diff options
| author | Michal Krol <michal@vmware.com> | 2009-09-17 12:44:24 +0200 |
|---|---|---|
| committer | Michal Krol <michal@vmware.com> | 2009-09-17 12:44:24 +0200 |
| commit | 2a661c383fee65bc4413541e706925fa3e9b9cf5 (patch) | |
| tree | dbea0c9d65d17b81720fe2f161604dfc91eb0546 /src/gallium/drivers | |
| parent | 90daefd1c474a6e0502df5053b581987c12b8673 (diff) | |
| parent | 21caa29fbd332a2ee05a58df91e1664fbbc4e61f (diff) | |
Merge commit 'origin/master' into glsl-pp-rework-2
Conflicts:
src/gallium/winsys/gdi/SConscript
Diffstat (limited to 'src/gallium/drivers')
58 files changed, 1254 insertions, 836 deletions
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c index bd48ce7005..9161747fdb 100644 --- a/src/gallium/drivers/cell/ppu/cell_screen.c +++ b/src/gallium/drivers/cell/ppu/cell_screen.c @@ -64,8 +64,6 @@ cell_get_param(struct pipe_screen *screen, int param) return 1; case PIPE_CAP_GLSL: return 1; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 0; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/i915simple/i915_screen.c b/src/gallium/drivers/i915simple/i915_screen.c index 9f017a14cc..a1dd43c1bc 100644 --- a/src/gallium/drivers/i915simple/i915_screen.c +++ b/src/gallium/drivers/i915simple/i915_screen.c @@ -101,8 +101,6 @@ i915_get_param(struct pipe_screen *screen, int param) return 1; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 0; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/i965simple/brw_screen.c b/src/gallium/drivers/i965simple/brw_screen.c index b22e105f10..fb68fd624b 100644 --- a/src/gallium/drivers/i965simple/brw_screen.c +++ b/src/gallium/drivers/i965simple/brw_screen.c @@ -85,8 +85,6 @@ brw_get_param(struct pipe_screen *screen, int param) return 1; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 0; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile index 06c586e6bb..cd7b6356d2 100644 --- a/src/gallium/drivers/llvmpipe/Makefile +++ b/src/gallium/drivers/llvmpipe/Makefile @@ -3,6 +3,8 @@ include $(TOP)/configs/current LIBNAME = llvmpipe +CFLAGS += -D__STDC_CONSTANT_MACROS -D__STDC_LIMIT_MACROS + C_SOURCES = \ lp_bld_alpha.c \ lp_bld_arit.c \ diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript index dea4b703c4..f4a9a3b22e 100644 --- a/src/gallium/drivers/llvmpipe/SConscript +++ b/src/gallium/drivers/llvmpipe/SConscript @@ -3,7 +3,7 @@ Import('*') env = env.Clone() env.Tool('llvm') -if env.has_key('LLVM_VERSION') is False: +if not env.has_key('LLVM_VERSION'): print 'warning: LLVM not found: not building llvmpipe' Return() diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c index 49c2f911af..2b4bc5c819 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.c @@ -45,7 +45,7 @@ void lp_build_alpha_test(LLVMBuilderRef builder, const struct pipe_alpha_state *state, - union lp_type type, + struct lp_type type, struct lp_build_mask_context *mask, LLVMValueRef alpha, LLVMValueRef ref) diff --git a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h index 9dbcdb4daa..634575670d 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_alpha.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_alpha.h @@ -38,14 +38,14 @@ #include <llvm-c/Core.h> struct pipe_alpha_state; -union lp_type; +struct lp_type; struct lp_build_mask_context; void lp_build_alpha_test(LLVMBuilderRef builder, const struct pipe_alpha_state *state, - union lp_type type, + struct lp_type type, struct lp_build_mask_context *mask, LLVMValueRef alpha, LLVMValueRef ref); diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.c b/src/gallium/drivers/llvmpipe/lp_bld_arit.c index ce3e5f91c0..0b115fc9b0 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_arit.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.c @@ -65,7 +65,7 @@ lp_build_min_simple(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; const char *intrinsic = NULL; LLVMValueRef cond; @@ -113,7 +113,7 @@ lp_build_max_simple(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; const char *intrinsic = NULL; LLVMValueRef cond; @@ -159,7 +159,7 @@ LLVMValueRef lp_build_comp(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; if(a == bld->one) return bld->zero; @@ -188,7 +188,7 @@ lp_build_add(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMValueRef res; if(a == bld->zero) @@ -241,7 +241,7 @@ lp_build_sub(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMValueRef res; if(b == bld->zero) @@ -405,7 +405,7 @@ lp_build_mul(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; if(a == bld->zero) return bld->zero; @@ -477,7 +477,7 @@ lp_build_div(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; if(a == bld->zero) return bld->zero; @@ -590,7 +590,7 @@ LLVMValueRef lp_build_abs(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); if(!type.sign) @@ -627,7 +627,7 @@ LLVMValueRef lp_build_sgn(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); LLVMValueRef cond; LLVMValueRef res; @@ -678,7 +678,7 @@ lp_build_round_sse41(struct lp_build_context *bld, LLVMValueRef a, enum lp_build_round_sse41_mode mode) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); const char *intrinsic; @@ -706,7 +706,7 @@ LLVMValueRef lp_build_round(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; assert(type.floating); @@ -724,7 +724,7 @@ LLVMValueRef lp_build_floor(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; assert(type.floating); @@ -742,7 +742,7 @@ LLVMValueRef lp_build_ceil(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; assert(type.floating); @@ -760,7 +760,7 @@ LLVMValueRef lp_build_trunc(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; assert(type.floating); @@ -782,7 +782,7 @@ LLVMValueRef lp_build_int(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); assert(type.floating); @@ -805,7 +805,7 @@ LLVMValueRef lp_build_sqrt(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); char intrinsic[32]; @@ -823,7 +823,7 @@ LLVMValueRef lp_build_rcp(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; if(a == bld->zero) return bld->undef; @@ -854,7 +854,7 @@ LLVMValueRef lp_build_rsqrt(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; assert(type.floating); @@ -875,7 +875,7 @@ LLVMValueRef lp_build_cos(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); char intrinsic[32]; @@ -895,7 +895,7 @@ LLVMValueRef lp_build_sin(struct lp_build_context *bld, LLVMValueRef a) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); char intrinsic[32]; @@ -966,7 +966,7 @@ lp_build_polynomial(struct lp_build_context *bld, const double *coeffs, unsigned num_coeffs) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMValueRef res = NULL; unsigned i; @@ -1014,7 +1014,7 @@ lp_build_exp2_approx(struct lp_build_context *bld, LLVMValueRef *p_frac_part, LLVMValueRef *p_exp2) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); LLVMValueRef ipart = NULL; @@ -1107,7 +1107,7 @@ lp_build_log2_approx(struct lp_build_context *bld, LLVMValueRef *p_floor_log2, LLVMValueRef *p_log2) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); diff --git a/src/gallium/drivers/llvmpipe/lp_bld_arit.h b/src/gallium/drivers/llvmpipe/lp_bld_arit.h index 5e083b847f..d68a97c4b8 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_arit.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_arit.h @@ -40,7 +40,7 @@ #include <llvm-c/Core.h> -union lp_type type; +struct lp_type type; struct lp_build_context; diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend.h b/src/gallium/drivers/llvmpipe/lp_bld_blend.h index d19e18846c..da272e549f 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_blend.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend.h @@ -46,7 +46,7 @@ struct pipe_blend_state; -union lp_type; +struct lp_type; struct lp_build_context; @@ -74,7 +74,7 @@ lp_build_blend_func(struct lp_build_context *bld, LLVMValueRef lp_build_blend_aos(LLVMBuilderRef builder, const struct pipe_blend_state *blend, - union lp_type type, + struct lp_type type, LLVMValueRef src, LLVMValueRef dst, LLVMValueRef const_, @@ -84,7 +84,7 @@ lp_build_blend_aos(LLVMBuilderRef builder, void lp_build_blend_soa(LLVMBuilderRef builder, const struct pipe_blend_state *blend, - union lp_type type, + struct lp_type type, LLVMValueRef src[4], LLVMValueRef dst[4], LLVMValueRef const_[4], diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c index c11a9398f8..d14f468ba9 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c @@ -303,7 +303,7 @@ lp_build_blend_func(struct lp_build_context *bld, LLVMValueRef lp_build_blend_aos(LLVMBuilderRef builder, const struct pipe_blend_state *blend, - union lp_type type, + struct lp_type type, LLVMValueRef src, LLVMValueRef dst, LLVMValueRef const_, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c index b92254a7d6..9511299d55 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_soa.c @@ -199,7 +199,7 @@ lp_build_blend_soa_factor(struct lp_build_blend_soa_context *bld, void lp_build_blend_soa(LLVMBuilderRef builder, const struct pipe_blend_state *blend, - union lp_type type, + struct lp_type type, LLVMValueRef src[4], LLVMValueRef dst[4], LLVMValueRef con[4], diff --git a/src/gallium/drivers/llvmpipe/lp_bld_const.c b/src/gallium/drivers/llvmpipe/lp_bld_const.c index 21487365ea..c8eaa8c394 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_const.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_const.c @@ -42,7 +42,7 @@ unsigned -lp_mantissa(union lp_type type) +lp_mantissa(struct lp_type type) { assert(type.floating); @@ -72,7 +72,7 @@ lp_mantissa(union lp_type type) * Same as lp_const_scale(), but in terms of shifts. */ unsigned -lp_const_shift(union lp_type type) +lp_const_shift(struct lp_type type) { if(type.floating) return 0; @@ -86,7 +86,7 @@ lp_const_shift(union lp_type type) unsigned -lp_const_offset(union lp_type type) +lp_const_offset(struct lp_type type) { if(type.floating || type.fixed) return 0; @@ -104,7 +104,7 @@ lp_const_offset(union lp_type type) * else for the fixed points types and normalized integers. */ double -lp_const_scale(union lp_type type) +lp_const_scale(struct lp_type type) { unsigned long long llscale; double dscale; @@ -122,7 +122,7 @@ lp_const_scale(union lp_type type) * Minimum value representable by the type. */ double -lp_const_min(union lp_type type) +lp_const_min(struct lp_type type) { unsigned bits; @@ -158,7 +158,7 @@ lp_const_min(union lp_type type) * Maximum value representable by the type. */ double -lp_const_max(union lp_type type) +lp_const_max(struct lp_type type) { unsigned bits; @@ -190,7 +190,7 @@ lp_const_max(union lp_type type) double -lp_const_eps(union lp_type type) +lp_const_eps(struct lp_type type) { if (type.floating) { switch(type.width) { @@ -211,7 +211,7 @@ lp_const_eps(union lp_type type) LLVMValueRef -lp_build_undef(union lp_type type) +lp_build_undef(struct lp_type type) { LLVMTypeRef vec_type = lp_build_vec_type(type); return LLVMGetUndef(vec_type); @@ -219,7 +219,7 @@ lp_build_undef(union lp_type type) LLVMValueRef -lp_build_zero(union lp_type type) +lp_build_zero(struct lp_type type) { LLVMTypeRef vec_type = lp_build_vec_type(type); return LLVMConstNull(vec_type); @@ -227,7 +227,7 @@ lp_build_zero(union lp_type type) LLVMValueRef -lp_build_one(union lp_type type) +lp_build_one(struct lp_type type) { LLVMTypeRef elem_type; LLVMValueRef elems[LP_MAX_VECTOR_LENGTH]; @@ -269,7 +269,7 @@ lp_build_one(union lp_type type) LLVMValueRef -lp_build_const_scalar(union lp_type type, +lp_build_const_scalar(struct lp_type type, double val) { LLVMTypeRef elem_type = lp_build_elem_type(type); @@ -295,7 +295,7 @@ lp_build_const_scalar(union lp_type type, LLVMValueRef -lp_build_int_const_scalar(union lp_type type, +lp_build_int_const_scalar(struct lp_type type, long long val) { LLVMTypeRef elem_type = lp_build_int_elem_type(type); @@ -312,7 +312,7 @@ lp_build_int_const_scalar(union lp_type type, LLVMValueRef -lp_build_const_aos(union lp_type type, +lp_build_const_aos(struct lp_type type, double r, double g, double b, double a, const unsigned char *swizzle) { @@ -352,8 +352,8 @@ lp_build_const_aos(union lp_type type, LLVMValueRef -lp_build_const_mask_aos(union lp_type type, - boolean cond[4]) +lp_build_const_mask_aos(struct lp_type type, + const boolean cond[4]) { LLVMTypeRef elem_type = LLVMIntType(type.width); LLVMValueRef masks[LP_MAX_VECTOR_LENGTH]; diff --git a/src/gallium/drivers/llvmpipe/lp_bld_const.h b/src/gallium/drivers/llvmpipe/lp_bld_const.h index 1934530ea3..ffb302f736 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_const.