLD_SAMPLE_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct pipe_texture; struct pipe_sampler_state; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c index a410688350..995c016b9d 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_sample_soa.c @@ -292,7 +292,7 @@ lp_build_sample_texel_soa(struct lp_build_sample_context *bld, int chan; for (chan = 0; chan < 4; chan++) { LLVMValueRef border_chan = - lp_build_const_scalar(bld->texel_type, + lp_build_const_vec(bld->texel_type, bld->static_state->border_color[chan]); texel[chan] = lp_build_select(&bld->texel_bld, use_border, border_chan, texel[chan]); @@ -457,8 +457,8 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, struct lp_build_context *coord_bld = &bld->coord_bld; struct lp_build_context *int_coord_bld = &bld->int_coord_bld; struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld; - LLVMValueRef two = lp_build_const_scalar(coord_bld->type, 2.0); - LLVMValueRef half = lp_build_const_scalar(coord_bld->type, 0.5); + LLVMValueRef two = lp_build_const_vec(coord_bld->type, 2.0); + LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5); LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length); LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one); LLVMValueRef length_f_minus_one = lp_build_sub(coord_bld, length_f, coord_bld->one); @@ -512,7 +512,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, else { LLVMValueRef min, max; /* clamp to [0.5, length - 0.5] */ - min = lp_build_const_scalar(coord_bld->type, 0.5F); + min = lp_build_const_vec(coord_bld->type, 0.5F); max = lp_build_sub(coord_bld, length_f, min); coord = lp_build_clamp(coord_bld, coord, min, max); } @@ -533,7 +533,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, if (bld->static_state->normalized_coords) { /* min = -1.0 / (2 * length) = -0.5 / length */ min = lp_build_mul(coord_bld, - lp_build_const_scalar(coord_bld->type, -0.5F), + lp_build_const_vec(coord_bld->type, -0.5F), lp_build_rcp(coord_bld, length_f)); /* max = 1.0 - min */ max = lp_build_sub(coord_bld, coord_bld->one, min); @@ -545,7 +545,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, } else { /* clamp to [-0.5, length + 0.5] */ - min = lp_build_const_scalar(coord_bld->type, -0.5F); + min = lp_build_const_vec(coord_bld->type, -0.5F); max = lp_build_sub(coord_bld, length_f, min); coord = lp_build_clamp(coord_bld, coord, min, max); coord = lp_build_sub(coord_bld, coord, half); @@ -620,7 +620,7 @@ lp_build_sample_wrap_linear(struct lp_build_sample_context *bld, LLVMValueRef min, max; /* min = -1.0 / (2 * length) = -0.5 / length */ min = lp_build_mul(coord_bld, - lp_build_const_scalar(coord_bld->type, -0.5F), + lp_build_const_vec(coord_bld->type, -0.5F), lp_build_rcp(coord_bld, length_f)); /* max = 1.0 - min */ max = lp_build_sub(coord_bld, coord_bld->one, min); @@ -665,7 +665,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld, struct lp_build_context *coord_bld = &bld->coord_bld; struct lp_build_context *int_coord_bld = &bld->int_coord_bld; struct lp_build_context *uint_coord_bld = &bld->uint_coord_bld; - LLVMValueRef two = lp_build_const_scalar(coord_bld->type, 2.0); + LLVMValueRef two = lp_build_const_vec(coord_bld->type, 2.0); LLVMValueRef length_f = lp_build_int_to_float(coord_bld, length); LLVMValueRef length_minus_one = lp_build_sub(uint_coord_bld, length, uint_coord_bld->one); LLVMValueRef length_f_minus_one = lp_build_sub(coord_bld, length_f, coord_bld->one); @@ -708,7 +708,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld, } else { /* clamp to [0.5, length - 0.5] */ - min = lp_build_const_scalar(coord_bld->type, 0.5F); + min = lp_build_const_vec(coord_bld->type, 0.5F); max = lp_build_sub(coord_bld, length_f, min); } /* coord = clamp(coord, min, max) */ @@ -724,7 +724,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld, if (bld->static_state->normalized_coords) { /* min = -1.0 / (2 * length) = -0.5 / length */ min = lp_build_mul(coord_bld, - lp_build_const_scalar(coord_bld->type, -0.5F), + lp_build_const_vec(coord_bld->type, -0.5F), lp_build_rcp(coord_bld, length_f)); /* max = length - min */ max = lp_build_sub(coord_bld, length_f, min); @@ -733,7 +733,7 @@ lp_build_sample_wrap_nearest(struct lp_build_sample_context *bld, } else { /* clamp to [-0.5, length + 0.5] */ - min = lp_build_const_scalar(coord_bld->type, -0.5F); + min = lp_build_const_vec(coord_bld->type, -0.5F); max = lp_build_sub(coord_bld, length_f, min); } /* coord = clamp(coord, min, max) */ @@ -843,87 +843,98 @@ lp_build_lod_selector(struct lp_build_sample_context *bld, LLVMValueRef depth) { - const int dims = texture_dims(bld->static_state->target); - struct lp_build_context *float_bld = &bld->float_bld; - LLVMValueRef lod_bias = LLVMConstReal(LLVMFloatType(), bld->static_state->lod_bias); - LLVMValueRef min_lod = LLVMConstReal(LLVMFloatType(), bld->static_state->min_lod); - LLVMValueRef max_lod = LLVMConstReal(LLVMFloatType(), bld->static_state->max_lod); - - LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0); - LLVMValueRef index1 = LLVMConstInt(LLVMInt32Type(), 1, 0); - LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0); - - LLVMValueRef s0, s1, s2; - LLVMValueRef t0, t1, t2; - LLVMValueRef r0, r1, r2; - LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy; - LLVMValueRef rho, lod; - - /* - * dsdx = abs(s[1] - s[0]); - * dsdy = abs(s[2] - s[0]); - * dtdx = abs(t[1] - t[0]); - * dtdy = abs(t[2] - t[0]); - * drdx = abs(r[1] - r[0]); - * drdy = abs(r[2] - r[0]); - * XXX we're assuming a four-element quad in 2x2 layout here. - */ - s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0"); - s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1"); - s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2"); - dsdx = LLVMBuildSub(bld->builder, s1, s0, ""); - dsdx = lp_build_abs(float_bld, dsdx); - dsdy = LLVMBuildSub(bld->builder, s2, s0, ""); - dsdy = lp_build_abs(float_bld, dsdy); - if (dims > 1) { - t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0"); - t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1"); - t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2"); - dtdx = LLVMBuildSub(bld->builder, t1, t0, ""); - dtdx = lp_build_abs(float_bld, dtdx); - dtdy = LLVMBuildSub(bld->builder, t2, t0, ""); - dtdy = lp_build_abs(float_bld, dtdy); - if (dims > 2) { - r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0"); - r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1"); - r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2"); - drdx = LLVMBuildSub(bld->builder, r1, r0, ""); - drdx = lp_build_abs(float_bld, drdx); - drdy = LLVMBuildSub(bld->builder, r2, r0, ""); - drdy = lp_build_abs(float_bld, drdy); - } + if (bld->static_state->min_lod == bld->static_state->max_lod) { + /* User is forcing sampling from a particular mipmap