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_const.h @@ -42,67 +42,67 @@ #include <pipe/p_compiler.h> -union lp_type type; +struct lp_type type; unsigned -lp_mantissa(union lp_type type); +lp_mantissa(struct lp_type type); unsigned -lp_const_shift(union lp_type type); +lp_const_shift(struct lp_type type); unsigned -lp_const_offset(union lp_type type); +lp_const_offset(struct lp_type type); double -lp_const_scale(union lp_type type); +lp_const_scale(struct lp_type type); double -lp_const_min(union lp_type type); +lp_const_min(struct lp_type type); double -lp_const_max(union lp_type type); +lp_const_max(struct lp_type type); double -lp_const_eps(union lp_type type); +lp_const_eps(struct lp_type type); LLVMValueRef -lp_build_undef(union lp_type type); +lp_build_undef(struct lp_type type); LLVMValueRef -lp_build_zero(union lp_type type); +lp_build_zero(struct lp_type type); LLVMValueRef -lp_build_one(union lp_type type); +lp_build_one(struct lp_type type); LLVMValueRef -lp_build_const_scalar(union lp_type type, +lp_build_const_scalar(struct lp_type type, double val); LLVMValueRef -lp_build_int_const_scalar(union lp_type type, +lp_build_int_const_scalar(struct lp_type type, long long val); LLVMValueRef -lp_build_const_aos(union lp_type type, +lp_build_const_aos(struct lp_type type, double r, double g, double b, double a, const unsigned char *swizzle); LLVMValueRef -lp_build_const_mask_aos(union lp_type type, - boolean cond[4]); +lp_build_const_mask_aos(struct lp_type type, + const boolean cond[4]); #endif /* !LP_BLD_CONST_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.c b/src/gallium/drivers/llvmpipe/lp_bld_conv.c index c5a71d2c72..186cac70f6 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_conv.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.c @@ -86,7 +86,7 @@ */ LLVMValueRef lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, - union lp_type src_type, + struct lp_type src_type, unsigned dst_width, LLVMValueRef src) { @@ -152,7 +152,7 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, LLVMValueRef lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, unsigned src_width, - union lp_type dst_type, + struct lp_type dst_type, LLVMValueRef src) { LLVMTypeRef vec_type = lp_build_vec_type(dst_type); @@ -248,8 +248,8 @@ lp_build_const_pack_shuffle(unsigned n) */ static void lp_build_expand(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, LLVMValueRef src, LLVMValueRef *dst, unsigned num_dsts) { @@ -266,7 +266,7 @@ lp_build_expand(LLVMBuilderRef builder, dst[0] = src; while(src_type.width < dst_type.width) { - union lp_type new_type = src_type; + struct lp_type new_type = src_type; LLVMTypeRef new_vec_type; new_type.width *= 2; @@ -314,8 +314,8 @@ lp_build_expand(LLVMBuilderRef builder, */ static LLVMValueRef lp_build_pack2(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, boolean clamped, LLVMValueRef lo, LLVMValueRef hi) @@ -392,8 +392,8 @@ lp_build_pack2(LLVMBuilderRef builder, */ static LLVMValueRef lp_build_pack(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, boolean clamped, const LLVMValueRef *src, unsigned num_srcs) { @@ -410,7 +410,7 @@ lp_build_pack(LLVMBuilderRef builder, tmp[i] = src[i]; while(src_type.width > dst_type.width) { - union lp_type new_type = src_type; + struct lp_type new_type = src_type; new_type.width /= 2; new_type.length *= 2; @@ -442,12 +442,12 @@ lp_build_pack(LLVMBuilderRef builder, */ void lp_build_conv(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, LLVMValueRef *dst, unsigned num_dsts) { - union lp_type tmp_type; + struct lp_type tmp_type; LLVMValueRef tmp[LP_MAX_VECTOR_LENGTH]; unsigned num_tmps; unsigned i; @@ -470,7 +470,7 @@ lp_build_conv(LLVMBuilderRef builder, * Clamp if necessary */ - if(src_type.value != dst_type.value) { + if(memcmp(&src_type, &dst_type, sizeof src_type) != 0) { struct lp_build_context bld; double src_min = lp_const_min(src_type); double dst_min = lp_const_min(dst_type); @@ -656,8 +656,8 @@ lp_build_conv(LLVMBuilderRef builder, */ void lp_build_conv_mask(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, LLVMValueRef *dst, unsigned num_dsts) { diff --git a/src/gallium/drivers/llvmpipe/lp_bld_conv.h b/src/gallium/drivers/llvmpipe/lp_bld_conv.h index 05c1ef2a10..ca378804d2 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_conv.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_conv.h @@ -40,33 +40,33 @@ #include <llvm-c/Core.h> -union lp_type type; +struct lp_type type; LLVMValueRef lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder, - union lp_type src_type, + struct lp_type src_type, unsigned dst_width, LLVMValueRef src); LLVMValueRef lp_build_unsigned_norm_to_float(LLVMBuilderRef builder, unsigned src_width, - union lp_type dst_type, + struct lp_type dst_type, LLVMValueRef src); void lp_build_conv(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, const LLVMValueRef *srcs, unsigned num_srcs, LLVMValueRef *dsts, unsigned num_dsts); void lp_build_conv_mask(LLVMBuilderRef builder, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, const LLVMValueRef *src, unsigned num_srcs, LLVMValueRef *dst, unsigned num_dsts); diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.c b/src/gallium/drivers/llvmpipe/lp_bld_depth.c index e5fe81193f..21c665c4d4 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_depth.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.c @@ -71,11 +71,11 @@ /** * Return a type appropriate for depth/stencil testing. */ -union lp_type +struct lp_type lp_depth_type(const struct util_format_description *format_desc, unsigned length) { - union lp_type type; + struct lp_type type; unsigned swizzle; assert(format_desc->colorspace == UTIL_FORMAT_COLORSPACE_ZS); @@ -85,7 +85,7 @@ lp_depth_type(const struct util_format_description *format_desc, swizzle = format_desc->swizzle[0]; assert(swizzle < 4); - type.value = 0; + memset(&type, 0, sizeof type); type.width = format_desc->block.bits; if(format_desc->channel[swizzle].type == UTIL_FORMAT_TYPE_FLOAT) { @@ -114,7 +114,7 @@ lp_depth_type(const struct util_format_description *format_desc, void lp_build_depth_test(LLVMBuilderRef builder, const struct pipe_depth_state *state, - union lp_type type, + struct lp_type type, const struct util_format_description *format_desc, struct lp_build_mask_context *mask, LLVMValueRef src, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_depth.h b/src/gallium/drivers/llvmpipe/lp_bld_depth.h index 5d2e042fcc..79d6981bb5 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_depth.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_depth.h @@ -41,11 +41,11 @@ struct pipe_depth_state; struct util_format_description; -union lp_type; +struct lp_type; struct lp_build_mask_context; -union lp_type +struct lp_type lp_depth_type(const struct util_format_description *format_desc, unsigned length); @@ -53,7 +53,7 @@ lp_depth_type(const struct util_format_description *format_desc, void lp_build_depth_test(LLVMBuilderRef builder, const struct pipe_depth_state *state, - union lp_type type, + struct lp_type type, const struct util_format_description *format_desc, struct lp_build_mask_context *mask, LLVMValueRef src, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_flow.c b/src/gallium/drivers/llvmpipe/lp_bld_flow.c index 69ed014ff3..dcc25fbff8 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_flow.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_flow.c @@ -405,7 +405,7 @@ lp_build_mask_check(struct lp_build_mask_context *mask) void lp_build_mask_begin(struct lp_build_mask_context *mask, struct lp_build_flow_context *flow, - union lp_type type, + struct lp_type type, LLVMValueRef value) { memset(mask, 0, sizeof *mask); diff --git a/src/gallium/drivers/llvmpipe/lp_bld_flow.h b/src/gallium/drivers/llvmpipe/lp_bld_flow.h index 9d76e3064d..e61999ff06 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_flow.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_flow.h @@ -38,7 +38,7 @@ #include <llvm-c/Core.h> -union lp_type; +struct lp_type; struct lp_build_flow_context; @@ -84,7 +84,7 @@ struct lp_build_mask_context void lp_build_mask_begin(struct lp_build_mask_context *mask, struct lp_build_flow_context *flow, - union lp_type type, + struct lp_type type, LLVMValueRef value); /** diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format.h b/src/gallium/drivers/llvmpipe/lp_bld_format.h index 5ee0656093..6d3f692619 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_format.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_format.h @@ -39,7 +39,7 @@ #include "pipe/p_format.h" struct util_format_description; -union lp_type; +struct lp_type; /** @@ -103,7 +103,7 @@ lp_build_gather(LLVMBuilderRef builder, void lp_build_unpack_rgba_soa(LLVMBuilderRef builder, const struct util_format_description *format_desc, - union lp_type type, + struct lp_type type, LLVMValueRef packed, LLVMValueRef *rgba); @@ -111,7 +111,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, void lp_build_load_rgba_soa(LLVMBuilderRef builder, const struct util_format_description *format_desc, - union lp_type type, + struct lp_type type, LLVMValueRef base_ptr, LLVMValueRef offsets, LLVMValueRef *rgba); diff --git a/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c index 569e8d10a3..b5ff434e1a 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_format_soa.c @@ -84,7 +84,7 @@ lp_build_gather(LLVMBuilderRef builder, static LLVMValueRef -lp_build_format_swizzle(union lp_type type, +lp_build_format_swizzle(struct lp_type type, const LLVMValueRef *inputs, enum util_format_swizzle swizzle) { @@ -110,7 +110,7 @@ lp_build_format_swizzle(union lp_type type, void lp_build_unpack_rgba_soa(LLVMBuilderRef builder, const struct util_format_description *format_desc, - union lp_type type, + struct lp_type type, LLVMValueRef packed, LLVMValueRef *rgba) { @@ -188,7 +188,7 @@ lp_build_unpack_rgba_soa(LLVMBuilderRef builder, void lp_build_load_rgba_soa(LLVMBuilderRef builder, const struct util_format_description *format_desc, - union lp_type type, + struct lp_type type, LLVMValueRef base_ptr, LLVMValueRef offsets, LLVMValueRef *rgba) diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.c b/src/gallium/drivers/llvmpipe/lp_bld_interp.c index cfe20a0d75..338dbca6d1 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_interp.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.c @@ -292,7 +292,7 @@ void lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld, const struct tgsi_token *tokens, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, LLVMValueRef a0_ptr, LLVMValueRef dadx_ptr, LLVMValueRef dady_ptr, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_interp.h b/src/gallium/drivers/llvmpipe/lp_bld_interp.h index 9194f6233a..9c57a10879 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_interp.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_interp.h @@ -83,7 +83,7 @@ void lp_build_interp_soa_init(struct lp_build_interp_soa_context *bld, const struct tgsi_token *tokens, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, LLVMValueRef a0_ptr, LLVMValueRef dadx_ptr, LLVMValueRef dady_ptr, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.c b/src/gallium/drivers/llvmpipe/lp_bld_logic.c index 995a69c0f4..6b6f820769 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_logic.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.c @@ -45,7 +45,7 @@ lp_build_cmp(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMTypeRef vec_type = lp_build_vec_type(type); LLVMTypeRef int_vec_type = lp_build_int_vec_type(type); LLVMValueRef zeros = LLVMConstNull(int_vec_type); @@ -301,7 +301,7 @@ lp_build_select(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b) { - union lp_type type = bld->type; + struct lp_type type = bld->type; LLVMValueRef res; if(a == b) @@ -339,7 +339,7 @@ lp_build_select_aos(struct lp_build_context *bld, LLVMValueRef b, const boolean cond[4]) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; const unsigned n = type.length; unsigned i, j; diff --git a/src/gallium/drivers/llvmpipe/lp_bld_logic.h b/src/gallium/drivers/llvmpipe/lp_bld_logic.h index 9099e0fb5b..a4ee7723b5 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_logic.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_logic.h @@ -42,7 +42,7 @@ #include "pipe/p_defines.h" /* For PIPE_FUNC_xxx */ -union lp_type type; +struct lp_type type; struct lp_build_context; diff --git a/src/gallium/drivers/llvmpipe/lp_bld_sample.h b/src/gallium/drivers/llvmpipe/lp_bld_sample.h index 6f565af76d..403d0e4836 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_sample.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_sample.h @@ -40,7 +40,7 @@ struct pipe_texture; struct pipe_sampler_state; -union lp_type; +struct lp_type; /** @@ -123,7 +123,7 @@ void lp_build_sample_soa(LLVMBuilderRef builder, const struct lp_sampler_static_state *static_state, struct lp_sampler_dynamic_state *dynamic_state, - union lp_type fp_type, + struct lp_type fp_type, unsigned unit, unsigned num_coords, const LLVMValueRef *coords, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c index 3ca25b0e76..8ca1be6f1b 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_sample_soa.c @@ -97,15 +97,15 @@ struct lp_build_sample_context const struct util_format_description *format_desc; /** Incoming coordinates type and build context */ - union lp_type coord_type; + struct lp_type coord_type; struct lp_build_context coord_bld; /** Integer coordinates */ - union lp_type int_coord_type; + struct lp_type int_coord_type; struct lp_build_context int_coord_bld; /** Output texels type and build context */ - union lp_type texel_type; + struct lp_type texel_type; struct lp_build_context texel_bld; }; @@ -208,6 +208,11 @@ lp_build_sample_wrap(struct lp_build_sample_context *bld, case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE: case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER: /* FIXME */ + _debug_printf("warning: failed to translate texture wrap mode %u\n", wrap_mode); + coord = lp_build_max(int_coord_bld, coord, int_coord_bld->zero); + coord = lp_build_min(int_coord_bld, coord, length_minus_one); + break; + default: assert(0); } @@ -337,7 +342,7 @@ void lp_build_sample_soa(LLVMBuilderRef builder, const struct lp_sampler_static_state *static_state, struct lp_sampler_dynamic_state *dynamic_state, - union lp_type type, + struct lp_type type, unsigned unit, unsigned num_coords, const LLVMValueRef *coords, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c index f35638be44..64e81f7b1f 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.c @@ -64,7 +64,7 @@ LLVMValueRef lp_build_broadcast_scalar(struct lp_build_context *bld, LLVMValueRef scalar) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; LLVMValueRef res; unsigned i; @@ -83,7 +83,7 @@ lp_build_broadcast_aos(struct lp_build_context *bld, LLVMValueRef a, unsigned channel) { - const union lp_type type = bld->type; + const struct lp_type type = bld->type; const unsigned n = type.length; unsigned i, j; @@ -115,7 +115,7 @@ lp_build_broadcast_aos(struct lp_build_context *bld, * YY00 YY00 .... YY00 * YYYY YYYY .... YYYY <= output */ - union lp_type type4 = type; + struct lp_type type4 = type; const char shifts[4][2] = { { 1, 2}, {-1, 2}, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h index cb0b6707ec..1f6da80448 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_swizzle.h @@ -40,7 +40,7 @@ #include <llvm-c/Core.h> -union lp_type type; +struct lp_type type; struct lp_build_context; diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h b/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h index 10c251c416..eddb7a83fa 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi.h @@ -39,7 +39,7 @@ struct tgsi_token; -union lp_type; +struct lp_type; struct lp_build_context; struct lp_build_mask_context; @@ -60,7 +60,7 @@ struct lp_build_sampler_soa void (*emit_fetch_texel)( struct lp_build_sampler_soa *sampler, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, unsigned unit, unsigned num_coords, const LLVMValueRef *coords, @@ -72,7 +72,7 @@ struct lp_build_sampler_soa void lp_build_tgsi_soa(LLVMBuilderRef builder, const struct tgsi_token *tokens, - union lp_type type, + struct lp_type type, struct lp_build_mask_context *mask, LLVMValueRef consts_ptr, const LLVMValueRef *pos, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c index b106ce2317..adc81569ed 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_tgsi_soa.c @@ -1415,7 +1415,7 @@ emit_instruction( void lp_build_tgsi_soa(LLVMBuilderRef builder, const struct tgsi_token *tokens, - union lp_type type, + struct lp_type type, struct lp_build_mask_context *mask, LLVMValueRef consts_ptr, const LLVMValueRef *pos, diff --git a/src/gallium/drivers/llvmpipe/lp_bld_type.c b/src/gallium/drivers/llvmpipe/lp_bld_type.c index 577644b7ab..606243d6c5 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_type.c +++ b/src/gallium/drivers/llvmpipe/lp_bld_type.c @@ -33,7 +33,7 @@ LLVMTypeRef -lp_build_elem_type(union lp_type type) +lp_build_elem_type(struct lp_type type) { if (type.floating) { switch(type.width) { @@ -55,7 +55,7 @@ lp_build_elem_type(union lp_type type) LLVMTypeRef -lp_build_vec_type(union lp_type type) +lp_build_vec_type(struct lp_type type) { LLVMTypeRef elem_type = lp_build_elem_type(type); return LLVMVectorType(elem_type, type.length); @@ -69,7 +69,7 @@ lp_build_vec_type(union lp_type type) * type and check for identity. */ boolean -lp_check_elem_type(union lp_type type, LLVMTypeRef elem_type) +lp_check_elem_type(struct lp_type type, LLVMTypeRef elem_type) { LLVMTypeKind elem_kind; @@ -107,7 +107,7 @@ lp_check_elem_type(union lp_type type, LLVMTypeRef elem_type) boolean -lp_check_vec_type(union lp_type type, LLVMTypeRef vec_type) +lp_check_vec_type(struct lp_type type, LLVMTypeRef vec_type) { LLVMTypeRef elem_type; @@ -128,7 +128,7 @@ lp_check_vec_type(union lp_type type, LLVMTypeRef vec_type) boolean -lp_check_value(union lp_type type, LLVMValueRef val) +lp_check_value(struct lp_type type, LLVMValueRef val) { LLVMTypeRef vec_type; @@ -143,25 +143,26 @@ lp_check_value(union lp_type type, LLVMValueRef val) LLVMTypeRef -lp_build_int_elem_type(union lp_type type) +lp_build_int_elem_type(struct lp_type type) { return LLVMIntType(type.width); } LLVMTypeRef -lp_build_int_vec_type(union lp_type type) +lp_build_int_vec_type(struct lp_type type) { LLVMTypeRef elem_type = lp_build_int_elem_type(type); return LLVMVectorType(elem_type, type.length); } -union lp_type -lp_int_type(union lp_type type) +struct lp_type +lp_int_type(struct lp_type type) { - union lp_type int_type; - int_type.value = 0; + struct lp_type int_type; + + memset(&int_type, 0, sizeof int_type); int_type.width = type.width; int_type.length = type.length; return int_type; @@ -171,7 +172,7 @@ lp_int_type(union lp_type type) void lp_build_context_init(struct lp_build_context *bld, LLVMBuilderRef builder, - union lp_type type) + struct lp_type type) { bld->builder = builder; bld->type = type; diff --git a/src/gallium/drivers/llvmpipe/lp_bld_type.h b/src/gallium/drivers/llvmpipe/lp_bld_type.h index 9933e0b45c..ee5ca3483c 100644 --- a/src/gallium/drivers/llvmpipe/lp_bld_type.h +++ b/src/gallium/drivers/llvmpipe/lp_bld_type.h @@ -56,58 +56,55 @@ * on the types used for intermediate computations, such as signed vs unsigned, * normalized values, or fixed point. */ -union lp_type { - struct { - /** - * Floating-point. Cannot be used with fixed. Integer numbers are - * represented by this zero. - */ - unsigned floating:1; +struct lp_type { + /** + * Floating-point. Cannot be used with fixed. Integer numbers are + * represented by this zero. + */ + unsigned floating:1; - /** - * Fixed-point. Cannot be used with floating. Integer numbers are - * represented by this zero. - */ - unsigned fixed:1; - - /** - * Whether it can represent negative values or not. - * - * If this is not set for floating point, it means that all values are - * assumed to be positive. - */ - unsigned sign:1; + /** + * Fixed-point. Cannot be used with floating. Integer numbers are + * represented by this zero. + */ + unsigned fixed:1; - /** - * Whether values are normalized to fit [0, 1] interval, or [-1, 1] - * interval for signed types. - * - * For integer types it means the representable integer range should be - * interpreted as the interval above. - * - * For floating and fixed point formats it means the values should be - * clamped to the interval above. - */ - unsigned norm:1; + /** + * Whether it can represent negative values or not. + * + * If this is not set for floating point, it means that all values are + * assumed to be positive. + */ + unsigned sign:1; - /** - * Element width. - * - * For fixed point values, the fixed point is assumed to be at half the - * width. - */ - unsigned width:14; + /** + * Whether values are normalized to fit [0, 1] interval, or [-1, 1] + * interval for signed types. + * + * For integer types it means the representable integer range should be + * interpreted as the interval above. + * + * For floating and fixed point formats it means the values should be + * clamped to the interval above. + */ + unsigned norm:1; + + /** + * Element width. + * + * For fixed point values, the fixed point is assumed to be at half the + * width. + */ + unsigned width:14; - /** - * Vector length. - * - * width*length should be a power of two greater or equal to eight. - * - * @sa LP_MAX_VECTOR_LENGTH - */ - unsigned length:14; - }; - uint32_t value; + /** + * Vector length. + * + * width*length should be a power of two greater or equal to eight. + * + * @sa LP_MAX_VECTOR_LENGTH + */ + unsigned length:14; }; @@ -124,7 +121,7 @@ struct lp_build_context * This not only describes the input/output LLVM types, but also whether * to normalize/clamp the results. */ - union lp_type type; + struct lp_type type; /** Same as lp_build_undef(type) */ LLVMValueRef undef; @@ -138,41 +135,41 @@ struct lp_build_context LLVMTypeRef -lp_build_elem_type(union lp_type type); +lp_build_elem_type(struct lp_type type); LLVMTypeRef -lp_build_vec_type(union lp_type type); +lp_build_vec_type(struct lp_type type); boolean -lp_check_elem_type(union lp_type type, LLVMTypeRef elem_type); +lp_check_elem_type(struct lp_type type, LLVMTypeRef elem_type); boolean -lp_check_vec_type(union lp_type type, LLVMTypeRef vec_type); +lp_check_vec_type(struct lp_type type, LLVMTypeRef vec_type); boolean -lp_check_value(union lp_type type, LLVMValueRef val); +lp_check_value(struct lp_type type, LLVMValueRef val); LLVMTypeRef -lp_build_int_elem_type(union lp_type type); +lp_build_int_elem_type(struct lp_type type); LLVMTypeRef -lp_build_int_vec_type(union lp_type type); +lp_build_int_vec_type(struct lp_type type); -union lp_type -lp_int_type(union lp_type type); +struct lp_type +lp_int_type(struct lp_type type); void lp_build_context_init(struct lp_build_context *bld, LLVMBuilderRef builder, - union lp_type type); + struct lp_type type); #endif /* !LP_BLD_TYPE_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c index 9465f763d5..b4a22ff4a9 100644 --- a/src/gallium/drivers/llvmpipe/lp_jit.c +++ b/src/gallium/drivers/llvmpipe/lp_jit.c @@ -152,7 +152,7 @@ lp_jit_screen_init(struct llvmpipe_screen *screen) screen->provider = LLVMCreateModuleProviderForExistingModule(screen->module); if (LLVMCreateJITCompiler(&screen->engine, screen->provider, 1, &error)) { - fprintf(stderr, "%s\n", error); + _debug_printf("%s\n", error); LLVMDisposeMessage(error); abort(); } diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c index 0ce1a37bd4..ff7ef8658a 100644 --- a/src/gallium/drivers/llvmpipe/lp_screen.c +++ b/src/gallium/drivers/llvmpipe/lp_screen.c @@ -65,8 +65,6 @@ llvmpipe_get_param(struct pipe_screen *screen, int param) return 1; case PIPE_CAP_GLSL: return 1; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 0; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c index 618cf1ffb8..9faed5a0b1 100644 --- a/src/gallium/drivers/llvmpipe/lp_state_fs.c +++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c @@ -133,13 +133,13 @@ generate_pos0(LLVMBuilderRef builder, static void generate_depth(LLVMBuilderRef builder, const struct lp_fragment_shader_variant_key *key, - union lp_type src_type, + struct lp_type src_type, struct lp_build_mask_context *mask, LLVMValueRef src, LLVMValueRef dst_ptr) { const struct util_format_description *format_desc; - union lp_type dst_type; + struct lp_type dst_type; if(!key->depth.enabled) return; @@ -181,7 +181,7 @@ generate_fs(struct llvmpipe_context *lp, struct lp_fragment_shader *shader, const struct lp_fragment_shader_variant_key *key, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, LLVMValueRef context_ptr, unsigned i, const struct lp_build_interp_soa_context *interp, @@ -299,7 +299,7 @@ generate_fs(struct llvmpipe_context *lp, static void generate_blend(const struct pipe_blend_state *blend, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, LLVMValueRef context_ptr, LLVMValueRef mask, LLVMValueRef *src, @@ -364,8 +364,8 @@ generate_fragment(struct llvmpipe_context *lp, { struct llvmpipe_screen *screen = llvmpipe_screen(lp->pipe.screen); struct lp_fragment_shader_variant *variant; - union lp_type fs_type; - union lp_type blend_type; + struct lp_type fs_type; + struct lp_type blend_type; LLVMTypeRef fs_elem_type; LLVMTypeRef fs_vec_type; LLVMTypeRef fs_int_vec_type; @@ -431,7 +431,7 @@ generate_fragment(struct llvmpipe_context *lp, /* TODO: actually pick these based on the fs and color buffer * characteristics. */ - fs_type.value = 0; + memset(&fs_type, 0, sizeof fs_type); fs_type.floating = TRUE; /* floating point values */ fs_type.sign = TRUE; /* values are signed */ fs_type.norm = FALSE; /* values are not limited to [0,1] or [-1,1] */ @@ -439,7 +439,7 @@ generate_fragment(struct llvmpipe_context *lp, fs_type.length = 4; /* 4 element per vector */ num_fs = 4; - blend_type.value = 0; + memset(&blend_type, 0, sizeof blend_type); blend_type.floating = FALSE; /* values are integers */ blend_type.sign = FALSE; /* values are unsigned */ blend_type.norm = TRUE; /* values are in [0,1] or [-1,1] */ diff --git a/src/gallium/drivers/llvmpipe/lp_test.h