level. + * This is hit during mipmap generation. + */ + return LLVMConstReal(LLVMFloatType(), bld->static_state->min_lod); } + else { + const int dims = texture_dims(bld->static_state->target); + struct lp_build_context *float_bld = &bld->float_bld; + LLVMValueRef lod_bias = LLVMConstReal(LLVMFloatType(), + bld->static_state->lod_bias); + LLVMValueRef min_lod = LLVMConstReal(LLVMFloatType(), + bld->static_state->min_lod); + LLVMValueRef max_lod = LLVMConstReal(LLVMFloatType(), + bld->static_state->max_lod); + + LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0); + LLVMValueRef index1 = LLVMConstInt(LLVMInt32Type(), 1, 0); + LLVMValueRef index2 = LLVMConstInt(LLVMInt32Type(), 2, 0); + + LLVMValueRef s0, s1, s2; + LLVMValueRef t0, t1, t2; + LLVMValueRef r0, r1, r2; + LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy; + LLVMValueRef rho, lod; + + /* + * dsdx = abs(s[1] - s[0]); + * dsdy = abs(s[2] - s[0]); + * dtdx = abs(t[1] - t[0]); + * dtdy = abs(t[2] - t[0]); + * drdx = abs(r[1] - r[0]); + * drdy = abs(r[2] - r[0]); + * XXX we're assuming a four-element quad in 2x2 layout here. + */ + s0 = LLVMBuildExtractElement(bld->builder, s, index0, "s0"); + s1 = LLVMBuildExtractElement(bld->builder, s, index1, "s1"); + s2 = LLVMBuildExtractElement(bld->builder, s, index2, "s2"); + dsdx = LLVMBuildSub(bld->builder, s1, s0, ""); + dsdx = lp_build_abs(float_bld, dsdx); + dsdy = LLVMBuildSub(bld->builder, s2, s0, ""); + dsdy = lp_build_abs(float_bld, dsdy); + if (dims > 1) { + t0 = LLVMBuildExtractElement(bld->builder, t, index0, "t0"); + t1 = LLVMBuildExtractElement(bld->builder, t, index1, "t1"); + t2 = LLVMBuildExtractElement(bld->builder, t, index2, "t2"); + dtdx = LLVMBuildSub(bld->builder, t1, t0, ""); + dtdx = lp_build_abs(float_bld, dtdx); + dtdy = LLVMBuildSub(bld->builder, t2, t0, ""); + dtdy = lp_build_abs(float_bld, dtdy); + if (dims > 2) { + r0 = LLVMBuildExtractElement(bld->builder, r, index0, "r0"); + r1 = LLVMBuildExtractElement(bld->builder, r, index1, "r1"); + r2 = LLVMBuildExtractElement(bld->builder, r, index2, "r2"); + drdx = LLVMBuildSub(bld->builder, r1, r0, ""); + drdx = lp_build_abs(float_bld, drdx); + drdy = LLVMBuildSub(bld->builder, r2, r0, ""); + drdy = lp_build_abs(float_bld, drdy); + } + } - /* Compute rho = max of all partial derivatives scaled by texture size. - * XXX this could be vectorized somewhat - */ - rho = LLVMBuildMul(bld->builder, - lp_build_max(float_bld, dsdx, dsdy), - lp_build_int_to_float(float_bld, width), ""); - if (dims > 1) { - LLVMValueRef max; - max = LLVMBuildMul(bld->builder, - lp_build_max(float_bld, dtdx, dtdy), - lp_build_int_to_float(float_bld, height), ""); - rho = lp_build_max(float_bld, rho, max); - if (dims > 2) { + /* Compute rho = max of all partial derivatives scaled by texture size. + * XXX this could be vectorized somewhat + */ + rho = LLVMBuildMul(bld->builder, + lp_build_max(float_bld, dsdx, dsdy), + lp_build_int_to_float(float_bld, width), ""); + if (dims > 1) { + LLVMValueRef max; max = LLVMBuildMul(bld->builder, - lp_build_max(float_bld, drdx, drdy), - lp_build_int_to_float(float_bld, depth), ""); + lp_build_max(float_bld, dtdx, dtdy), + lp_build_int_to_float(float_bld, height), ""); rho = lp_build_max(float_bld, rho, max); + if (dims > 2) { + max = LLVMBuildMul(bld->builder, + lp_build_max(float_bld, drdx, drdy), + lp_build_int_to_float(float_bld, depth), ""); + rho = lp_build_max(float_bld, rho, max); + } } - } - /* compute lod = log2(rho) */ - lod = lp_build_log2(float_bld, rho); + /* compute lod = log2(rho) */ + lod = lp_build_log2(float_bld, rho); - /* add lod bias */ - lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "LOD bias"); + /* add lod bias */ + lod = LLVMBuildAdd(bld->builder, lod, lod_bias, "LOD bias"); - /* clamp lod */ - lod = lp_build_clamp(float_bld, lod, min_lod, max_lod); + /* clamp lod */ + lod = lp_build_clamp(float_bld, lod, min_lod, max_lod); - return lod; + return lod; + } } @@ -986,7 +997,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld, last_level); /* compute level 1 and clamp to legal range of levels */ *level1_out = lp_build_add(int_bld, *level0_out, int_bld->one); - *level1_out = lp_build_min(int_bld, *level1_out, int_bld->zero); + *level1_out = lp_build_min(int_bld, *level1_out, last_level); *weight_out = lp_build_fract(float_bld, lod); } @@ -994,6 +1005,7 @@ lp_build_linear_mip_levels(struct lp_build_sample_context *bld, /** * Generate code to sample a mipmap level with nearest filtering. + * If sampling a cube texture, r = cube face in [0,5]. */ static void lp_build_sample_image_nearest(struct lp_build_sample_context *bld, @@ -1031,6 +1043,9 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, bld->static_state->wrap_r); lp_build_name(z, "tex.z.wrapped"); } + else if (bld->static_state->target == PIPE_TEXTURE_CUBE) { + z = r; + } else { z = NULL; } @@ -1051,7 +1066,7 @@ lp_build_sample_image_nearest(struct lp_build_sample_context *bld, /** * Generate code to sample a mipmap level with linear filtering. - * 1D, 2D and 3D images are suppored. + * If sampling a cube texture, r = cube face in [0,5]. */ static void lp_build_sample_image_linear(struct lp_build_sample_context *bld, @@ -1098,8 +1113,13 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, lp_build_name(z0, "tex.z0.wrapped"); lp_build_name(z1, "tex.z1.wrapped"); } + else if (bld->static_state->target == PIPE_TEXTURE_CUBE) { + z0 = z1 = r; /* cube face */ + r_fpart = NULL; + } else { - z0 = z1 = r_fpart = NULL; + z0 = z1 = NULL; + r_fpart = NULL; } } else { @@ -1201,6 +1221,70 @@ lp_build_sample_image_linear(struct lp_build_sample_context *bld, } +/** Helper used by lp_build_cube_lookup() */ +static LLVMValueRef +lp_build_cube_ima(struct lp_build_context *coord_bld, LLVMValueRef coord) +{ + /* ima = -0.5 / abs(coord); */ + LLVMValueRef negHalf = lp_build_const_vec(coord_bld->type, -0.5); + LLVMValueRef absCoord = lp_build_abs(coord_bld, coord); + LLVMValueRef ima = lp_build_mul(coord_bld, negHalf, + lp_build_rcp(coord_bld, absCoord)); + return ima; +} + + +/** + * Helper used by lp_build_cube_lookup() + * \param sign scalar +1 or -1 + * \param coord float vector + * \param ima float vector + */ +static LLVMValueRef +lp_build_cube_coord(struct lp_build_context *coord_bld, + LLVMValueRef sign, int negate_coord, + LLVMValueRef coord, LLVMValueRef ima) +{ + /* return negate(coord) * ima * sign + 0.5; */ + LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5); + LLVMValueRef res; + + assert(negate_coord == +1 || negate_coord == -1); + + if (negate_coord == -1) { + coord = lp_build_negate(coord_bld, coord); + } + + res = lp_build_mul(coord_bld, coord, ima); + if (sign) { + sign = lp_build_broadcast_scalar(coord_bld, sign); + res = lp_build_mul(coord_bld, res, sign); + } + res = lp_build_add(coord_bld, res, half); + + return