b/src/gallium/drivers/llvmpipe/lp_test.h index 69aaae26e0..a88e110c66 100644 --- a/src/gallium/drivers/llvmpipe/lp_test.h +++ b/src/gallium/drivers/llvmpipe/lp_test.h @@ -86,43 +86,43 @@ random_float(void); void -dump_type(FILE *fp, union lp_type type); +dump_type(FILE *fp, struct lp_type type); double -read_elem(union lp_type type, const void *src, unsigned index); +read_elem(struct lp_type type, const void *src, unsigned index); void -write_elem(union lp_type type, void *dst, unsigned index, double src); +write_elem(struct lp_type type, void *dst, unsigned index, double src); void -random_elem(union lp_type type, void *dst, unsigned index); +random_elem(struct lp_type type, void *dst, unsigned index); void -read_vec(union lp_type type, const void *src, double *dst); +read_vec(struct lp_type type, const void *src, double *dst); void -write_vec(union lp_type type, void *dst, const double *src); +write_vec(struct lp_type type, void *dst, const double *src); void -random_vec(union lp_type type, void *dst); +random_vec(struct lp_type type, void *dst); boolean -compare_vec_with_eps(union lp_type type, const void *res, const void *ref, double eps); +compare_vec_with_eps(struct lp_type type, const void *res, const void *ref, double eps); boolean -compare_vec(union lp_type type, const void *res, const void *ref); +compare_vec(struct lp_type type, const void *res, const void *ref); void -dump_vec(FILE *fp, union lp_type type, const void *src); +dump_vec(FILE *fp, struct lp_type type, const void *src); #endif /* !LP_TEST_H */ diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c index 8dfad468e3..94b661dcba 100644 --- a/src/gallium/drivers/llvmpipe/lp_test_blend.c +++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c @@ -80,7 +80,7 @@ static void write_tsv_row(FILE *fp, const struct pipe_blend_state *blend, enum vector_mode mode, - union lp_type type, + struct lp_type type, double cycles, boolean success) { @@ -125,7 +125,7 @@ static void dump_blend_type(FILE *fp, const struct pipe_blend_state *blend, enum vector_mode mode, - union lp_type type) + struct lp_type type) { fprintf(fp, "%s", mode ? "soa" : "aos"); @@ -153,7 +153,7 @@ static LLVMValueRef add_blend_test(LLVMModuleRef module, const struct pipe_blend_state *blend, enum vector_mode mode, - union lp_type type) + struct lp_type type) { LLVMTypeRef ret_type; LLVMTypeRef vec_type; @@ -467,7 +467,7 @@ test_one(unsigned verbose, FILE *fp, const struct pipe_blend_state *blend, enum vector_mode mode, - union lp_type type) + struct lp_type type) { LLVMModuleRef module = NULL; LLVMValueRef func = NULL; @@ -765,10 +765,10 @@ blend_funcs[] = { }; -const union lp_type blend_types[] = { +const struct lp_type blend_types[] = { /* float, fixed, sign, norm, width, len */ - {{ TRUE, FALSE, FALSE, TRUE, 32, 4 }}, /* f32 x 4 */ - {{ FALSE, FALSE, FALSE, TRUE, 8, 16 }}, /* u8n x 16 */ + { TRUE, FALSE, FALSE, TRUE, 32, 4 }, /* f32 x 4 */ + { FALSE, FALSE, FALSE, TRUE, 8, 16 }, /* u8n x 16 */ }; @@ -788,7 +788,7 @@ test_all(unsigned verbose, FILE *fp) const unsigned *alpha_dst_factor; struct pipe_blend_state blend; enum vector_mode mode; - const union lp_type *type; + const struct lp_type *type; bool success = TRUE; for(rgb_func = blend_funcs; rgb_func < &blend_funcs[num_funcs]; ++rgb_func) { @@ -841,27 +841,27 @@ test_some(unsigned verbose, FILE *fp, unsigned long n) const unsigned *alpha_dst_factor; struct pipe_blend_state blend; enum vector_mode mode; - const union lp_type *type; + const struct lp_type *type; unsigned long i; bool success = TRUE; for(i = 0; i < n; ++i) { - rgb_func = &blend_funcs[random() % num_funcs]; - alpha_func = &blend_funcs[random() % num_funcs]; - rgb_src_factor = &blend_factors[random() % num_factors]; - alpha_src_factor = &blend_factors[random() % num_factors]; + rgb_func = &blend_funcs[rand() % num_funcs]; + alpha_func = &blend_funcs[rand() % num_funcs]; + rgb_src_factor = &blend_factors[rand() % num_factors]; + alpha_src_factor = &blend_factors[rand() % num_factors]; do { - rgb_dst_factor = &blend_factors[random() % num_factors]; + rgb_dst_factor = &blend_factors[rand() % num_factors]; } while(*rgb_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE); do { - alpha_dst_factor = &blend_factors[random() % num_factors]; + alpha_dst_factor = &blend_factors[rand() % num_factors]; } while(*alpha_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE); - mode = random() & 1; + mode = rand() & 1; - type = &blend_types[random() % num_types]; + type = &blend_types[rand() % num_types]; memset(&blend, 0, sizeof blend); blend.blend_enable = 1; diff --git a/src/gallium/drivers/llvmpipe/lp_test_conv.c b/src/gallium/drivers/llvmpipe/lp_test_conv.c index e6489834af..9dcf58e5dc 100644 --- a/src/gallium/drivers/llvmpipe/lp_test_conv.c +++ b/src/gallium/drivers/llvmpipe/lp_test_conv.c @@ -59,8 +59,8 @@ write_tsv_header(FILE *fp) static void write_tsv_row(FILE *fp, - union lp_type src_type, - union lp_type dst_type, + struct lp_type src_type, + struct lp_type dst_type, double cycles, boolean success) { @@ -80,8 +80,8 @@ write_tsv_row(FILE *fp, static void dump_conv_types(FILE *fp, - union lp_type src_type, - union lp_type dst_type) + struct lp_type src_type, + struct lp_type dst_type) { fprintf(fp, "src_type="); dump_type(fp, src_type); @@ -96,8 +96,8 @@ dump_conv_types(FILE *fp, static LLVMValueRef add_conv_test(LLVMModuleRef module, - union lp_type src_type, unsigned num_srcs, - union lp_type dst_type, unsigned num_dsts) + struct lp_type src_type, unsigned num_srcs, + struct lp_type dst_type, unsigned num_dsts) { LLVMTypeRef args[2]; LLVMValueRef func; @@ -145,8 +145,8 @@ add_conv_test(LLVMModuleRef module, static boolean test_one(unsigned verbose, FILE *fp, - union lp_type src_type, - union lp_type dst_type) + struct lp_type src_type, + struct lp_type dst_type) { LLVMModuleRef module = NULL; LLVMValueRef func = NULL; @@ -343,35 +343,35 @@ test_one(unsigned verbose, } -const union lp_type conv_types[] = { +const struct lp_type conv_types[] = { /* float, fixed, sign, norm, width, len */ - {{ TRUE, FALSE, TRUE, TRUE, 32, 4 }}, - {{ TRUE, FALSE, TRUE, FALSE, 32, 4 }}, - {{ TRUE, FALSE, FALSE, TRUE, 32, 4 }}, - {{ TRUE, FALSE, FALSE, FALSE, 32, 4 }}, + { TRUE, FALSE, TRUE, TRUE, 32, 4 }, + { TRUE, FALSE, TRUE, FALSE, 32, 4 }, + { TRUE, FALSE, FALSE, TRUE, 32, 4 }, + { TRUE, FALSE, FALSE, FALSE, 32, 4 }, /* TODO: test fixed formats too */ - {{ FALSE, FALSE, TRUE, TRUE, 16, 8 }}, - {{ FALSE, FALSE, TRUE, FALSE, 16, 8 }}, - {{ FALSE, FALSE, FALSE, TRUE, 16, 8 }}, - {{ FALSE, FALSE, FALSE, FALSE, 16, 8 }}, + { FALSE, FALSE, TRUE, TRUE, 16, 8 }, + { FALSE, FALSE, TRUE, FALSE, 16, 8 }, + { FALSE, FALSE, FALSE, TRUE, 16, 8 }, + { FALSE, FALSE, FALSE, FALSE, 16, 8 }, - {{ FALSE, FALSE, TRUE, TRUE, 32, 4 }}, - {{ FALSE, FALSE, TRUE, FALSE, 32, 4 }}, - {{ FALSE, FALSE, FALSE, TRUE, 32, 4 }}, - {{ FALSE, FALSE, FALSE, FALSE, 32, 4 }}, + { FALSE, FALSE, TRUE, TRUE, 32, 4 }, + { FALSE, FALSE, TRUE, FALSE, 32, 4 }, + { FALSE, FALSE, FALSE, TRUE, 32, 4 }, + { FALSE, FALSE, FALSE, FALSE, 32, 4 }, - {{ FALSE, FALSE, TRUE, TRUE, 16, 8 }}, - {{ FALSE, FALSE, TRUE, FALSE, 16, 8 }}, - {{ FALSE, FALSE, FALSE, TRUE, 16, 8 }}, - {{ FALSE, FALSE, FALSE, FALSE, 16, 8 }}, + { FALSE, FALSE, TRUE, TRUE, 16, 8 }, + { FALSE, FALSE, TRUE, FALSE, 16, 8 }, + { FALSE, FALSE, FALSE, TRUE, 16, 8 }, + { FALSE, FALSE, FALSE, FALSE, 16, 8 }, - {{ FALSE, FALSE, TRUE, TRUE, 8, 16 }}, - {{ FALSE, FALSE, TRUE, FALSE, 8, 16 }}, - {{ FALSE, FALSE, FALSE, TRUE, 8, 16 }}, - {{ FALSE, FALSE, FALSE, FALSE, 8, 16 }}, + { FALSE, FALSE, TRUE, TRUE, 8, 16 }, + { FALSE, FALSE, TRUE, FALSE, 8, 16 }, + { FALSE, FALSE, FALSE, TRUE, 8, 16 }, + { FALSE, FALSE, FALSE, FALSE, 8, 16 }, }; @@ -381,8 +381,8 @@ const unsigned num_types = sizeof(conv_types)/sizeof(conv_types[0]); boolean test_all(unsigned verbose, FILE *fp) { - const union lp_type *src_type; - const union lp_type *dst_type; + const struct lp_type *src_type; + const struct lp_type *dst_type; bool success = TRUE; for(src_type = conv_types; src_type < &conv_types[num_types]; ++src_type) { @@ -407,16 +407,16 @@ test_all(unsigned verbose, FILE *fp) boolean test_some(unsigned verbose, FILE *fp, unsigned long n) { - const union lp_type *src_type; - const union lp_type *dst_type; + const struct lp_type *src_type; + const struct lp_type *dst_type; unsigned long i; bool success = TRUE; for(i = 0; i < n; ++i) { - src_type = &conv_types[random() % num_types]; + src_type = &conv_types[rand() % num_types]; do { - dst_type = &conv_types[random() % num_types]; + dst_type = &conv_types[rand() % num_types]; } while (src_type == dst_type || src_type->norm != dst_type->norm); if(!test_one(verbose, fp, *src_type, *dst_type)) diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c index 49213fb4f0..4592dc0b2d 100644 --- a/src/gallium/drivers/llvmpipe/lp_test_main.c +++ b/src/gallium/drivers/llvmpipe/lp_test_main.c @@ -40,7 +40,7 @@ void dump_type(FILE *fp, - union lp_type type) + struct lp_type type) { fprintf(fp, "%s%s%u%sx%u", type.sign ? (type.floating || type.fixed ? "" : "s") : "u", @@ -52,7 +52,7 @@ dump_type(FILE *fp, double -read_elem(union lp_type type, const void *src, unsigned index) +read_elem(struct lp_type type, const void *src, unsigned index) { double scale = lp_const_scale(type); double value; @@ -115,7 +115,7 @@ read_elem(union lp_type type, const void *src, unsigned index) void -write_elem(union lp_type type, void *dst, unsigned index, double value) +write_elem(struct lp_type type, void *dst, unsigned index, double value) { assert(index < type.length); if(!type.sign && value < 0.0) @@ -184,11 +184,11 @@ write_elem(union lp_type type, void *dst, unsigned index, double value) void -random_elem(union lp_type type, void *dst, unsigned index) +random_elem(struct lp_type type, void *dst, unsigned index) { double value; assert(index < type.length); - value = (double)random()/(double)RAND_MAX; + value = (double)rand()/(double)RAND_MAX; if(!type.norm) { unsigned long long mask; if (type.floating) @@ -199,17 +199,17 @@ random_elem(union lp_type type, void *dst, unsigned index) mask = ((unsigned long long)1 << (type.width - 1)) - 1; else mask = ((unsigned long long)1 << type.width) - 1; - value += (double)(mask & random()); + value += (double)(mask & rand()); } if(!type.sign) - if(random() & 1) + if(rand() & 1) value = -value; write_elem(type, dst, index, value); } void -read_vec(union lp_type type, const void *src, double *dst) +read_vec(struct lp_type type, const void *src, double *dst) { unsigned i; for (i = 0; i < type.length; ++i) @@ -218,7 +218,7 @@ read_vec(union lp_type type, const void *src, double *dst) void -write_vec(union lp_type type, void *dst, const double *src) +write_vec(struct lp_type type, void *dst, const double *src) { unsigned i; for (i = 0; i < type.length; ++i) @@ -229,12 +229,12 @@ write_vec(union lp_type type, void *dst, const double *src) float random_float(void) { - return (float)((double)random()/(double)RAND_MAX); + return (float)((double)rand()/(double)RAND_MAX); } void -random_vec(union lp_type type, void *dst) +random_vec(struct lp_type type, void *dst) { unsigned i; for (i = 0; i < type.length; ++i) @@ -243,7 +243,7 @@ random_vec(union lp_type type, void *dst) boolean -compare_vec_with_eps(union lp_type type, const void *res, const void *ref, double eps) +compare_vec_with_eps(struct lp_type type, const void *res, const void *ref, double eps) { unsigned i; for (i = 0; i < type.length; ++i) { @@ -259,7 +259,7 @@ compare_vec_with_eps(union lp_type type, const void *res, const void *ref, doubl boolean -compare_vec(union lp_type type, const void *res, const void *ref) +compare_vec(struct lp_type type, const void *res, const void *ref) { double eps = lp_const_eps(type); return compare_vec_with_eps(type, res, ref, eps); @@ -267,7 +267,7 @@ compare_vec(union lp_type type, const void *res, const void *ref) void -dump_vec(FILE *fp, union lp_type type, const void *src) +dump_vec(FILE *fp, struct lp_type type, const void *src) { unsigned i; for (i = 0; i < type.length; ++i) { diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_c.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_c.c index 9a876f404d..a1365a045f 100644 --- a/src/gallium/drivers/llvmpipe/lp_tex_sample_c.c +++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_c.c @@ -1654,7 +1654,7 @@ lp_c_sampler_soa_destroy(struct lp_build_sampler_soa *sampler) static void lp_c_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa *_sampler, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, unsigned unit, unsigned num_coords, const LLVMValueRef *coords, diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c index 7d31705d01..d2a6ae21f5 100644 --- a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c +++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c @@ -149,7 +149,7 @@ lp_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler) static void lp_llvm_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa *base, LLVMBuilderRef builder, - union lp_type type, + struct lp_type type, unsigned unit, unsigned num_coords, const LLVMValueRef *coords, diff --git a/src/gallium/drivers/nv04/nv04_screen.c b/src/gallium/drivers/nv04/nv04_screen.c index ff2febb668..170ce3eb7e 100644 --- a/src/gallium/drivers/nv04/nv04_screen.c +++ b/src/gallium/drivers/nv04/nv04_screen.c @@ -16,8 +16,6 @@ nv04_screen_get_param(struct pipe_screen *screen, int param) return 0; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 0; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/nv10/nv10_screen.c b/src/gallium/drivers/nv10/nv10_screen.c index 4469b22d91..ee5901e743 100644 --- a/src/gallium/drivers/nv10/nv10_screen.c +++ b/src/gallium/drivers/nv10/nv10_screen.c @@ -15,8 +15,6 @@ nv10_screen_get_param(struct pipe_screen *screen, int param) return 0; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 1; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/nv20/nv20_screen.c b/src/gallium/drivers/nv20/nv20_screen.c index e6924ad71e..4eeacd1afd 100644 --- a/src/gallium/drivers/nv20/nv20_screen.c +++ b/src/gallium/drivers/nv20/nv20_screen.c @@ -15,8 +15,6 @@ nv20_screen_get_param(struct pipe_screen *screen, int param) return 0; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 1; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c index f8285e4455..41af38450b 100644 --- a/src/gallium/drivers/nv30/nv30_screen.c +++ b/src/gallium/drivers/nv30/nv30_screen.c @@ -22,8 +22,6 @@ nv30_screen_get_param(struct pipe_screen *pscreen, int param) return 1; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 1; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c index 5d2a4216c5..bd13dfddd1 100644 --- a/src/gallium/drivers/nv40/nv40_screen.c +++ b/src/gallium/drivers/nv40/nv40_screen.c @@ -21,8 +21,6 @@ nv40_screen_get_param(struct pipe_screen *pscreen, int param) return 1; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 1; case PIPE_CAP_ANISOTROPIC_FILTER: return 1; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c index 6e8f4f9750..fca078b174 100644 --- a/src/gallium/drivers/nv50/nv50_context.c +++ b/src/gallium/drivers/nv50/nv50_context.c @@ -37,11 +37,12 @@ nv50_flush(struct pipe_context *pipe, unsigned flags, /* We need this in the ddx for reliable composite, not sure what we're * actually flushing. We generate all our own flushes with flags = 0. */ - WAIT_RING(chan, 3); + WAIT_RING(chan, 2); BEGIN_RING(chan, eng2d, 0x0110, 1); OUT_RING (chan, 0); - FIRE_RING(chan); + if (flags & PIPE_FLUSH_FRAME) + FIRE_RING(chan); } static void @@ -110,6 +111,9 @@ nv50_create(struct pipe_screen *pscreen, unsigned pctx_id) nv50->pipe.is_texture_referenced = nv50_is_texture_referenced; nv50->pipe.is_buffer_referenced = nv50_is_buffer_referenced; + screen->base.channel->user_private = nv50; + screen->base.channel->flush_notify = nv50_state_flush_notify; + nv50_init_surface_functions(nv50); nv50_init_state_functions(nv50); nv50_init_query_functions(nv50); diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h index 1e9e8e49bf..4608854d71 100644 --- a/src/gallium/drivers/nv50/nv50_context.h +++ b/src/gallium/drivers/nv50/nv50_context.h @@ -116,6 +116,7 @@ struct nv50_state { unsigned miptree_nr; struct nouveau_stateobj *vertprog; struct nouveau_stateobj *fragprog; + struct nouveau_stateobj *programs; struct nouveau_stateobj *vtxfmt; struct nouveau_stateobj *vtxbuf; struct nouveau_stateobj *vtxattr; @@ -190,10 +191,12 @@ extern void nv50_clear(struct pipe_context *pipe, unsigned buffers, /* nv50_program.c */ extern void nv50_vertprog_validate(struct nv50_context *nv50); extern void nv50_fragprog_validate(struct nv50_context *nv50); +extern void nv50_linkage_validate(struct nv50_context *nv50); extern void nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p); /* nv50_state_validate.c */ extern boolean nv50_state_validate(struct nv50_context *nv50); +extern void nv50_state_flush_notify(struct nouveau_channel *chan); /* nv50_tex.c */ extern void nv50_tex_validate(struct nv50_context *); diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c index 4a838529de..eb90d5e66f 100644 --- a/src/gallium/drivers/nv50/nv50_program.c +++ b/src/gallium/drivers/nv50/nv50_program.c @@ -112,6 +112,10 @@ struct nv50_pc { struct nv50_reg *temp_temp[16]; unsigned temp_temp_nr; + /* broadcast and destination replacement regs */ + struct nv50_reg *r_brdc; + struct nv50_reg *r_dst[4]; + unsigned interp_mode[32]; /* perspective interpolation registers */ struct nv50_reg *iv_p; @@ -124,6 +128,25 @@ struct nv50_pc { boolean allow32; }; +static INLINE void +ctor_reg(struct nv50_reg *reg, unsigned type, int index, int hw) +{ + reg->type = type; + reg->index = index; + reg->hw = hw; + reg->neg = 0; + reg->rhw = -1; + reg->acc = 0; +} + +static INLINE unsigned +popcnt4(uint32_t val) +{ + static const unsigned cnt[16] + = { 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4 }; + return cnt[val & 0xf]; +} + static void alloc_reg(struct nv50_pc *pc, struct nv50_reg *reg) { @@ -184,11 +207,8 @@ alloc_temp(struct nv50_pc *pc, struct nv50_reg *dst) for (i = 0; i < NV50_SU_MAX_TEMP; i++) { if (!pc->r_temp[i]) { - r = CALLOC_STRUCT(nv50_reg); - r->type = P_TEMP; - r->index = -1; - r->hw = i; - r->rhw = -1; + r = MALLOC_STRUCT(nv50_reg); + ctor_reg(r, P_TEMP, -1, i); pc->r_temp[i] = r; return r; } @@ -254,10 +274,8 @@ alloc_temp4(struct nv50_pc *pc, struct nv50_reg *dst[4], int idx) return alloc_temp4(pc, dst, idx + 4); for (i = 0; i < 4; i++) { - dst[i] = CALLOC_STRUCT(nv50_reg); - dst[i]->type = P_TEMP; - dst[i]->index = -1; - dst[i]->hw = idx + i; + dst[i] = MALLOC_STRUCT(nv50_reg); + ctor_reg(dst[i], P_TEMP, -1, idx + i); pc->r_temp[idx + i] = dst[i]; } @@ -309,7 +327,7 @@ ctor_immd(struct nv50_pc *pc, float x, float y, float z, float w) static struct nv50_reg * alloc_immd(struct nv50_pc *pc, float f) { - struct nv50_reg *r = CALLOC_STRUCT(nv50_reg); + struct nv50_reg *r = MALLOC_STRUCT(nv50_reg); unsigned hw; for (hw = 0; hw < pc->immd_nr * 4; hw++) @@ -319,9 +337,7 @@ alloc_immd(struct nv50_pc *pc, float f) if (hw == pc->immd_nr * 4) hw = ctor_immd(pc, f, -f, 0.5 * f, 0) * 4; - r->type = P_IMMD; - r->hw = hw; - r->index = -1; + ctor_reg(r, P_IMMD, -1, hw); return r; } @@ -786,6 +802,9 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) #define CVTOP_SAT 0x08 #define CVTOP_ABS 0x10 +/* 0x04 == 32 bit */ +/* 0x40 == dst is float */ +/* 0x80 == src is float */ #define CVT_F32_F32 0xc4 #define CVT_F32_S32 0x44 #define CVT_F32_U32 0x64 @@ -795,7 +814,7 @@ emit_precossin(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) static void emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src, - int wp, unsigned cop, unsigned fmt) + int wp, unsigned cvn, unsigned fmt) { struct nv50_program_exec *e; @@ -804,7 +823,7 @@ emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src, e->inst[0] |= 0xa0000000; e->inst[1] |= 0x00004000; - e->inst[1] |= (cop << 16); + e->inst[1] |= (cvn << 16); e->inst[1] |= (fmt << 24); set_src_0(pc, src, e); @@ -821,49 +840,80 @@ emit_cvt(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src, emit(pc, e); } +/* nv50 Condition codes: + * 0x1 = LT + * 0x2 = EQ + * 0x3 = LE + * 0x4 = GT + * 0x5 = NE + * 0x6 = GE + * 0x7 = set condition code ? (used before bra.lt/le/gt/ge) + * 0x8 = unordered bit (allows NaN) + */ static void -emit_set(struct nv50_pc *pc, unsigned c_op, struct nv50_reg *dst, +emit_set(struct nv50_pc *pc, unsigned ccode, struct nv50_reg *dst, int wp, struct nv50_reg *src0, struct nv50_reg *src1) { + static const unsigned cc_swapped[8] = { 0, 4, 2, 6, 1, 5, 3, 7 }; + struct nv50_program_exec *e = exec(pc); - unsigned inv_cop[8] = { 0, 4, 2, 6, 1, 5, 3, 7 }; struct nv50_reg *rdst; - assert(c_op <= 7); + assert(ccode < 16); if (check_swap_src_0_1(pc, &src0, &src1)) - c_op = inv_cop[c_op]; + ccode = cc_swapped[ccode & 7] | (ccode & 8); rdst = dst; - if (dst->type != P_TEMP) + if (dst && dst->type != P_TEMP) dst = alloc_temp(pc, NULL); /* set.u32 */ set_long(pc, e); e->inst[0] |= 0xb0000000; - e->inst[1] |= (3 << 29); - e->inst[1] |= (c_op << 14); - /*XXX: breaks things, .u32 by default? - * decuda will disasm as .u16 and use .lo/.hi regs, but this - * doesn't seem to match what the hw actually does. - inst[1] |= 0x04000000; << breaks things.. .u32 by default? + e->inst[1] |= 0x60000000 | (ccode << 14); + + /* XXX: decuda will disasm as .u16 and use .lo/.hi regs, but + * that doesn't seem to match what the hw actually does + e->inst[1] |= 0x04000000; << breaks things, u32 by default ? */ - set_dst(pc, dst, e); + + if (wp >= 0) + set_pred_wr(pc, 1, wp, e); + if (dst) + set_dst(pc, dst, e); + else { + e->inst[0] |= 0x000001fc; + e->inst[1] |= 0x00000008; + } + set_src_0(pc, src0, e); set_src_1(pc, src1, e); - emit(pc, e); - /* cvt.f32.u32 */ - e = exec(pc); - e->inst[0] = 0xa0000001; - e->inst[1] = 0x64014780; - set_dst(pc, rdst, e); - set_src_0(pc, dst, e); emit(pc, e); - if (dst != rdst) + /* cvt.f32.u32/s32 (?) if we didn't only write the predicate */ + if (rdst) + emit_cvt(pc, rdst, dst, -1, CVTOP_ABS | CVTOP_RN, CVT_F32_S32); + if (rdst && rdst != dst) free_temp(pc, dst); } +static INLINE unsigned +map_tgsi_setop_cc(unsigned op) +{ + switch (op) { + case TGSI_OPCODE_SLT: return 0x1; + case TGSI_OPCODE_SGE: return 0x6; + case TGSI_OPCODE_SEQ: return 0x2; + case TGSI_OPCODE_SGT: return 0x4; + case TGSI_OPCODE_SLE: return 0x3; + case TGSI_OPCODE_SNE: return 0xd; + default: + assert(0); + return 0; + } +} + static INLINE void emit_flr(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) { @@ -890,6 +940,12 @@ emit_abs(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) emit_cvt(pc, dst, src, -1, CVTOP_ABS, CVT_F32_F32); } +static INLINE void +emit_sat(struct nv50_pc *pc, struct nv50_reg *dst, struct nv50_reg *src) +{ + emit_cvt(pc, dst, src, -1, CVTOP_SAT, CVT_F32_F32); +} + static void emit_lit(struct nv50_pc *pc, struct nv50_reg **dst, unsigned mask, struct nv50_reg **src) @@ -1159,6 +1215,70 @@ negate_supported(const struct tgsi_full_instruction *insn, int i) } } +/* Return a read mask for source registers deduced from opcode & write mask. */ +static unsigned +nv50_tgsi_src_mask(const struct tgsi_full_instruction *insn, int c) +{ + unsigned x, mask = insn->FullDstRegisters[0].DstRegister.WriteMask; + + switch (insn->Instruction.Opcode) { + case TGSI_OPCODE_COS: + case TGSI_OPCODE_SIN: + return (mask & 0x8) | ((mask & 0x7) ? 0x1 : 0x0); + case TGSI_OPCODE_DP3: + return 0x7; + case TGSI_OPCODE_DP4: + case TGSI_OPCODE_DPH: + case TGSI_OPCODE_KIL: /* WriteMask ignored */ + return 0xf; + case TGSI_OPCODE_DST: + return mask & (c ? 0xa : 0x6); + case TGSI_OPCODE_EX2: + case TGSI_OPCODE_LG2: + case TGSI_OPCODE_POW: + case TGSI_OPCODE_RCP: + case TGSI_OPCODE_RSQ: + case TGSI_OPCODE_SCS: + return 0x1; + case TGSI_OPCODE_LIT: + return 0xb; + case TGSI_OPCODE_TEX: + case TGSI_OPCODE_TXP: + { + const struct tgsi_instruction_ext_texture *tex; + + assert(insn->Instruction.Extended); + tex = &insn->InstructionExtTexture; + + mask = 0x7; + if (insn->Instruction.Opcode == TGSI_OPCODE_TXP) + mask |= 0x8; + + switch (tex->Texture) { + case TGSI_TEXTURE_1D: + mask &= 0x9; + break; + case TGSI_TEXTURE_2D: + mask &= 0xb; + break; + default: + break; + } + } + return mask; + case TGSI_OPCODE_XPD: + x = 0; + if (mask & 1) x |= 0x6; + if (mask & 2) x |= 0x5; + if (mask & 4) x |= 0x3; + return x; + default: + break; + } + + return mask; +} + static struct nv50_reg * tgsi_dst(struct nv50_pc *pc, int c, const struct tgsi_full_dst_register *dst) { @@ -1258,93 +1378,126 @@ tgsi_src(struct nv50_pc *pc, int chan, const struct tgsi_full_src_register *src, return r; } -/* returns TRUE if instruction can overwrite sources before they're read */ +/* return TRUE for ops that produce only a single result */ static boolean -direct2dest_op(const struct tgsi_full_instruction *insn) +is_scalar_op(unsigned op) { - if (insn->Instruction.Saturate) - return FALSE; - - switch (insn->Instruction.Opcode) { + switch (op) { case TGSI_OPCODE_COS: + case TGSI_OPCODE_DP2: case TGSI_OPCODE_DP3: case TGSI_OPCODE_DP4: case TGSI_OPCODE_DPH: - case TGSI_OPCODE_KIL: - case TGSI_OPCODE_LIT: + case TGSI_OPCODE_EX2: + case TGSI_OPCODE_LG2: case TGSI_OPCODE_POW: case TGSI_OPCODE_RCP: case TGSI_OPCODE_RSQ: - case TGSI_OPCODE_SCS: case TGSI_OPCODE_SIN: + /* + case TGSI_OPCODE_KIL: + case TGSI_OPCODE_LIT: + case TGSI_OPCODE_SCS: + */ + return TRUE; + default: + return FALSE; + } +} + +/* Returns a bitmask indicating which dst components depend + * on source s, component c (reverse of nv50_tgsi_src_mask). + */ +static unsigned +nv50_tgsi_dst_revdep(unsigned op, int s, int c) +{ + if (is_scalar_op(op)) + return 0x1; + + switch (op) { + case TGSI_OPCODE_DST: + return (1 << c) & (s ? 