res; +} + + +/** Helper used by lp_build_cube_lookup() + * Return (major_coord >= 0) ? pos_face : neg_face; + */ +static LLVMValueRef +lp_build_cube_face(struct lp_build_sample_context *bld, + LLVMValueRef major_coord, + unsigned pos_face, unsigned neg_face) +{ + LLVMValueRef cmp = LLVMBuildFCmp(bld->builder, LLVMRealUGE, + major_coord, + bld->float_bld.zero, ""); + LLVMValueRef pos = LLVMConstInt(LLVMInt32Type(), pos_face, 0); + LLVMValueRef neg = LLVMConstInt(LLVMInt32Type(), neg_face, 0); + LLVMValueRef res = LLVMBuildSelect(bld->builder, cmp, pos, neg, ""); + return res; +} + + + /** * Generate code to do cube face selection and per-face texcoords. */ @@ -1213,11 +1297,10 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, LLVMValueRef *face_s, LLVMValueRef *face_t) { -#if 0 struct lp_build_context *float_bld = &bld->float_bld; + struct lp_build_context *coord_bld = &bld->coord_bld; LLVMValueRef rx, ry, rz; LLVMValueRef arx, ary, arz; - LLVMValueRef sc, tc, ma; LLVMValueRef c25 = LLVMConstReal(LLVMFloatType(), 0.25); LLVMValueRef arx_ge_ary, arx_ge_arz; LLVMValueRef ary_ge_arx, ary_ge_arz; @@ -1257,34 +1340,165 @@ lp_build_cube_lookup(struct lp_build_sample_context *bld, { struct lp_build_flow_context *flow_ctx; - struct lp_build_if_state if_ctx, if2_ctx; + struct lp_build_if_state if_ctx; flow_ctx = lp_build_flow_create(bld->builder); + lp_build_flow_scope_begin(flow_ctx); + + *face_s = bld->coord_bld.undef; + *face_t = bld->coord_bld.undef; + *face = bld->int_bld.undef; + + lp_build_name(*face_s, "face_s"); + lp_build_name(*face_t, "face_t"); + lp_build_name(*face, "face"); + + lp_build_flow_scope_declare(flow_ctx, face_s); + lp_build_flow_scope_declare(flow_ctx, face_t); + lp_build_flow_scope_declare(flow_ctx, face); lp_build_if(&if_ctx, flow_ctx, bld->builder, arx_ge_ary_arz); { -#if 0 - lp_build_if(&if2_ctx, flow_ctx, bld->builder, rx_pos); + /* +/- X face */ + LLVMValueRef sign = lp_build_sgn(float_bld, rx); + LLVMValueRef ima = lp_build_cube_ima(coord_bld, s); + *face_s = lp_build_cube_coord(coord_bld, sign, +1, r, ima); + *face_t = lp_build_cube_coord(coord_bld, NULL, +1, t, ima); + *face = lp_build_cube_face(bld, rx, + PIPE_TEX_FACE_POS_X, + PIPE_TEX_FACE_NEG_X); + } + lp_build_else(&if_ctx); + { + struct lp_build_flow_context *flow_ctx2; + struct lp_build_if_state if_ctx2; + + LLVMValueRef face_s2 = bld->coord_bld.undef; + LLVMValueRef face_t2 = bld->coord_bld.undef; + LLVMValueRef face2 = bld->int_bld.undef; + + flow_ctx2 = lp_build_flow_create(bld->builder); + lp_build_flow_scope_begin(flow_ctx2); + lp_build_flow_scope_declare(flow_ctx2, &face_s2); + lp_build_flow_scope_declare(flow_ctx2, &face_t2); + lp_build_flow_scope_declare(flow_ctx2, &face2); + + ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, ""); + + lp_build_if(&if_ctx2, flow_ctx2, bld->builder, ary_ge_arx_arz); { - /* Positive X face */ + /* +/- Y face */ + LLVMValueRef sign = lp_build_sgn(float_bld, ry); + LLVMValueRef ima = lp_build_cube_ima(coord_bld, t); + face_s2 = lp_build_cube_coord(coord_bld, NULL, -1, s, ima); + face_t2 = lp_build_cube_coord(coord_bld, sign, -1, r, ima); + face2 = lp_build_cube_face(bld, ry, + PIPE_TEX_FACE_POS_Y, + PIPE_TEX_FACE_NEG_Y); } - lp_build_else(&if2_ctx); + lp_build_else(&if_ctx2); { - /* Negative X face */ + /* +/- Z face */ + LLVMValueRef sign = lp_build_sgn(float_bld, rz); + LLVMValueRef ima = lp_build_cube_ima(coord_bld, r); + face_s2 = lp_build_cube_coord(coord_bld, sign, -1, s, ima); + face_t2 = lp_build_cube_coord(coord_bld, NULL, +1, t, ima); + face2 = lp_build_cube_face(bld, rz, + PIPE_TEX_FACE_POS_Z, + PIPE_TEX_FACE_NEG_Z); } - lp_build_endif(&if2_ctx); -#endif + lp_build_endif(&if_ctx2); + lp_build_flow_scope_end(flow_ctx2); + lp_build_flow_destroy(flow_ctx2); + + *face_s = face_s2; + *face_t = face_t2; + *face = face2; } - lp_build_else(&if_ctx); - { + lp_build_endif(&if_ctx); + lp_build_flow_scope_end(flow_ctx); + lp_build_flow_destroy(flow_ctx); + } +} + + + +/** + * Sample the texture/mipmap using given image filter and mip filter. + * data0_ptr and data1_ptr point to the two mipmap levels to sample + * from. width0/1_vec, height0/1_vec, depth0/1_vec indicate their sizes. + * If we're using nearest miplevel sampling the '1' values will be null/unused. + */ +static void +lp_build_sample_mipmap(struct lp_build_sample_context *bld, + unsigned img_filter, + unsigned mip_filter, + LLVMValueRef s, + LLVMValueRef t, + LLVMValueRef r, + LLVMValueRef lod_fpart, + LLVMValueRef width0_vec, + LLVMValueRef width1_vec, + LLVMValueRef height0_vec, + LLVMValueRef height1_vec, + LLVMValueRef depth0_vec, + LLVMValueRef depth1_vec, + LLVMValueRef row_stride0_vec, + LLVMValueRef row_stride1_vec, + LLVMValueRef img_stride0_vec, + LLVMValueRef img_stride1_vec, + LLVMValueRef data_ptr0, + LLVMValueRef data_ptr1, + LLVMValueRef *colors_out) +{ + LLVMValueRef colors0[4], colors1[4]; + int chan; + + if (img_filter == PIPE_TEX_FILTER_NEAREST) { + lp_build_sample_image_nearest(bld, + width0_vec, height0_vec, depth0_vec, + row_stride0_vec, img_stride0_vec, + data_ptr0, s, t, r, colors0); + if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { + /* sample the second mipmap level, and interp */ + lp_build_sample_image_nearest(bld, + width1_vec, height1_vec, depth1_vec, + row_stride1_vec, img_stride1_vec, + data_ptr1, s, t, r, colors1); } - lp_build_endif(&if_ctx); + } + else { + assert(img_filter == PIPE_TEX_FILTER_LINEAR); - lp_build_flow_destroy(flow_ctx); + lp_build_sample_image_linear(bld, + width0_vec, height0_vec, depth0_vec, + row_stride0_vec, img_stride0_vec, + data_ptr0, s, t, r, colors0); + + if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { + /* sample the second mipmap level, and interp */ + lp_build_sample_image_linear(bld, + width1_vec, height1_vec, depth1_vec, + row_stride1_vec, img_stride1_vec, + data_ptr1, s, t, r, colors1); + } + } + + if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { + /* interpolate samples from the two mipmap levels */ + for (chan = 0; chan < 4; chan++) { + colors_out[chan] = lp_build_lerp(&bld->texel_bld, lod_fpart, + colors0[chan], colors1[chan]); + } + } + else { + /* use first/only level's colors */ + for (chan = 0; chan < 4; chan++) { + colors_out[chan] = colors0[chan]; + } } -#endif } @@ -1311,6 +1525,7 @@ lp_build_sample_general(struct lp_build_sample_context *bld, LLVMValueRef data_array, LLVMValueRef *colors_out) { + struct lp_build_context *float_bld = &bld->float_bld; const unsigned mip_filter = bld->static_state->min_mip_filter; const unsigned min_filter = bld->static_state->min_img_filter; const unsigned mag_filter = bld->static_state->mag_img_filter; @@ -1322,7 +1537,6 @@ lp_build_sample_general(struct lp_build_sample_context *bld, LLVMValueRef row_stride0_vec = NULL, row_stride1_vec = NULL; LLVMValueRef img_stride0_vec = NULL, img_stride1_vec = NULL; LLVMValueRef data_ptr0, data_ptr1; - int chan; /* printf("%s mip %d min %d mag %d\n", __FUNCTION__, @@ -1330,16 +1544,24 @@ lp_build_sample_general(struct lp_build_sample_context *bld, */ /* - * Compute the level of detail (mipmap level index(es)). + * Compute the level of detail (float). + */ + if (min_filter != mag_filter || + mip_filter != PIPE_TEX_MIPFILTER_NONE) { + /* Need to compute lod either to choose mipmap levels or to + * distinguish between minification/magnification with one mipmap level. + */ + lod = lp_build_lod_selector(bld, s, t, r, width, height, depth); + } + + /* + * Compute integer mipmap level(s) to fetch texels from. */ if (mip_filter == PIPE_TEX_MIPFILTER_NONE) { /* always use mip level 0 */ ilevel0 = LLVMConstInt(LLVMInt32Type(), 0, 0); } else { - /* compute float LOD */ - lod = lp_build_lod_selector(bld, s, t, r, width, height, depth); - if (mip_filter == PIPE_TEX_MIPFILTER_NEAREST) { lp_build_nearest_mip_level(bld, unit, lod, &ilevel0); } @@ -1366,33 +1588,41 @@ lp_build_sample_general(struct lp_build_sample_context *bld, height0_vec = lp_build_minify(bld, height_vec, ilevel0_vec); row_stride0_vec = lp_build_get_level_stride_vec(bld, row_stride_array, ilevel0); - if (dims == 3) { - depth0_vec = lp_build_minify(bld, depth_vec, ilevel0_vec); + if (dims == 3 || bld->static_state->target == PIPE_TEXTURE_CUBE) { img_stride0_vec = lp_build_mul(&bld->int_coord_bld, row_stride0_vec, height0_vec); + if (dims == 3) { + depth0_vec = lp_build_minify(bld, depth_vec, ilevel0_vec); + } } } if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { - /* compute width, height, depth for second mipmap level at ilevel1 */ + /* compute width, height, depth for second mipmap level at 'ilevel1' */ width1_vec = lp_build_minify(bld, width_vec, ilevel1_vec); if (dims >= 2) { height1_vec = lp_build_minify(bld, height_vec, ilevel1_vec); row_stride1_vec = lp_build_get_level_stride_vec(bld, row_stride_array, ilevel1); - if (dims == 3) { - depth1_vec = lp_build_minify(bld, depth_vec, ilevel1_vec); + if (dims == 3 || bld->static_state->target == PIPE_TEXTURE_CUBE) { img_stride1_vec = lp_build_mul(&bld->int_coord_bld, row_stride1_vec, height1_vec); + if (dims ==3) { + depth1_vec = lp_build_minify(bld, depth_vec, ilevel1_vec); + } } } } /* - * Choose cube face, recompute texcoords. + * Choose cube face, recompute per-face texcoords. */ if (bld->static_state->target == PIPE_TEXTURE_CUBE) { LLVMValueRef face, face_s, face_t; lp_build_cube_lookup(bld, s, t, r, &face, &face_s, &face_t); + s = face_s; /* vec */ + t = face_t; /* vec */ + /* use 'r' to indicate cube face */ + r = lp_build_broadcast_scalar(&bld->int_coord_bld, face); /* vec */ } /* @@ -1406,62 +1636,67 @@ lp_build_sample_general(struct lp_build_sample_context *bld, /* * Get/interpolate texture colors. */ - /* XXX temporarily force this path: */ - if (1 /*min_filter == mag_filter*/) { - /* same filter for minification or magnification */ - LLVMValueRef colors0[4], colors1[4]; - - if (min_filter == PIPE_TEX_FILTER_NEAREST) { - lp_build_sample_image_nearest(bld, - width0_vec, height0_vec, depth0_vec, - row_stride0_vec, img_stride0_vec, - data_ptr0, s, t, r, colors0); - - if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { - /* sample the second mipmap level, and interp */ - lp_build_sample_image_nearest(bld, - width1_vec, height1_vec, depth1_vec, - row_stride1_vec, img_stride1_vec, - data_ptr1, s, t, r, colors1); - } - } - else { - assert(min_filter == PIPE_TEX_FILTER_LINEAR); + if (min_filter == mag_filter) { + /* no need to distinquish between minification and magnification */ + lp_build_sample_mipmap(bld, min_filter, mip_filter, s, t, r, lod_fpart, + width0_vec, width1_vec, + height0_vec, height1_vec, + depth0_vec, depth1_vec, + row_stride0_vec, row_stride1_vec, + img_stride0_vec, img_stride1_vec, + data_ptr0, data_ptr1, + colors_out); + } + else { + /* Emit conditional to choose min image filter or mag image filter + * depending on the lod being >0 or <= 0, respectively. + */ + struct lp_build_flow_context *flow_ctx; + struct lp_build_if_state if_ctx; + LLVMValueRef minify; - lp_build_sample_image_linear(bld, - width0_vec, height0_vec, depth0_vec, - row_stride0_vec, img_stride0_vec, - data_ptr0, s, t, r, colors0); + flow_ctx = lp_build_flow_create(bld->builder); + lp_build_flow_scope_begin(flow_ctx); + lp_build_flow_scope_declare(flow_ctx, &colors_out[0]); + lp_build_flow_scope_declare(flow_ctx, &colors_out[1]); + lp_build_flow_scope_declare(flow_ctx, &colors_out[2]); + lp_build_flow_scope_declare(flow_ctx, &colors_out[3]); - if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { - /* sample the second mipmap level, and interp */ - lp_build_sample_image_linear(bld, - width1_vec, height1_vec, depth1_vec, - row_stride1_vec, img_stride1_vec, - data_ptr1, s, t, r, colors1); - } - } + /* minify = lod > 0.0 */ + minify = LLVMBuildFCmp(bld->builder, LLVMRealUGE, + lod, float_bld->zero, ""); - if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) { - /* interpolate samples from the two mipmap levels */ - for (chan = 0; chan < 4; chan++) { - colors_out[chan] = lp_build_lerp(&bld->texel_bld, lod_fpart, - colors0[chan], colors1[chan]); - } + lp_build_if(&if_ctx, flow_ctx, bld->builder, minify); + { + /* Use the minification filter */ + lp_build_sample_mipmap(bld, min_filter, mip_filter, + s, t, r, lod_fpart, + width0_vec, width1_vec, + height0_vec, height1_vec, + depth0_vec, depth1_vec, + row_stride0_vec, row_stride1_vec, + img_stride0_vec, img_stride1_vec, + data_ptr0, data_ptr1, + colors_out); } - else { - /* use first/only level's colors */ - for (chan = 0; chan < 4; chan++) { - colors_out[chan] = colors0[chan]; - } + lp_build_else(&if_ctx); + { + /* Use the magnification filter */ + lp_build_sample_mipmap(bld, mag_filter, mip_filter, + s, t, r, lod_fpart, + width0_vec, width1_vec, + height0_vec, height1_vec, + depth0_vec, depth1_vec, + row_stride0_vec, row_stride1_vec, + img_stride0_vec, img_stride1_vec, + data_ptr0, data_ptr1, + colors_out); } - } - else { - /* emit conditional to choose min image filter or mag image filter - * depending on the lod being >0 or <= 0, respectively. - */ - abort(); + lp_build_endif(&if_ctx); + + lp_build_flow_scope_end(flow_ctx); + lp_build_flow_destroy(flow_ctx); } } @@ -1473,7 +1708,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, LLVMValueRef packed, LLVMValueRef *rgba) { - LLVMValueRef mask = lp_build_int_const_scalar(dst_type, 0xff); + LLVMValueRef mask = lp_build_const_int_vec(dst_type, 0xff); unsigned chan; /* Decode the input vector components */ @@ -1485,7 +1720,7 @@ lp_build_rgba8_to_f32_soa(LLVMBuilderRef builder, input = packed; if(start) - input = LLVMBuildLShr(builder, input, lp_build_int_const_scalar(dst_type, start), ""); + input = LLVMBuildLShr(builder, input, lp_build_const_int_vec(dst_type, start), ""); if(stop < 32) input = LLVMBuildAnd(builder, input, mask, ""); @@ -1547,17 +1782,17 @@ lp_build_sample_2d_linear_aos(struct lp_build_sample_context *bld, t = LLVMBuildFPToSI(builder, t, i32_vec_type, ""); /* subtract 0.5 (add -128) */ - i32_c128 = lp_build_int_const_scalar(i32.type, -128); + i32_c128 = lp_build_const_int_vec(i32.type, -128); s = LLVMBuildAdd(builder, s, i32_c128, ""); t = LLVMBuildAdd(builder, t, i32_c128, ""); /* compute floor (shift right 8) */ - i32_c8 = lp_build_int_const_scalar(i32.type, 8); + i32_c8 = lp_build_const_int_vec(i32.type, 8); s_ipart = LLVMBuildAShr(builder, s, i32_c8, ""); t_ipart = LLVMBuildAShr(builder, t, i32_c8, ""); /* compute fractional part (AND with 0xff) */ - i32_c255 = lp_build_int_const_scalar(i32.type, 255); + i32_c255 = lp_build_const_int_vec(i32.type, 255); s_fpart = LLVMBuildAnd(builder, s, i32_c255, ""); t_fpart = LLVMBuildAnd(builder, t, i32_c255, ""); @@ -1724,7 +1959,7 @@ lp_build_sample_compare(struct lp_build_sample_context *bld, } assert(res); - res = lp_build_mul(texel_bld, res, lp_build_const_scalar(texel_bld->type, 0.25)); + res = lp_build_mul(texel_bld, res, lp_build_const_vec(texel_bld->type, 0.25)); /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */ for(chan = 0; chan < 3; ++chan) diff --git a/src/gallium/auxiliary/gallivm/lp_bld_struct.h b/src/gallium/auxiliary/gallivm/lp_bld_struct.h index 34478c10f5..147336edb4 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_struct.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_struct.h @@ -37,7 +37,7 @@ #define LP_BLD_STRUCT_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" #include <llvm-c/Target.h> #include "util/u_debug.h" diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c index 64e81f7b1f..278c838eac 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.c @@ -144,9 +144,9 @@ lp_build_broadcast_aos(struct lp_build_context *bld, #endif if(shift > 0) - tmp = LLVMBuildLShr(bld->builder, a, lp_build_int_const_scalar(type4, shift*type.width), ""); + tmp = LLVMBuildLShr(bld->builder, a, lp_build_const_int_vec(type4, shift*type.width), ""); if(shift < 0) - tmp = LLVMBuildShl(bld->builder, a, lp_build_int_const_scalar(type4, -shift*type.width), ""); + tmp = LLVMBuildShl(bld->builder, a, lp_build_const_int_vec(type4, -shift*type.width), ""); assert(tmp); if(tmp) diff --git a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h index 57b5cc079f..138ca620e6 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_swizzle.h @@ -37,7 +37,7 @@ #define LP_BLD_SWIZZLE_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct lp_type; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h index 0f2f8a65b1..63b938bfa9 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi.h @@ -35,7 +35,7 @@ #ifndef LP_BLD_TGSI_H #define LP_BLD_TGSI_H -#include "os/os_llvm.h" +#include "gallivm/lp_bld.h" struct tgsi_token; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c index 5ec59d636c..f160be878f 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c +++ b/src/gallium/auxiliary/gallivm/lp_bld_tgsi_soa.c @@ -475,7 +475,7 @@ emit_store( break; case TGSI_SAT_MINUS_PLUS_ONE: - value = lp_build_max(&bld->base, value, lp_build_const_scalar(bld->base.type, -1.0)); + value = lp_build_max(&bld->base, value, lp_build_const_vec(bld->base.type, -1.0)); value = lp_build_min(&bld->base, value, bld->base.one); break; @@ -996,7 +996,7 @@ emit_instruction( src0 = emit_fetch( bld, inst, 0, chan_index ); src1 = emit_fetch( bld, inst, 1, chan_index ); src2 = emit_fetch( bld, inst, 2, chan_index ); - tmp1 = lp_build_const_scalar(bld->base.type, 0.5); + tmp1 = lp_build_const_vec(bld->base.type, 0.5); tmp0 = lp_build_cmp( &bld->base, PIPE_FUNC_GREATER, src2, tmp1); dst0[chan_index] = lp_build_select( &bld->base, tmp0, src0, src1 ); } @@ -1713,7 +1713,7 @@ lp_build_tgsi_soa(LLVMBuilderRef builder, assert(num_immediates < LP_MAX_IMMEDIATES); for( i = 0; i < size; ++i ) bld.immediates[num_immediates][i] = - lp_build_const_scalar(type, parse.FullToken.FullImmediate.u[i].Float); + lp_build_const_vec(type, parse.FullToken.FullImmediate.u[i].Float); for( i = size; i < 4; ++i ) bld.immediates[num_immediates][i] = bld.base.undef; num_immediates++; diff --git a/src/gallium/auxiliary/gallivm/lp_bld_type.h b/src/gallium/auxiliary/gallivm/lp_bld_type.h index 5b351476ac..cd59d2faa6 100644 --- a/src/gallium/auxiliary/gallivm/lp_bld_type.h +++ b/src/gallium/auxiliary/gallivm/lp_bld_type.h @@ -37,9 +37,9 @@ #define LP_BLD_TYPE_H -#include "os/os_llvm.h" +#include "pipe/p_compiler.h" +#include "gallivm/lp_bld.h" -#include <pipe/p_compiler.h> /** diff --git a/src/gallium/auxiliary/os/os_time.h b/src/gallium/auxiliary/os/os_time.h index 5b55c1b374..7e0f67a76b 100644 --- a/src/gallium/auxiliary/os/os_time.h +++ b/src/gallium/auxiliary/os/os_time.h @@ -71,7 +71,7 @@ os_time_sleep(int64_t usecs); /* * Helper function for detecting time outs, taking in account overflow. * - * Returns true the the current time has elapsed beyond the specified interval. + * Returns true if the current time has elapsed beyond the specified interval. */ static INLINE boolean os_time_timeout(int64_t start, diff --git a/src/gallium/auxiliary/translate/translate_generic.c b/src/gallium/auxiliary/translate/translate_generic.c index c9ec2b32bf..c3ec9ae3f4 100644 --- a/src/gallium/auxiliary/translate/translate_generic.c +++ b/src/gallium/auxiliary/translate/translate_generic.c @@ -393,10 +393,10 @@ static fetch_func get_fetch_func( enum pipe_format format ) return &fetch_R8G8B8A8_SSCALED; case PIPE_FORMAT_B8G8R8A8_UNORM: - return &fetch_A8R8G8B8_UNORM; + return &fetch_B8G8R8A8_UNORM; case PIPE_FORMAT_A8R8G8B8_UNORM: - return &fetch_B8G8R8A8_UNORM; + return &fetch_A8R8G8B8_UNORM; case PIPE_FORMAT_R32_FIXED: return &fetch_R32_FIXED; @@ -552,10 +552,10 @@ static emit_func get_emit_func( enum pipe_format format ) return &emit_R8G8B8A8_SSCALED; case PIPE_FORMAT_B8G8R8A8_UNORM: - return &emit_A8R8G8B8_UNORM; + return &emit_B8G8R8A8_UNORM; case PIPE_FORMAT_A8R8G8B8_UNORM: - return &emit_B8G8R8A8_UNORM; + return &emit_A8R8G8B8_UNORM; default: assert(0); diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c index 03e093c11e..c13e742738 100644 --- a/src/gallium/auxiliary/translate/translate_sse.c +++ b/src/gallium/auxiliary/translate/translate_sse.c @@ -336,7 +336,7 @@ static boolean translate_attr( struct translate_sse *p, case PIPE_FORMAT_R32G32B32A32_FLOAT: emit_load_R32G32B32A32(p, dataXMM, srcECX); break; - case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_B8G8R8A8_UNORM: emit_load_R8G8B8A8_UNORM(p, dataXMM, srcECX); emit_swizzle(p, dataXMM, dataXMM, SHUF(Z,Y,X,W)); break; @@ -360,7 +360,7 @@ static boolean