0xa : 0x6); + case TGSI_OPCODE_XPD: + switch (c) { + case 0: return 0x6; + case 1: return 0x5; + case 2: return 0x3; + case 3: return 0x0; + default: + assert(0); + return 0x0; + } + case TGSI_OPCODE_LIT: + case TGSI_OPCODE_SCS: case TGSI_OPCODE_TEX: case TGSI_OPCODE_TXP: - return FALSE; + /* these take care of dangerous swizzles themselves */ + return 0x0; + case TGSI_OPCODE_IF: + case TGSI_OPCODE_KIL: + /* don't call this function for these ops */ + assert(0); + return 0; default: - return TRUE; + /* linear vector instruction */ + return (1 << c); } } static boolean -nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) +nv50_program_tx_insn(struct nv50_pc *pc, + const struct tgsi_full_instruction *inst) { - const struct tgsi_full_instruction *inst = &tok->FullInstruction; - struct nv50_reg *rdst[4], *dst[4], *src[3][4], *temp; + struct nv50_reg *rdst[4], *dst[4], *brdc, *src[3][4], *temp; unsigned mask, sat, unit; - boolean assimilate = FALSE; int i, c; mask = inst->FullDstRegisters[0].DstRegister.WriteMask; sat = inst->Instruction.Saturate == TGSI_SAT_ZERO_ONE; + memset(src, 0, sizeof(src)); + for (c = 0; c < 4; c++) { - if (mask & (1 << c)) + if ((mask & (1 << c)) && !pc->r_dst[c]) dst[c] = tgsi_dst(pc, c, &inst->FullDstRegisters[0]); else - dst[c] = NULL; - rdst[c] = NULL; - src[0][c] = NULL; - src[1][c] = NULL; - src[2][c] = NULL; + dst[c] = pc->r_dst[c]; + rdst[c] = dst[c]; } for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { const struct tgsi_full_src_register *fs = &inst->FullSrcRegisters[i]; + unsigned src_mask; + boolean neg_supp; + + src_mask = nv50_tgsi_src_mask(inst, i); + neg_supp = negate_supported(inst, i); if (fs->SrcRegister.File == TGSI_FILE_SAMPLER) unit = fs->SrcRegister.Index; for (c = 0; c < 4; c++) - src[i][c] = tgsi_src(pc, c, fs, - negate_supported(inst, i)); + if (src_mask & (1 << c)) + src[i][c] = tgsi_src(pc, c, fs, neg_supp); } - if (sat) { - for (c = 0; c < 4; c++) { - rdst[c] = dst[c]; - dst[c] = temp_temp(pc); - } + brdc = temp = pc->r_brdc; + if (brdc && brdc->type != P_TEMP) { + temp = temp_temp(pc); + if (sat) + brdc = temp; } else - if (direct2dest_op(inst)) { + if (sat) { for (c = 0; c < 4; c++) { - if (!dst[c] || dst[c]->type != P_TEMP) + if (!(mask & (1 << c)) || dst[c]->type == P_TEMP) continue; - - for (i = c + 1; i < 4; i++) { - if (dst[c] == src[0][i] || - dst[c] == src[1][i] || - dst[c] == src[2][i]) - break; - } - if (i == 4) - continue; - - assimilate = TRUE; rdst[c] = dst[c]; - dst[c] = alloc_temp(pc, NULL); + dst[c] = temp_temp(pc); } } + assert(brdc || !is_scalar_op(inst->Instruction.Opcode)); + switch (inst->Instruction.Opcode) { case TGSI_OPCODE_ABS: for (c = 0; c < 4; c++) { @@ -1360,74 +1513,56 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) emit_add(pc, dst[c], src[0][c], src[1][c]); } break; - case TGSI_OPCODE_COS: - temp = temp_temp(pc); - emit_precossin(pc, temp, src[0][0]); - emit_flop(pc, 5, temp, temp); + case TGSI_OPCODE_CEIL: for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - emit_mov(pc, dst[c], temp); + emit_cvt(pc, dst[c], src[0][c], -1, + CVTOP_CEIL, CVT_F32_F32); + } + break; + case TGSI_OPCODE_COS: + if (mask & 8) { + emit_precossin(pc, temp, src[0][3]); + emit_flop(pc, 5, dst[3], temp); + if (!(mask &= 7)) + break; + if (temp == dst[3]) + temp = brdc = temp_temp(pc); } + emit_precossin(pc, temp, src[0][0]); + emit_flop(pc, 5, brdc, temp); break; case TGSI_OPCODE_DP3: - temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); - emit_mad(pc, temp, src[0][2], src[1][2], temp); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); - } + emit_mad(pc, brdc, src[0][2], src[1][2], temp); break; case TGSI_OPCODE_DP4: - temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); - emit_mad(pc, temp, src[0][3], src[1][3], temp); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); - } + emit_mad(pc, brdc, src[0][3], src[1][3], temp); break; case TGSI_OPCODE_DPH: - temp = temp_temp(pc); emit_mul(pc, temp, src[0][0], src[1][0]); emit_mad(pc, temp, src[0][1], src[1][1], temp); emit_mad(pc, temp, src[0][2], src[1][2], temp); - emit_add(pc, temp, src[1][3], temp); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); - } + emit_add(pc, brdc, src[1][3], temp); break; case TGSI_OPCODE_DST: - { - struct nv50_reg *one = alloc_immd(pc, 1.0); - if (mask & (1 << 0)) - emit_mov(pc, dst[0], one); if (mask & (1 << 1)) emit_mul(pc, dst[1], src[0][1], src[1][1]); if (mask & (1 << 2)) emit_mov(pc, dst[2], src[0][2]); if (mask & (1 << 3)) emit_mov(pc, dst[3], src[1][3]); - FREE(one); - } + if (mask & (1 << 0)) + emit_mov_immdval(pc, dst[0], 1.0f); break; case TGSI_OPCODE_EX2: - temp = temp_temp(pc); emit_preex2(pc, temp, src[0][0]); - emit_flop(pc, 6, temp, temp); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); - } + emit_flop(pc, 6, brdc, temp); break; case TGSI_OPCODE_FLR: for (c = 0; c < 4; c++) { @@ -1450,19 +1585,12 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) emit_kil(pc, src[0][1]); emit_kil(pc, src[0][2]); emit_kil(pc, src[0][3]); - pc->p->cfg.fp.regs[2] |= 0x00100000; break; case TGSI_OPCODE_LIT: emit_lit(pc, &dst[0], mask, &src[0][0]); break; case TGSI_OPCODE_LG2: - temp = temp_temp(pc); - emit_flop(pc, 3, temp, src[0][0]); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); - } + emit_flop(pc, 3, brdc, src[0][0]); break; case TGSI_OPCODE_LRP: temp = temp_temp(pc); @@ -1510,31 +1638,18 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) } break; case TGSI_OPCODE_POW: - temp = temp_temp(pc); - emit_pow(pc, temp, src[0][0], src[1][0]); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); - } + emit_pow(pc, brdc, src[0][0], src[1][0]); break; case TGSI_OPCODE_RCP: - for (c = 3; c >= 0; c--) { - if (!(mask & (1 << c))) - continue; - emit_flop(pc, 0, dst[c], src[0][0]); - } + emit_flop(pc, 0, brdc, src[0][0]); break; case TGSI_OPCODE_RSQ: - for (c = 3; c >= 0; c--) { - if (!(mask & (1 << c))) - continue; - emit_flop(pc, 2, dst[c], src[0][0]); - } + emit_flop(pc, 2, brdc, src[0][0]); break; case TGSI_OPCODE_SCS: temp = temp_temp(pc); - emit_precossin(pc, temp, src[0][0]); + if (mask & 3) + emit_precossin(pc, temp, src[0][0]); if (mask & (1 << 0)) emit_flop(pc, 5, dst[0], temp); if (mask & (1 << 1)) @@ -1544,28 +1659,29 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) if (mask & (1 << 3)) emit_mov_immdval(pc, dst[3], 1.0); break; - case TGSI_OPCODE_SGE: - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_set(pc, 6, dst[c], src[0][c], src[1][c]); - } - break; case TGSI_OPCODE_SIN: - temp = temp_temp(pc); - emit_precossin(pc, temp, src[0][0]); - emit_flop(pc, 4, temp, temp); - for (c = 0; c < 4; c++) { - if (!(mask & (1 << c))) - continue; - emit_mov(pc, dst[c], temp); + if (mask & 8) { + emit_precossin(pc, temp, src[0][3]); + emit_flop(pc, 4, dst[3], temp); + if (!(mask &= 7)) + break; + if (temp == dst[3]) + temp = brdc = temp_temp(pc); } + emit_precossin(pc, temp, src[0][0]); + emit_flop(pc, 4, brdc, temp); break; case TGSI_OPCODE_SLT: + case TGSI_OPCODE_SGE: + case TGSI_OPCODE_SEQ: + case TGSI_OPCODE_SGT: + case TGSI_OPCODE_SLE: + case TGSI_OPCODE_SNE: + i = map_tgsi_setop_cc(inst->Instruction.Opcode); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - emit_set(pc, 1, dst[c], src[0][c], src[1][c]); + emit_set(pc, i, dst[c], -1, src[0][c], src[1][c]); } break; case TGSI_OPCODE_SUB: @@ -1583,6 +1699,14 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) emit_tex(pc, dst, mask, src[0], unit, inst->InstructionExtTexture.Texture, TRUE); break; + case TGSI_OPCODE_TRUNC: + for (c = 0; c < 4; c++) { + if (!(mask & (1 << c))) + continue; + emit_cvt(pc, dst[c], src[0][c], -1, + CVTOP_TRUNC, CVT_F32_F32); + } + break; case TGSI_OPCODE_XPD: temp = temp_temp(pc); if (mask & (1 << 0)) { @@ -1607,17 +1731,22 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) return FALSE; } + if (brdc) { + if (sat) + emit_sat(pc, brdc, brdc); + for (c = 0; c < 4; c++) + if ((mask & (1 << c)) && dst[c] != brdc) + emit_mov(pc, dst[c], brdc); + } else if (sat) { for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - emit_cvt(pc, rdst[c], dst[c], -1, CVTOP_SAT, - CVT_F32_F32); + /* in this case we saturate later */ + if (dst[c]->type == P_TEMP && dst[c]->index < 0) + continue; + emit_sat(pc, rdst[c], dst[c]); } - } else if (assimilate) { - for (c = 0; c < 4; c++) - if (rdst[c]) - assimilate_temp(pc, rdst[c], dst[c]); } for (i = 0; i < inst->Instruction.NumSrcRegs; i++) { @@ -1626,9 +1755,6 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) continue; if (src[i][c]->index == -1 && src[i][c]->type == P_IMMD) FREE(src[i][c]); - else - if (src[i][c]->acc == pc->insn_cur) - release_hw(pc, src[i][c]); } } @@ -1636,180 +1762,271 @@ nv50_program_tx_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) return TRUE; } -/* Adjust a bitmask that indicates what components of a source are used, - * we use this in tx_prep so we only load interpolants that are needed. - */ static void -insn_adjust_mask(const struct tgsi_full_instruction *insn, unsigned *mask) +prep_inspect_insn(struct nv50_pc *pc, const struct tgsi_full_instruction *insn) { - const struct tgsi_instruction_ext_texture *tex; - - switch (insn->Instruction.Opcode) { - case TGSI_OPCODE_DP3: - *mask = 0x7; - break; - case TGSI_OPCODE_DP4: - case TGSI_OPCODE_DPH: - *mask = 0xF; - break; - case TGSI_OPCODE_LIT: - *mask = 0xB; - break; - case TGSI_OPCODE_RCP: - case TGSI_OPCODE_RSQ: - *mask = 0x1; - break; - case TGSI_OPCODE_TEX: - case TGSI_OPCODE_TXP: - assert(insn->Instruction.Extended); - tex = &insn->InstructionExtTexture; - - *mask = 0x7; - if (tex->Texture == TGSI_TEXTURE_1D) - *mask = 0x1; - else - if (tex->Texture == TGSI_TEXTURE_2D) - *mask = 0x3; - - if (insn->Instruction.Opcode == TGSI_OPCODE_TXP) - *mask |= 0x8; - break; - default: - break; - } -} - -static void -prep_inspect_insn(struct nv50_pc *pc, const union tgsi_full_token *tok, - unsigned *r_usage[2]) -{ - const struct tgsi_full_instruction *insn; + struct nv50_reg *reg = NULL; const struct tgsi_full_src_register *src; const struct tgsi_dst_register *dst; + unsigned i, c, k, mask; - unsigned i, c, k, n, mask, *acc_p; - - insn = &tok->FullInstruction; dst = &insn->FullDstRegisters[0].DstRegister; mask = dst->WriteMask; - if (!r_usage[0]) - r_usage[0] = CALLOC(pc->temp_nr * 4, sizeof(unsigned)); - if (!r_usage[1]) - r_usage[1] = CALLOC(pc->attr_nr * 4, sizeof(unsigned)); + if (dst->File == TGSI_FILE_TEMPORARY) + reg = pc->temp; + else + if (dst->File == TGSI_FILE_OUTPUT) + reg = pc->result; - if (dst->File == TGSI_FILE_TEMPORARY) { + if (reg) { for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - r_usage[0][dst->Index * 4 + c] = pc->insn_nr; + reg[dst->Index * 4 + c].acc = pc->insn_nr; } } for (i = 0; i < insn->Instruction.NumSrcRegs; i++) { src = &insn->FullSrcRegisters[i]; - switch (src->SrcRegister.File) { - case TGSI_FILE_TEMPORARY: - acc_p = r_usage[0]; - break; - case TGSI_FILE_INPUT: - acc_p = r_usage[1]; - break; - default: + if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) + reg = pc->temp; + else + if (src->SrcRegister.File == TGSI_FILE_INPUT) + reg = pc->attr; + else continue; - } - insn_adjust_mask(insn, &mask); + mask = nv50_tgsi_src_mask(insn, i); for (c = 0; c < 4; c++) { if (!(mask & (1 << c))) continue; - k = tgsi_util_get_full_src_register_extswizzle(src, c); - switch (k) { - case TGSI_EXTSWIZZLE_X: - case TGSI_EXTSWIZZLE_Y: - case TGSI_EXTSWIZZLE_Z: - case TGSI_EXTSWIZZLE_W: - n = src->SrcRegister.Index * 4 + k; - acc_p[n] = pc->insn_nr; - break; - default: - break; - } + + if (k > TGSI_EXTSWIZZLE_W) + continue; + + reg[src->SrcRegister.Index * 4 + k].acc = pc->insn_nr; } } } +/* Returns a bitmask indicating which dst components need to be + * written to temporaries first to avoid 'corrupting' sources. + * + * m[i] (out) indicate component to write in the i-th position + * rdep[c] (in) bitmasks of dst[i] that require dst[c] as source + */ static unsigned -load_fp_attrib(struct nv50_pc *pc, int i, unsigned *acc, int *mid, - int *aid, int *p_oid) +nv50_revdep_reorder(unsigned m[4], unsigned rdep[4]) { - struct nv50_reg *iv; - int oid, c, n; - unsigned mask = 0; + unsigned i, c, x, unsafe; - iv = (pc->interp_mode[i] & INTERP_CENTROID) ? pc->iv_c : pc->iv_p; + for (c = 0; c < 4; c++) + m[c] = c; - for (c = 0, n = i * 4; c < 4; c++, n++) { - oid = (*p_oid)++; - pc->attr[n].type = P_TEMP; - pc->attr[n].index = i; + /* Swap as long as a dst component written earlier is depended on + * by one written later, but the next one isn't depended on by it. + */ + for (c = 0; c < 3; c++) { + if (rdep[m[c + 1]] & (1 << m[c])) + continue; /* if next one is depended on by us */ + for (i = c + 1; i < 4; i++) + /* if we are depended on by a later one */ + if (rdep[m[c]] & (1 << m[i])) + break; + if (i == 4) + continue; + /* now, swap */ + x = m[c]; + m[c] = m[c + 1]; + m[c + 1] = x; + + /* restart */ + c = 0; + } + + /* mark dependencies that could not be resolved by reordering */ + for (i = 0; i < 3; ++i) + for (c = i + 1; c < 4; ++c) + if (rdep[m[i]] & (1 << m[c])) + unsafe |= (1 << i); + + /* NOTE: $unsafe is with respect to order, not component */ + return unsafe; +} - if (pc->attr[n].acc == acc[n]) +/* Select a suitable dst register for broadcasting scalar results, + * or return NULL if we have to allocate an extra TEMP. + * + * If e.g. only 1 component is written, we may also emit the final + * result to a write-only register. + */ +static struct nv50_reg * +tgsi_broadcast_dst(struct nv50_pc *pc, + const struct tgsi_full_dst_register *fd, unsigned mask) +{ + if (fd->DstRegister.File == TGSI_FILE_TEMPORARY) { + int c = ffs(~mask & fd->DstRegister.WriteMask); + if (c) + return tgsi_dst(pc, c - 1, fd); + } else { + int c = ffs(fd->DstRegister.WriteMask) - 1; + if ((1 << c) == fd->DstRegister.WriteMask) + return tgsi_dst(pc, c, fd); + } + + return NULL; +} + +/* Scan source swizzles and return a bitmask indicating dst regs that + * also occur among the src regs, and fill rdep for nv50_revdep_reoder. + */ +static unsigned +nv50_tgsi_scan_swizzle(const struct tgsi_full_instruction *insn, + unsigned rdep[4]) +{ + const struct tgsi_full_dst_register *fd = &insn->FullDstRegisters[0]; + const struct tgsi_full_src_register *fs; + unsigned i, deqs = 0; + + for (i = 0; i < 4; ++i) + rdep[i] = 0; + + for (i = 0; i < insn->Instruction.NumSrcRegs; i++) { + unsigned chn, mask = nv50_tgsi_src_mask(insn, i); + boolean neg_supp = negate_supported(insn, i); + + fs = &insn->FullSrcRegisters[i]; + if (fs->SrcRegister.File != fd->DstRegister.File || + fs->SrcRegister.Index != fd->DstRegister.Index) continue; - mask |= (1 << c); - pc->attr[n].acc = acc[n]; - pc->attr[n].rhw = pc->attr[n].hw = -1; - alloc_reg(pc, &pc->attr[n]); + for (chn = 0; chn < 4; ++chn) { + unsigned s, c; + + if (!(mask & (1 << chn))) /* src is not read */ + continue; + c = tgsi_util_get_full_src_register_extswizzle(fs, chn); + s = tgsi_util_get_full_src_register_sign_mode(fs, chn); - pc->attr[n].rhw = (*aid)++; - emit_interp(pc, &pc->attr[n], iv, pc->interp_mode[i]); + if (c > TGSI_EXTSWIZZLE_W || + !(fd->DstRegister.WriteMask & (1 << c))) + continue; - pc->p->cfg.fp.map[(*mid) / 4] |= oid << (8 * ((*mid) % 4)); - (*mid)++; - pc->p->cfg.fp.regs[1] += 0x00010001; + /* no danger if src is copied to TEMP first */ + if ((s != TGSI_UTIL_SIGN_KEEP) && + (s != TGSI_UTIL_SIGN_TOGGLE || !neg_supp)) + continue; + + rdep[c] |= nv50_tgsi_dst_revdep( + insn->Instruction.Opcode, i, chn); + deqs |= (1 << c); + } } - return mask; + return deqs; } static boolean -nv50_program_tx_prep(struct nv50_pc *pc) +nv50_tgsi_insn(struct nv50_pc *pc, const union tgsi_full_token *tok) { - struct tgsi_parse_context p; - boolean ret = FALSE; - unsigned i, c; - unsigned fcol, bcol, fcrd, depr; + struct tgsi_full_instruction insn = tok->FullInstruction; + const struct tgsi_full_dst_register *fd; + unsigned i, deqs, rdep[4], m[4]; + + fd = &tok->FullInstruction.FullDstRegisters[0]; + deqs = nv50_tgsi_scan_swizzle(&insn, rdep); + + if (is_scalar_op(insn.Instruction.Opcode)) { + pc->r_brdc = tgsi_broadcast_dst(pc, fd, deqs); + if (!pc->r_brdc) + pc->r_brdc = temp_temp(pc); + return nv50_program_tx_insn(pc, &insn); + } + pc->r_brdc = NULL; + + if (!deqs) + return nv50_program_tx_insn(pc, &insn); + + deqs = nv50_revdep_reorder(m, rdep); - /* count (centroid) perspective interpolations */ - unsigned centroid_loads = 0; - unsigned perspect_loads = 0; + for (i = 0; i < 4; ++i) { + assert(pc->r_dst[m[i]] == NULL); - /* track register access for temps and attrs */ - unsigned *r_usage[2]; - r_usage[0] = NULL; - r_usage[1] = NULL; + insn.FullDstRegisters[0].DstRegister.WriteMask = + fd->DstRegister.WriteMask & (1 << m[i]); - depr = fcol = bcol = fcrd = 0xffff; + if (!insn.FullDstRegisters[0].DstRegister.WriteMask) + continue; + + if (deqs & (1 << i)) + pc->r_dst[m[i]] = alloc_temp(pc, NULL); - if (pc->p->type == PIPE_SHADER_FRAGMENT) { - pc->p->cfg.fp.regs[0] = 0x01000404; - pc->p->cfg.fp.regs[1] = 0x00000400; + if (!nv50_program_tx_insn(pc, &insn)) + return FALSE; } - tgsi_parse_init(&p, pc->p->pipe.tokens); - while (!tgsi_parse_end_of_tokens(&p)) { - const union tgsi_full_token *tok = &p.FullToken; + for (i = 0; i < 4; i++) { + struct nv50_reg *reg = pc->r_dst[i]; + if (!reg) + continue; + pc->r_dst[i] = NULL; + + if (insn.Instruction.Saturate == TGSI_SAT_ZERO_ONE) + emit_sat(pc, tgsi_dst(pc, i, fd), reg); + else + emit_mov(pc, tgsi_dst(pc, i, fd), reg); + free_temp(pc, reg); + } - tgsi_parse_token(&p); + return TRUE; +} + +static void +load_interpolant(struct nv50_pc *pc, struct nv50_reg *reg) +{ + struct nv50_reg *iv, **ppiv; + unsigned mode = pc->interp_mode[reg->index]; + + ppiv = (mode & INTERP_CENTROID) ? &pc->iv_c : &pc->iv_p; + iv = *ppiv; + + if ((mode & INTERP_PERSPECTIVE) && !iv) { + iv = *ppiv = alloc_temp(pc, NULL); + iv->rhw = popcnt4(pc->p->cfg.regs[1] >> 24) - 1; + + emit_interp(pc, iv, NULL, mode & INTERP_CENTROID); + emit_flop(pc, 0, iv, iv); + + /* XXX: when loading interpolants dynamically, move these + * to the program head, or make sure it can't be skipped. + */ + } + + emit_interp(pc, reg, iv, mode); +} + +static boolean +nv50_program_tx_prep(struct nv50_pc *pc) +{ + struct tgsi_parse_context tp; + struct nv50_program *p = pc->p; + boolean ret = FALSE; + unsigned i, c, flat_nr = 0; + + tgsi_parse_init(&tp, pc->p->pipe.tokens); + while (!tgsi_parse_end_of_tokens(&tp)) { + const union tgsi_full_token *tok = &tp.FullToken; + + tgsi_parse_token(&tp); switch (tok->Token.Type) { case TGSI_TOKEN_TYPE_IMMEDIATE: { const struct tgsi_full_immediate *imm = - &p.FullToken.FullImmediate; + &tp.FullToken.FullImmediate; ctor_immd(pc, imm->u[0].Float, imm->u[1].Float, @@ -1820,78 +2037,61 @@ nv50_program_tx_prep(struct nv50_pc *pc) case TGSI_TOKEN_TYPE_DECLARATION: { const struct tgsi_full_declaration *d; - unsigned last, first, mode; + unsigned si, last, first, mode; - d = &p.FullToken.FullDeclaration; + d = &tp.FullToken.FullDeclaration; first = d->DeclarationRange.First; last = d->DeclarationRange.Last; switch (d->Declaration.File) { case TGSI_FILE_TEMPORARY: - if (pc->temp_nr < (last + 1)) - pc->temp_nr = last + 1; break; case TGSI_FILE_OUTPUT: - if (pc->result_nr < (last + 1)) - pc->result_nr = last + 1; - - if (!d->Declaration.Semantic) + if (!d->Declaration.Semantic || + p->type == PIPE_SHADER_FRAGMENT) break; + si = d->Semantic.SemanticIndex; switch (d->Semantic.SemanticName) { - case TGSI_SEMANTIC_POSITION: - depr = first; - pc->p->cfg.fp.regs[2] |= 0x00000100; - pc->p->cfg.fp.regs[3] |= 0x00000011; + case TGSI_SEMANTIC_BCOLOR: + p->cfg.two_side[si].hw = first; + if (p->cfg.io_nr > first) + p->cfg.io_nr = first; + break; + case TGSI_SEMANTIC_PSIZE: + p->cfg.psiz = first; + if (p->cfg.io_nr > first) + p->cfg.io_nr = first; + break; + /* + case TGSI_SEMANTIC_CLIP_DISTANCE: + p->cfg.clpd = MIN2(p->cfg.clpd, first); break; + */ default: break; } - break; case TGSI_FILE_INPUT: { - if (pc->attr_nr < (last + 1)) - pc->attr_nr = last + 1; - - if (pc->p->type != PIPE_SHADER_FRAGMENT) + if (p->type != PIPE_SHADER_FRAGMENT) break; switch (d->Declaration.Interpolate) { case TGSI_INTERPOLATE_CONSTANT: mode = INTERP_FLAT; + flat_nr++; break; case TGSI_INTERPOLATE_PERSPECTIVE: mode = INTERP_PERSPECTIVE; + p->cfg.regs[1] |= 0x08 << 24; break; default: mode = INTERP_LINEAR; break; } - - if (d->Declaration.Semantic) { - switch (d->Semantic.SemanticName) { - case TGSI_SEMANTIC_POSITION: - fcrd = first; - break; - case TGSI_SEMANTIC_COLOR: - fcol = first; - mode = INTERP_PERSPECTIVE; - break; - case TGSI_SEMANTIC_BCOLOR: - bcol = first; - mode = INTERP_PERSPECTIVE; - break; - } - } - - if (d->Declaration.Centroid) { + if (d->Declaration.Centroid) mode |= INTERP_CENTROID; - if (mode & INTERP_PERSPECTIVE) - centroid_loads++; - } else - if (mode & INTERP_PERSPECTIVE) - perspect_loads++; assert(last < 32); for (i = first; i <= last; i++) @@ -1899,8 +2099,6 @@ nv50_program_tx_prep(struct nv50_pc *pc) } break; case TGSI_FILE_CONSTANT: - if (pc->param_nr < (last + 1)) - pc->param_nr = last + 1; break; case TGSI_FILE_SAMPLER: break; @@ -1913,182 +2111,155 @@ nv50_program_tx_prep(struct nv50_pc *pc) break; case TGSI_TOKEN_TYPE_INSTRUCTION: pc->insn_nr++; - prep_inspect_insn(pc, tok, r_usage); + prep_inspect_insn(pc, &tok->FullInstruction); break; default: break; } } - if (pc->temp_nr) { - pc->temp = CALLOC(pc->temp_nr * 4, sizeof(struct nv50_reg)); - if (!pc->temp) - goto out_err; + if (p->type == PIPE_SHADER_VERTEX) { + int rid = 0; - for (i = 0; i < pc->temp_nr; i++) { - for (c = 0; c < 4; c++) { - pc->temp[i*4+c].type = P_TEMP; - pc->temp[i*4+c].hw = -1; - pc->temp[i*4+c].rhw = -1; - pc->temp[i*4+c].index = i; - pc->temp[i*4+c].acc = r_usage[0][i*4+c]; + for (i = 0; i < pc->attr_nr * 4; ++i) { + if (pc->attr[i].acc) { + pc->attr[i].hw = rid++; + p->cfg.attr[i / 32] |= 1 << (i % 32); } } - } - - if (pc->attr_nr) { - int oid = 4, mid = 4, aid = 0; - /* oid = VP output id - * aid = FP attribute/interpolant id - * mid = VP output mapping field ID - */ - pc->attr = CALLOC(pc->attr_nr * 4, sizeof(struct nv50_reg)); - if (!pc->attr) - goto out_err; + for (i = 0, rid = 0; i < pc->result_nr; ++i) { + p->cfg.io[i].hw = rid; + p->cfg.io[i].id_vp = i; - if (pc->p->type == PIPE_SHADER_FRAGMENT) { - /* position should be loaded first */ - if (fcrd != 0xffff) { - unsigned mask; - mid = 0; - mask = load_fp_attrib(pc, fcrd, r_usage[1], - &mid, &aid, &oid); - oid = 0; - pc->p->cfg.fp.regs[1] |= (mask << 24); - pc->p->cfg.fp.map[0] = 0x04040404 * fcrd; + for (c = 0; c < 4; ++c) { + int n = i * 4 + c; + if (!pc->result[n].acc) + continue; + pc->result[n].hw = rid++; + p->cfg.io[i].mask |= 1 << c; } - pc->p->cfg.fp.map[0] += 0x03020100; + } - /* should do MAD fcrd.xy, fcrd, SOME_CONST, fcrd */ + for (c = 0; c < 2; ++c) + if (p->cfg.two_side[c].hw < 0x40) + p->cfg.two_side[c] = p->cfg.io[ + p->cfg.two_side[c].hw]; - if (perspect_loads) { - pc->iv_p = alloc_temp(pc, NULL); + if (p->cfg.psiz < 0x40) + p->cfg.psiz = p->cfg.io[p->cfg.psiz].hw; + } else + if (p->type == PIPE_SHADER_FRAGMENT) { + int rid, aid; + unsigned n = 0, m = pc->attr_nr - flat_nr; - if (!(pc->p->cfg.fp.regs[1] & 0x08000000)) { - pc->p->cfg.fp.regs[1] |= 0x08000000; - pc->iv_p->rhw = aid++; - emit_interp(pc, pc->iv_p, NULL, - INTERP_LINEAR); - emit_flop(pc, 0, pc->iv_p, pc->iv_p); - } else { - pc->iv_p->rhw = aid - 1; - emit_flop(pc, 0, pc->iv_p, - &pc->attr[fcrd * 4 + 3]); - } - } + int base = (TGSI_SEMANTIC_POSITION == + p->info.input_semantic_name[0]) ? 0 : 1; - if (centroid_loads) { - pc->iv_c = alloc_temp(pc, NULL); - pc->iv_c->rhw = pc->iv_p ? aid - 1 : aid++; - emit_interp(pc, pc->iv_c, NULL, - INTERP_CENTROID); - emit_flop(pc, 0, pc->iv_c, pc->iv_c); - pc->p->cfg.fp.regs[1] |= 0x08000000; + /* non-flat interpolants have to be mapped to + * the lower hardware IDs, so sort them: + */ + for (i = 0; i < pc->attr_nr; i++) { + if (pc->interp_mode[i] == INTERP_FLAT) { + p->cfg.io[m].id_vp = i + base; + p->cfg.io[m++].id_fp = i; + } else { + if (!(pc->interp_mode[i] & INTERP_PERSPECTIVE)) + p->cfg.io[n].linear = TRUE; + p->cfg.io[n].id_vp = i + base; + p->cfg.io[n++].id_fp = i; } + } - for (c = 0; c < 4; c++) { - /* I don't know what these values do, but - * let's set them like the blob does: - */ - if (fcol != 0xffff && r_usage[1][fcol * 4 + c]) - pc->p->cfg.fp.regs[0] += 0x00010000; - if (bcol != 0xffff && r_usage[1][bcol * 4 + c]) - pc->p->cfg.fp.regs[0] += 0x00010000; - } + if (!base) /* set w-coordinate mask from perspective interp */ + p->cfg.io[0].mask |= p->cfg.regs[1] >> 24; - for (i = 0; i < pc->attr_nr; i++) - load_fp_attrib(pc, i, r_usage[1], - &mid, &aid, &oid); + aid = popcnt4( /* if fcrd isn't contained in cfg.io */ + base ? (p->cfg.regs[1] >> 24) : p->cfg.io[0].mask); - if (pc->iv_p) - free_temp(pc, pc->iv_p); - if (pc->iv_c) - free_temp(pc, pc->iv_c); + for (n = 0; n < pc->attr_nr; ++n) { + p->cfg.io[n].hw = rid = aid; + i = p->cfg.io[n].id_fp; - pc->p->cfg.fp.high_map = (mid / 4); - pc->p->cfg.fp.high_map += ((mid % 4) ? 