translate_attr( struct translate_sse *p, case PIPE_FORMAT_R32G32B32A32_FLOAT: emit_store_R32G32B32A32(p, dstEAX, dataXMM); break; - case PIPE_FORMAT_A8R8G8B8_UNORM: + case PIPE_FORMAT_B8G8R8A8_UNORM: emit_swizzle(p, dataXMM, dataXMM, SHUF(Z,Y,X,W)); emit_store_R8G8B8A8_UNORM(p, dstEAX, dataXMM); break; diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c index 94be682c4b..e997cfa8a3 100644 --- a/src/gallium/auxiliary/util/u_debug.c +++ b/src/gallium/auxiliary/util/u_debug.c @@ -421,26 +421,31 @@ void debug_dump_image(const char *prefix, #endif } -void debug_dump_surface(const char *prefix, +void debug_dump_surface(struct pipe_context *pipe, + const char *prefix, struct pipe_surface *surface) { struct pipe_texture *texture; - struct pipe_screen *screen; struct pipe_transfer *transfer; void *data; if (!surface) return; + /* XXX: this doesn't necessarily work, as the driver may be using + * temporary storage for the surface which hasn't been propagated + * back into the texture. Need to nail down the semantics of views + * and transfers a bit better before we can say if extra work needs + * to be done here: + */ texture = surface->texture; - screen = texture->screen; - transfer = screen->get_tex_transfer(screen, texture, surface->face, - surface->level, surface->zslice, - PIPE_TRANSFER_READ, 0, 0, surface->width, - surface->height); + transfer = pipe->get_tex_transfer(pipe, texture, surface->face, + surface->level, surface->zslice, + PIPE_TRANSFER_READ, 0, 0, surface->width, + surface->height); - data = screen->transfer_map(screen, transfer); + data = pipe->transfer_map(pipe, transfer); if(!data) goto error; @@ -452,13 +457,14 @@ void debug_dump_surface(const char *prefix, transfer->stride, data); - screen->transfer_unmap(screen, transfer); + pipe->transfer_unmap(pipe, transfer); error: - screen->tex_transfer_destroy(transfer); + pipe->tex_transfer_destroy(pipe, transfer); } -void debug_dump_texture(const char *prefix, +void debug_dump_texture(struct pipe_context *pipe, + const char *prefix, struct pipe_texture *texture) { struct pipe_surface *surface; @@ -473,7 +479,7 @@ void debug_dump_texture(const char *prefix, surface = screen->get_tex_surface(screen, texture, 0, 0, 0, PIPE_TEXTURE_USAGE_SAMPLER); if (surface) { - debug_dump_surface(prefix, surface); + debug_dump_surface(pipe, prefix, surface); screen->tex_surface_destroy(surface); } } @@ -511,27 +517,28 @@ struct bmp_rgb_quad { }; void -debug_dump_surface_bmp(const char *filename, +debug_dump_surface_bmp(struct pipe_context *pipe, + const char *filename, struct pipe_surface *surface) { #ifndef PIPE_SUBSYSTEM_WINDOWS_MINIPORT struct pipe_transfer *transfer; struct pipe_texture *texture = surface->texture; - struct pipe_screen *screen = texture->screen; - transfer = screen->get_tex_transfer(screen, texture, surface->face, - surface->level, surface->zslice, - PIPE_TRANSFER_READ, 0, 0, surface->width, - surface->height); + transfer = pipe->get_tex_transfer(pipe, texture, surface->face, + surface->level, surface->zslice, + PIPE_TRANSFER_READ, 0, 0, surface->width, + surface->height); - debug_dump_transfer_bmp(filename, transfer); + debug_dump_transfer_bmp(pipe, filename, transfer); - screen->tex_transfer_destroy(transfer); + pipe->tex_transfer_destroy(pipe, transfer); #endif } void -debug_dump_transfer_bmp(const char *filename, +debug_dump_transfer_bmp(struct pipe_context *pipe, + const char *filename, struct pipe_transfer *transfer) { #ifndef PIPE_SUBSYSTEM_WINDOWS_MINIPORT @@ -544,7 +551,7 @@ debug_dump_transfer_bmp(const char *filename, if(!rgba) goto error1; - pipe_get_tile_rgba(transfer, 0, 0, + pipe_get_tile_rgba(pipe, transfer, 0, 0, transfer->width, transfer->height, rgba); diff --git a/src/gallium/auxiliary/util/u_debug.h b/src/gallium/auxiliary/util/u_debug.h index 0f4768f344..98addeb372 100644 --- a/src/gallium/auxiliary/util/u_debug.h +++ b/src/gallium/auxiliary/util/u_debug.h @@ -312,6 +312,7 @@ debug_memory_end(unsigned long beginning); #ifdef DEBUG +struct pipe_context; struct pipe_surface; struct pipe_transfer; struct pipe_texture; @@ -321,21 +322,25 @@ void debug_dump_image(const char *prefix, unsigned width, unsigned height, unsigned stride, const void *data); -void debug_dump_surface(const char *prefix, +void debug_dump_surface(struct pipe_context *pipe, + const char *prefix, struct pipe_surface *surface); -void debug_dump_texture(const char *prefix, +void debug_dump_texture(struct pipe_context *pipe, + const char *prefix, struct pipe_texture *texture); -void debug_dump_surface_bmp(const char *filename, +void debug_dump_surface_bmp(struct pipe_context *pipe, + const char *filename, struct pipe_surface *surface); -void debug_dump_transfer_bmp(const char *filename, +void debug_dump_transfer_bmp(struct pipe_context *pipe, + const char *filename, struct pipe_transfer *transfer); void debug_dump_float_rgba_bmp(const char *filename, unsigned width, unsigned height, float *rgba, unsigned stride); #else #define debug_dump_image(prefix, format, cpp, width, height, stride, data) ((void)0) -#define debug_dump_surface(prefix, surface) ((void)0) -#define debug_dump_surface_bmp(filename, surface) ((void)0) +#define debug_dump_surface(pipe, prefix, surface) ((void)0) +#define debug_dump_surface_bmp(pipe, filename, surface) ((void)0) #define debug_dump_transfer_bmp(filename, transfer) ((void)0) #define debug_dump_float_rgba_bmp(filename, width, height, rgba, stride) ((void)0) #endif diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c index d421bee8ef..5c51b53d7b 100644 --- a/src/gallium/auxiliary/util/u_gen_mipmap.c +++ b/src/gallium/auxiliary/util/u_gen_mipmap.c @@ -1119,7 +1119,6 @@ make_1d_mipmap(struct gen_mipmap_state *ctx, uint face, uint baseLevel, uint lastLevel) { struct pipe_context *pipe = ctx->pipe; - struct pipe_screen *screen = pipe->screen; const uint zslice = 0; uint dstLevel; @@ -1128,27 +1127,27 @@ make_1d_mipmap(struct gen_mipmap_state *ctx, struct pipe_transfer *srcTrans, *dstTrans; void *srcMap, *dstMap; - srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, + srcTrans = pipe->get_tex_transfer(pipe, pt, face, srcLevel, zslice, PIPE_TRANSFER_READ, 0, 0, u_minify(pt->width0, srcLevel), u_minify(pt->height0, srcLevel)); - dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, + dstTrans = pipe->get_tex_transfer(pipe, pt, face, dstLevel, zslice, PIPE_TRANSFER_WRITE, 0, 0, u_minify(pt->width0, dstLevel), u_minify(pt->height0, dstLevel)); - srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); - dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); + srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); + dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); reduce_1d(pt->format, srcTrans->width, srcMap, dstTrans->width, dstMap); - screen->transfer_unmap(screen, srcTrans); - screen->transfer_unmap(screen, dstTrans); + pipe->transfer_unmap(pipe, srcTrans); + pipe->transfer_unmap(pipe, dstTrans); - screen->tex_transfer_destroy(srcTrans); - screen->tex_transfer_destroy(dstTrans); + pipe->tex_transfer_destroy(pipe, srcTrans); + pipe->tex_transfer_destroy(pipe, dstTrans); } } @@ -1159,7 +1158,6 @@ make_2d_mipmap(struct gen_mipmap_state *ctx, uint face, uint baseLevel, uint lastLevel) { struct pipe_context *pipe = ctx->pipe; - struct pipe_screen *screen = pipe->screen; const uint zslice = 0; uint dstLevel; @@ -1171,17 +1169,17 @@ make_2d_mipmap(struct gen_mipmap_state *ctx, struct pipe_transfer *srcTrans, *dstTrans; ubyte *srcMap, *dstMap; - srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, + srcTrans = pipe->get_tex_transfer(pipe, pt, face, srcLevel, zslice, PIPE_TRANSFER_READ, 0, 0, u_minify(pt->width0, srcLevel), u_minify(pt->height0, srcLevel)); - dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, + dstTrans = pipe->get_tex_transfer(pipe, pt, face, dstLevel, zslice, PIPE_TRANSFER_WRITE, 0, 0, u_minify(pt->width0, dstLevel), u_minify(pt->height0, dstLevel)); - srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); - dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); + srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); + dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); reduce_2d(pt->format, srcTrans->width, srcTrans->height, @@ -1189,11 +1187,11 @@ make_2d_mipmap(struct gen_mipmap_state *ctx, dstTrans->width, dstTrans->height, dstTrans->stride, dstMap); - screen->transfer_unmap(screen, srcTrans); - screen->transfer_unmap(screen, dstTrans); + pipe->transfer_unmap(pipe, srcTrans); + pipe->transfer_unmap(pipe, dstTrans); - screen->tex_transfer_destroy(srcTrans); - screen->tex_transfer_destroy(dstTrans); + pipe->tex_transfer_destroy(pipe, srcTrans); + pipe->tex_transfer_destroy(pipe, dstTrans); } } @@ -1216,17 +1214,17 @@ make_3d_mipmap(struct gen_mipmap_state *ctx, struct pipe_transfer *srcTrans, *dstTrans; ubyte *srcMap, *dstMap; - srcTrans = screen->get_tex_transfer(screen, pt, face, srcLevel, zslice, + srcTrans = pipe->get_tex_transfer(pipe, pt, face, srcLevel, zslice, PIPE_TRANSFER_READ, 0, 0, u_minify(pt->width0, srcLevel), u_minify(pt->height0, srcLevel)); - dstTrans = screen->get_tex_transfer(screen, pt, face, dstLevel, zslice, + dstTrans = pipe->get_tex_transfer(pipe, pt, face, dstLevel, zslice, PIPE_TRANSFER_WRITE, 0, 0, u_minify(pt->width0, dstLevel), u_minify(pt->height0, dstLevel)); - srcMap = (ubyte *) screen->transfer_map(screen, srcTrans); - dstMap = (ubyte *) screen->transfer_map(screen, dstTrans); + srcMap = (ubyte *) pipe->transfer_map(pipe, srcTrans); + dstMap = (ubyte *) pipe->transfer_map(pipe, dstTrans); reduce_3d(pt->format, srcTrans->width, srcTrans->height, @@ -1234,11 +1232,11 @@ make_3d_mipmap(struct gen_mipmap_state *ctx, dstTrans->width, dstTrans->height, dstTrans->stride, dstMap); - screen->transfer_unmap(screen, srcTrans); - screen->transfer_unmap(screen, dstTrans); + pipe->transfer_unmap(pipe, srcTrans); + pipe->transfer_unmap(pipe, dstTrans); - screen->tex_transfer_destroy(srcTrans); - screen->tex_transfer_destroy(dstTrans); + pipe->tex_transfer_destroy(pipe, srcTrans); + pipe->tex_transfer_destroy(pipe, dstTrans); } #else (void) reduce_3d; diff --git a/src/gallium/auxiliary/util/u_inlines.h b/src/gallium/auxiliary/util/u_inlines.h index 0cb3432c6e..e7255e3baa 100644 --- a/src/gallium/auxiliary/util/u_inlines.h +++ b/src/gallium/auxiliary/util/u_inlines.h @@ -264,24 +264,24 @@ pipe_buffer_read(struct pipe_screen *screen, } static INLINE void * -pipe_transfer_map( struct pipe_transfer *transf ) +pipe_transfer_map( struct pipe_context *context, + struct pipe_transfer *transf ) { - struct pipe_screen *screen = transf->texture->screen; - return screen->transfer_map(screen, transf); + return context->transfer_map(context, transf); } static INLINE void -pipe_transfer_unmap( struct pipe_transfer *transf ) +pipe_transfer_unmap( struct pipe_context *context, + struct pipe_transfer *transf ) { - struct pipe_screen *screen = transf->texture->screen; - screen->transfer_unmap(screen, transf); + context->transfer_unmap(context, transf); } static INLINE void -pipe_transfer_destroy( struct pipe_transfer *transf ) +pipe_transfer_destroy( struct pipe_context *context, + struct pipe_transfer *transfer ) { - struct pipe_screen *screen = transf->texture->screen; - screen->tex_transfer_destroy(transf); + context->tex_transfer_destroy(context, transfer); } static INLINE unsigned diff --git a/src/gallium/auxiliary/util/u_rect.c b/src/gallium/auxiliary/util/u_rect.c index 8479161c74..e73797f1b7 100644 --- a/src/gallium/auxiliary/util/u_rect.c +++ b/src/gallium/auxiliary/util/u_rect.c @@ -169,7 +169,6 @@ util_surface_copy(struct pipe_context *pipe, unsigned src_x, unsigned src_y, unsigned w, unsigned h) { - struct pipe_screen *screen = pipe->screen; struct pipe_transfer *src_trans, *dst_trans; void *dst_map; const void *src_map; @@ -182,7 +181,7 @@ util_surface_copy(struct pipe_context *pipe, src_format = src->texture->format; dst_format = dst->texture->format; - src_trans = screen->get_tex_transfer(screen, + src_trans = pipe->get_tex_transfer(pipe, src->texture, src->face, src->level, @@ -190,7 +189,7 @@ util_surface_copy(struct pipe_context *pipe, PIPE_TRANSFER_READ, src_x, src_y, w, h); - dst_trans = screen->get_tex_transfer(screen, + dst_trans = pipe->get_tex_transfer(pipe, dst->texture, dst->face, dst->level, @@ -202,8 +201,8 @@ util_surface_copy(struct pipe_context *pipe, assert(util_format_get_blockwidth(dst_format) == util_format_get_blockwidth(src_format)); assert(util_format_get_blockheight(dst_format) == util_format_get_blockheight(src_format)); - src_map = pipe->screen->transfer_map(screen, src_trans); - dst_map = pipe->screen->transfer_map(screen, dst_trans); + src_map = pipe->transfer_map(pipe, src_trans); + dst_map = pipe->transfer_map(pipe, dst_trans); assert(src_map); assert(dst_map); @@ -221,11 +220,11 @@ util_surface_copy(struct pipe_context *pipe, do_flip ? h - 1 : 0); } - pipe->screen->transfer_unmap(pipe->screen, src_trans); - pipe->screen->transfer_unmap(pipe->screen, dst_trans); + pipe->transfer_unmap(pipe, src_trans); + pipe->transfer_unmap(pipe, dst_trans); - screen->tex_transfer_destroy(src_trans); - screen->tex_transfer_destroy(dst_trans); + pipe->tex_transfer_destroy(pipe, src_trans); + pipe->tex_transfer_destroy(pipe, dst_trans); } @@ -243,14 +242,13 @@ util_surface_fill(struct pipe_context *pipe, unsigned dstx, unsigned dsty, unsigned width, unsigned height, unsigned value) { - struct pipe_screen *screen = pipe->screen; struct pipe_transfer *dst_trans; void *dst_map; assert(dst->texture); if (!dst->texture) return; - dst_trans = screen->get_tex_transfer(screen, + dst_trans = pipe->get_tex_transfer(pipe, dst->texture, dst->face, dst->level, @@ -258,7 +256,7 @@ util_surface_fill(struct pipe_context *pipe, PIPE_TRANSFER_WRITE, dstx, dsty, width, height); - dst_map = pipe->screen->transfer_map(screen, dst_trans); + dst_map = pipe->transfer_map(pipe, dst_trans); assert(dst_map); @@ -302,6 +300,6 @@ util_surface_fill(struct pipe_context *pipe, } } - pipe->screen->transfer_unmap(pipe->screen, dst_trans); - screen->tex_transfer_destroy(dst_trans); + pipe->transfer_unmap(pipe, dst_trans); + pipe->tex_transfer_destroy(pipe, dst_trans); } diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c index 79481b710b..e445895efc 100644 --- a/src/gallium/auxiliary/util/u_tile.c +++ b/src/gallium/auxiliary/util/u_tile.c @@ -45,11 +45,11 @@ * Move raw block of pixels from transfer object to user memory. */ void -pipe_get_tile_raw(struct pipe_transfer *pt, +pipe_get_tile_raw(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, void *dst, int dst_stride) { - struct pipe_screen *screen = pt->texture->screen; const void *src; if (dst_stride == 0) @@ -58,14 +58,14 @@ pipe_get_tile_raw(struct pipe_transfer *pt, if (pipe_clip_tile(x, y, &w, &h, pt)) return; - src = screen->transfer_map(screen, pt); + src = pipe->transfer_map(pipe, pt); assert(src); if(!src) return; util_copy_rect(dst, pt->texture->format, dst_stride, 0, 0, w, h, src, pt->stride, x, y); - screen->transfer_unmap(screen, pt); + pipe->transfer_unmap(pipe, pt); } @@ -73,11 +73,11 @@ pipe_get_tile_raw(struct pipe_transfer *pt, * Move raw block of pixels from user memory to transfer object. */ void -pipe_put_tile_raw(struct pipe_transfer *pt, +pipe_put_tile_raw(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, const void *src, int src_stride) { - struct pipe_screen *screen = pt->texture->screen; void *dst; enum pipe_format format = pt->texture->format; @@ -87,14 +87,14 @@ pipe_put_tile_raw(struct pipe_transfer *pt, if (pipe_clip_tile(x, y, &w, &h, pt)) return; - dst = screen->transfer_map(screen, pt); + dst = pipe->transfer_map(pipe, pt); assert(dst); if(!dst) return; util_copy_rect(dst, format, pt->stride, x, y, w, h, src, src_stride, 0, 0); - screen->transfer_unmap(screen, pt); + pipe->transfer_unmap(pipe, pt); } @@ -1246,7 +1246,8 @@ pipe_tile_raw_to_rgba(enum pipe_format format, void -pipe_get_tile_rgba(struct pipe_transfer *pt, +pipe_get_tile_rgba(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, float *p) { @@ -1265,7 +1266,7 @@ pipe_get_tile_rgba(struct pipe_transfer *pt, if(format == PIPE_FORMAT_UYVY || format == PIPE_FORMAT_YUYV) assert((x & 1) == 0); - pipe_get_tile_raw(pt, x, y, w, h, packed, 0); + pipe_get_tile_raw(pipe, pt, x, y, w, h, packed, 0); pipe_tile_raw_to_rgba(format, packed, w, h, p, dst_stride); @@ -1274,7 +1275,8 @@ pipe_get_tile_rgba(struct pipe_transfer *pt, void -pipe_put_tile_rgba(struct pipe_transfer *pt, +pipe_put_tile_rgba(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, const float *p) { @@ -1363,7 +1365,7 @@ pipe_put_tile_rgba(struct pipe_transfer *pt, 0, 0, w, h); } - pipe_put_tile_raw(pt, x, y, w, h, packed, 0); + pipe_put_tile_raw(pipe, pt, x, y, w, h, packed, 0); FREE(packed); } @@ -1373,11 +1375,11 @@ pipe_put_tile_rgba(struct pipe_transfer *pt, * Get a block of Z values, converted to 32-bit range. */ void -pipe_get_tile_z(struct pipe_transfer *pt, +pipe_get_tile_z(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, uint *z) { - struct pipe_screen *screen = pt->texture->screen; const uint dstStride = w; ubyte *map; uint *pDest = z; @@ -1387,7 +1389,7 @@ pipe_get_tile_z(struct pipe_transfer *pt, if (pipe_clip_tile(x, y, &w, &h, pt)) return; - map = (ubyte *)screen->transfer_map(screen, pt); + map = (ubyte *)pipe->transfer_map(pipe, pt); if (!map) { assert(0); return; @@ -1453,16 +1455,16 @@ pipe_get_tile_z(struct pipe_transfer *pt, assert(0); } - screen->transfer_unmap(screen, pt); + pipe->transfer_unmap(pipe, pt); } void -pipe_put_tile_z(struct pipe_transfer *pt, +pipe_put_tile_z(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, const uint *zSrc) { - struct pipe_screen *screen = pt->texture->screen; const uint srcStride = w; const uint *ptrc = zSrc; ubyte *map; @@ -1472,7 +1474,7 @@ pipe_put_tile_z(struct pipe_transfer *pt, if (pipe_clip_tile(x, y, &w, &h, pt)) return; - map = (ubyte *)screen->transfer_map(screen, pt); + map = (ubyte *)pipe->transfer_map(pipe, pt); if (!map) { assert(0); return; @@ -1560,7 +1562,7 @@ pipe_put_tile_z(struct pipe_transfer *pt, assert(0); } - screen->transfer_unmap(screen, pt); + pipe->transfer_unmap(pipe, pt); } diff --git a/src/gallium/auxiliary/util/u_tile.h b/src/gallium/auxiliary/util/u_tile.h index 1453af38b8..8329087cfa 100644 --- a/src/gallium/auxiliary/util/u_tile.h +++ b/src/gallium/auxiliary/util/u_tile.h @@ -56,34 +56,40 @@ extern "C" { #endif void -pipe_get_tile_raw(struct pipe_transfer *pt, +pipe_get_tile_raw(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, void *p, int dst_stride); void -pipe_put_tile_raw(struct pipe_transfer *pt, +pipe_put_tile_raw(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, const void *p, int src_stride); void -pipe_get_tile_rgba(struct pipe_transfer *pt, +pipe_get_tile_rgba(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, float *p); void -pipe_put_tile_rgba(struct pipe_transfer *pt, +pipe_put_tile_rgba(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, const float *p); void -pipe_get_tile_z(struct pipe_transfer *pt, +pipe_get_tile_z(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, uint *z); void -pipe_put_tile_z(struct pipe_transfer *pt, +pipe_put_tile_z(struct pipe_context *pipe, + struct pipe_transfer *pt, uint x, uint y, uint w, uint h, const uint *z); diff --git a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c index 0763b5bb0e..beb4722901 100644 --- a/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c +++ b/src/gallium/auxiliary/vl/vl_mpeg12_mc_renderer.c @@ -680,14 +680,14 @@ xfer_buffers_map(struct vl_mpeg12_mc_renderer *r) assert(r); for (i = 0; i < 3; ++i) { - r->tex_transfer[i] = r->pipe->screen->get_tex_transfer + r->tex_transfer[i] = r->pipe->get_tex_transfer ( - r->pipe->screen, r->textures.all[i], + r->pipe, r->textures.all[i], 0, 0, 0, PIPE_TRANSFER_WRITE, 0, 0, r->textures.all[i]->width0, r->textures.all[i]->height0 ); - r->texels[i] = r->pipe->screen->transfer_map(r->pipe->screen, r->tex_transfer[i]); + r->texels[i] = r->pipe->transfer_map(r->pipe, r->tex_transfer[i]); } } @@ -699,8 +699,8 @@ xfer_buffers_unmap(struct vl_mpeg12_mc_renderer *r) assert(r); for (i = 0; i < 3; ++i) { - r->pipe->screen->transfer_unmap(r->pipe->screen, r->tex_transfer[i]); - r->pipe->screen->tex_transfer_destroy(r->tex_transfer[i]); + r->pipe->transfer_unmap(r->pipe, r->tex_transfer[i]); + r->pipe->tex_transfer_destroy(r->pipe, r->tex_transfer[i]); } } |