1 : 0); - } else { - /* vertex program */ - for (i = 0; i < pc->attr_nr * 4; i++) { - pc->p->cfg.vp.attr[aid / 32] |= - (1 << (aid % 32)); - pc->attr[i].type = P_ATTR; - pc->attr[i].hw = aid++; - pc->attr[i].index = i / 4; + for (c = 0; c < 4; ++c) { + if (!pc->attr[i * 4 + c].acc) + continue; + pc->attr[i * 4 + c].rhw = rid++; + p->cfg.io[n].mask |= 1 << c; + + load_interpolant(pc, &pc->attr[i * 4 + c]); } + aid += popcnt4(p->cfg.io[n].mask); } - } - if (pc->result_nr) { - int rid = 0; + if (!base) + p->cfg.regs[1] |= p->cfg.io[0].mask << 24; - pc->result = CALLOC(pc->result_nr * 4, sizeof(struct nv50_reg)); - if (!pc->result) - goto out_err; + m = popcnt4(p->cfg.regs[1] >> 24); - for (i = 0; i < pc->result_nr; i++) { - for (c = 0; c < 4; c++) { - if (pc->p->type == PIPE_SHADER_FRAGMENT) { - pc->result[i*4+c].type = P_TEMP; - pc->result[i*4+c].hw = -1; - pc->result[i*4+c].rhw = (i == depr) ? - -1 : rid++; - } else { - pc->result[i*4+c].type = P_RESULT; - pc->result[i*4+c].hw = rid++; - } - pc->result[i*4+c].index = i; - } + /* set count of non-position inputs and of non-flat + * non-position inputs for FP_INTERPOLANT_CTRL + */ + p->cfg.regs[1] |= aid - m; - if (pc->p->type == PIPE_SHADER_FRAGMENT && - depr != 0xffff) { - pc->result[depr * 4 + 2].rhw = - (pc->result_nr - 1) * 4; - } - } - } + if (flat_nr) { + i = p->cfg.io[pc->attr_nr - flat_nr].hw; + p->cfg.regs[1] |= (i - m) << 16; + } else + p->cfg.regs[1] |= p->cfg.regs[1] << 16; - if (pc->param_nr) { - int rid = 0; + /* mark color semantic for light-twoside */ + n = 0x40; + for (i = 0; i < pc->attr_nr; i++) { + ubyte si, sn; - pc->param = CALLOC(pc->param_nr * 4, sizeof(struct nv50_reg)); - if (!pc->param) - goto out_err; + sn = p->info.input_semantic_name[p->cfg.io[i].id_fp]; + si = p->info.input_semantic_index[p->cfg.io[i].id_fp]; + + if (sn == TGSI_SEMANTIC_COLOR) { + p->cfg.two_side[si] = p->cfg.io[i]; + + /* increase colour count */ + p->cfg.regs[0] += popcnt4( + p->cfg.two_side[si].mask) << 16; - for (i = 0; i < pc->param_nr; i++) { - for (c = 0; c < 4; c++) { - pc->param[i*4+c].type = P_CONST; - pc->param[i*4+c].hw = rid++; - pc->param[i*4+c].index = i; + n = MIN2(n, p->cfg.io[i].hw - m); } } + if (n < 0x40) + p->cfg.regs[0] += n; + + /* Initialize FP results: + * FragDepth is always first TGSI and last hw output + */ + i = p->info.writes_z ? 4 : 0; + for (rid = 0; i < pc->result_nr * 4; i++) + pc->result[i].rhw = rid++; + if (p->info.writes_z) + pc->result[2].rhw = rid; } if (pc->immd_nr) { int rid = 0; - pc->immd = CALLOC(pc->immd_nr * 4, sizeof(struct nv50_reg)); + pc->immd = MALLOC(pc->immd_nr * 4 * sizeof(struct nv50_reg)); if (!pc->immd) goto out_err; for (i = 0; i < pc->immd_nr; i++) { - for (c = 0; c < 4; c++) { - pc->immd[i*4+c].type = P_IMMD; - pc->immd[i*4+c].hw = rid++; - pc->immd[i*4+c].index = i; - } + for (c = 0; c < 4; c++, rid++) + ctor_reg(&pc->immd[rid], P_IMMD, i, rid); } } ret = TRUE; out_err: - if (r_usage[0]) - FREE(r_usage[0]); - if (r_usage[1]) - FREE(r_usage[1]); + if (pc->iv_p) + free_temp(pc, pc->iv_p); + if (pc->iv_c) + free_temp(pc, pc->iv_c); - tgsi_parse_free(&p); + tgsi_parse_free(&tp); return ret; } @@ -2110,6 +2281,88 @@ free_nv50_pc(struct nv50_pc *pc) } static boolean +ctor_nv50_pc(struct nv50_pc *pc, struct nv50_program *p) +{ + int i, c; + unsigned rtype[2] = { P_ATTR, P_RESULT }; + + pc->p = p; + pc->temp_nr = p->info.file_max[TGSI_FILE_TEMPORARY] + 1; + pc->attr_nr = p->info.file_max[TGSI_FILE_INPUT] + 1; + pc->result_nr = p->info.file_max[TGSI_FILE_OUTPUT] + 1; + pc->param_nr = p->info.file_max[TGSI_FILE_CONSTANT] + 1; + + p->cfg.high_temp = 4; + + p->cfg.two_side[0].hw = 0x40; + p->cfg.two_side[1].hw = 0x40; + + switch (p->type) { + case PIPE_SHADER_VERTEX: + p->cfg.psiz = 0x40; + p->cfg.clpd = 0x40; + p->cfg.io_nr = pc->result_nr; + break; + case PIPE_SHADER_FRAGMENT: + rtype[0] = rtype[1] = P_TEMP; + + p->cfg.regs[0] = 0x01000004; + p->cfg.io_nr = pc->attr_nr; + + if (p->info.writes_z) { + p->cfg.regs[2] |= 0x00000100; + p->cfg.regs[3] |= 0x00000011; + } + if (p->info.uses_kill) + p->cfg.regs[2] |= 0x00100000; + break; + } + + if (pc->temp_nr) { + pc->temp = MALLOC(pc->temp_nr * 4 * sizeof(struct nv50_reg)); + if (!pc->temp) + return FALSE; + + for (i = 0; i < pc->temp_nr * 4; ++i) + ctor_reg(&pc->temp[i], P_TEMP, i / 4, -1); + } + + if (pc->attr_nr) { + pc->attr = MALLOC(pc->attr_nr * 4 * sizeof(struct nv50_reg)); + if (!pc->attr) + return FALSE; + + for (i = 0; i < pc->attr_nr * 4; ++i) + ctor_reg(&pc->attr[i], rtype[0], i / 4, -1); + } + + if (pc->result_nr) { + unsigned nr = pc->result_nr * 4; + + pc->result = MALLOC(nr * sizeof(struct nv50_reg)); + if (!pc->result) + return FALSE; + + for (i = 0; i < nr; ++i) + ctor_reg(&pc->result[i], rtype[1], i / 4, -1); + } + + if (pc->param_nr) { + int rid = 0; + + pc->param = MALLOC(pc->param_nr * 4 * sizeof(struct nv50_reg)); + if (!pc->param) + return FALSE; + + for (i = 0; i < pc->param_nr; ++i) + for (c = 0; c < 4; ++c, ++rid) + ctor_reg(&pc->param[rid], P_CONST, i, rid); + } + + return TRUE; +} + +static boolean nv50_program_tx(struct nv50_program *p) { struct tgsi_parse_context parse; @@ -2120,8 +2373,10 @@ nv50_program_tx(struct nv50_program *p) pc = CALLOC_STRUCT(nv50_pc); if (!pc) return FALSE; - pc->p = p; - pc->p->cfg.high_temp = 4; + + ret = ctor_nv50_pc(pc, p); + if (ret == FALSE) + goto out_cleanup; ret = nv50_program_tx_prep(pc); if (ret == FALSE) @@ -2141,7 +2396,7 @@ nv50_program_tx(struct nv50_program *p) switch (tok->Token.Type) { case TGSI_TOKEN_TYPE_INSTRUCTION: ++pc->insn_cur; - ret = nv50_program_tx_insn(pc, tok); + ret = nv50_tgsi_insn(pc, tok); if (ret == FALSE) goto out_err; break; @@ -2152,8 +2407,8 @@ nv50_program_tx(struct nv50_program *p) if (p->type == PIPE_SHADER_FRAGMENT) { struct nv50_reg out; + ctor_reg(&out, P_TEMP, -1, -1); - out.type = P_TEMP; for (k = 0; k < pc->result_nr * 4; k++) { if (pc->result[k].rhw == -1) continue; @@ -2258,30 +2513,19 @@ nv50_program_validate_data(struct nv50_context *nv50, struct nv50_program *p) p->immd_nr, NV50_CB_PMISC); } - if (!p->data[1] && p->param_nr) { - struct nouveau_resource *heap = - nv50->screen->parm_heap[p->type]; - - if (nouveau_resource_alloc(heap, p->param_nr, p, &p->data[1])) { - while (heap->next && heap->size < p->param_nr) { - struct nv50_program *evict = heap->next->priv; - nouveau_resource_free(&evict->data[1]); - } - - if (nouveau_resource_alloc(heap, p->param_nr, p, - &p->data[1])) - assert(0); - } - } + assert(p->param_nr <= 128); if (p->param_nr) { - unsigned cbuf = NV50_CB_PVP; + unsigned cb; float *map = pipe_buffer_map(pscreen, nv50->constbuf[p->type], PIPE_BUFFER_USAGE_CPU_READ); - if (p->type == PIPE_SHADER_FRAGMENT) - cbuf = NV50_CB_PFP; - nv50_program_upload_data(nv50, map, p->data[1]->start, - p->param_nr, cbuf); + + if (p->type == PIPE_SHADER_VERTEX) + cb = NV50_CB_PVP; + else + cb = NV50_CB_PFP; + + nv50_program_upload_data(nv50, map, 0, p->param_nr, cb); pipe_buffer_unmap(pscreen, nv50->constbuf[p->type]); } } @@ -2303,32 +2547,30 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p) upload = TRUE; } - if ((p->data[0] && p->data[0]->start != p->data_start[0]) || - (p->data[1] && p->data[1]->start != p->data_start[1])) { - for (e = p->exec_head; e; e = e->next) { - unsigned ei, ci, bs; + if (p->data[0] && p->data[0]->start != p->data_start[0]) + upload = TRUE; - if (e->param.index < 0) - continue; - bs = (e->inst[1] >> 22) & 0x07; - assert(bs < 2); - ei = e->param.shift >> 5; - ci = e->param.index + p->data[bs]->start; + if (!upload) + return; - e->inst[ei] &= ~e->param.mask; - e->inst[ei] |= (ci << e->param.shift); - } + for (e = p->exec_head; e; e = e->next) { + unsigned ei, ci, bs; - if (p->data[0]) - p->data_start[0] = p->data[0]->start; - if (p->data[1]) - p->data_start[1] = p->data[1]->start; + if (e->param.index < 0) + continue; + bs = (e->inst[1] >> 22) & 0x07; + assert(bs < 2); + ei = e->param.shift >> 5; + ci = e->param.index; + if (bs == 0) + ci += p->data[bs]->start; - upload = TRUE; + e->inst[ei] &= ~e->param.mask; + e->inst[ei] |= (ci << e->param.shift); } - if (!upload) - return; + if (p->data[0]) + p->data_start[0] = p->data[0]->start; #ifdef NV50_PROGRAM_DUMP NOUVEAU_ERR("-------\n"); @@ -2402,8 +2644,8 @@ nv50_vertprog_validate(struct nv50_context *nv50) so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); so_method(so, tesla, NV50TCL_VP_ATTR_EN_0, 2); - so_data (so, p->cfg.vp.attr[0]); - so_data (so, p->cfg.vp.attr[1]); + so_data (so, p->cfg.attr[0]); + so_data (so, p->cfg.attr[1]); so_method(so, tesla, NV50TCL_VP_REG_ALLOC_RESULT, 1); so_data (so, p->cfg.high_result); so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 2); @@ -2421,7 +2663,6 @@ nv50_fragprog_validate(struct nv50_context *nv50) struct nouveau_grobj *tesla = nv50->screen->tesla; struct nv50_program *p = nv50->fragprog; struct nouveau_stateobj *so; - unsigned i; if (!p->translated) { nv50_program_validate(nv50, p); @@ -2438,29 +2679,186 @@ nv50_fragprog_validate(struct nv50_context *nv50) NOUVEAU_BO_HIGH, 0, 0); so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD | NOUVEAU_BO_LOW, 0, 0); - so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4); - so_data (so, p->cfg.fp.regs[0]); /* 0x01000404 / 0x00040404 */ - so_data (so, 0x00000004); - so_data (so, 0x00000000); - so_data (so, 0x00000000); - so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), p->cfg.fp.high_map); - for (i = 0; i < p->cfg.fp.high_map; i++) - so_data(so, p->cfg.fp.map[i]); - so_method(so, tesla, NV50TCL_FP_INTERPOLANT_CTRL, 2); - so_data (so, p->cfg.fp.regs[1]); /* 0x08040404 / 0x0f000401 */ + so_method(so, tesla, NV50TCL_FP_REG_ALLOC_TEMP, 1); so_data (so, p->cfg.high_temp); so_method(so, tesla, NV50TCL_FP_RESULT_COUNT, 1); so_data (so, p->cfg.high_result); so_method(so, tesla, NV50TCL_FP_CTRL_UNK19A8, 1); - so_data (so, p->cfg.fp.regs[2]); + so_data (so, p->cfg.regs[2]); so_method(so, tesla, NV50TCL_FP_CTRL_UNK196C, 1); - so_data (so, p->cfg.fp.regs[3]); + so_data (so, p->cfg.regs[3]); so_method(so, tesla, NV50TCL_FP_START_ID, 1); so_data (so, 0); /* program start offset */ so_ref(so, &nv50->state.fragprog); so_ref(NULL, &so); } +static void +nv50_pntc_replace(struct nv50_context *nv50, uint32_t pntc[8], unsigned base) +{ + struct nv50_program *fp = nv50->fragprog; + struct nv50_program *vp = nv50->vertprog; + unsigned i, c, m = base; + + /* XXX: This can't work correctly in all cases yet, we either + * have to create TGSI_SEMANTIC_PNTC or sprite_coord_mode has + * to be per FP input instead of per VP output + */ + memset(pntc, 0, 8 * sizeof(uint32_t)); + + for (i = 0; i < fp->cfg.io_nr; i++) { + uint8_t sn, si; + uint8_t j = fp->cfg.io[i].id_vp, k = fp->cfg.io[i].id_fp; + unsigned n = popcnt4(fp->cfg.io[i].mask); + + if (fp->info.input_semantic_name[k] != TGSI_SEMANTIC_GENERIC) { + m += n; + continue; + } + + sn = vp->info.input_semantic_name[j]; + si = vp->info.input_semantic_index[j]; + + if (j < fp->cfg.io_nr && sn == TGSI_SEMANTIC_GENERIC) { + ubyte mode = + nv50->rasterizer->pipe.sprite_coord_mode[si]; + + if (mode == PIPE_SPRITE_COORD_NONE) { + m += n; + continue; + } + } + + /* this is either PointCoord or replaced by sprite coords */ + for (c = 0; c < 4; c++) { + if (!(fp->cfg.io[i].mask & (1 << c))) + continue; + pntc[m / 8] |= (c + 1) << ((m % 8) * 4); + ++m; + } + } +} + +static int +nv50_sreg4_map(uint32_t *p_map, int mid, uint32_t lin[4], + struct nv50_sreg4 *fpi, struct nv50_sreg4 *vpo) +{ + int c; + uint8_t mv = vpo->mask, mf = fpi->mask, oid = vpo->hw; + uint8_t *map = (uint8_t *)p_map; + + for (c = 0; c < 4; ++c) { + if (mf & 1) { + if (fpi->linear == TRUE) + lin[mid / 32] |= 1 << (mid % 32); + map[mid++] = (mv & 1) ? oid : ((c == 3) ? 0x41 : 0x40); + } + + oid += mv & 1; + mf >>= 1; + mv >>= 1; + } + + return mid; +} + +void +nv50_linkage_validate(struct nv50_context *nv50) +{ + struct nouveau_grobj *tesla = nv50->screen->tesla; + struct nv50_program *vp = nv50->vertprog; + struct nv50_program *fp = nv50->fragprog; + struct nouveau_stateobj *so; + struct nv50_sreg4 dummy, *vpo; + int i, n, c, m = 0; + uint32_t map[16], lin[4], reg[5], pcrd[8]; + + memset(map, 0, sizeof(map)); + memset(lin, 0, sizeof(lin)); + + reg[1] = 0x00000004; /* low and high clip distance map ids */ + reg[2] = 0x00000000; /* layer index map id (disabled, GP only) */ + reg[3] = 0x00000000; /* point size map id & enable */ + reg[0] = fp->cfg.regs[0]; /* colour semantic reg */ + reg[4] = fp->cfg.regs[1]; /* interpolant info */ + + dummy.linear = FALSE; + dummy.mask = 0xf; /* map all components of HPOS */ + m = nv50_sreg4_map(map, m, lin, &dummy, &vp->cfg.io[0]); + + dummy.mask = 0x0; + + if (vp->cfg.clpd < 0x40) { + for (c = 0; c < vp->cfg.clpd_nr; ++c) + map[m++] = vp->cfg.clpd + c; + reg[1] = (m << 8); + } + + reg[0] |= m << 8; /* adjust BFC0 id */ + + /* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */ + if (nv50->rasterizer->pipe.light_twoside) { + vpo = &vp->cfg.two_side[0]; + + m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[0], &vpo[0]); + m = nv50_sreg4_map(map, m, lin, &fp->cfg.two_side[1], &vpo[1]); + } + + reg[0] += m - 4; /* adjust FFC0 id */ + reg[4] |= m << 8; /* set mid where 'normal' FP inputs start */ + + i = 0; + if (fp->info.input_semantic_name[0] == TGSI_SEMANTIC_POSITION) + i = 1; + for (; i < fp->cfg.io_nr; i++) { + ubyte sn = fp->info.input_semantic_name[fp->cfg.io[i].id_fp]; + ubyte si = fp->info.input_semantic_index[fp->cfg.io[i].id_fp]; + + n = fp->cfg.io[i].id_vp; + if (n >= vp->cfg.io_nr || + vp->info.output_semantic_name[n] != sn || + vp->info.output_semantic_index[n] != si) + vpo = &dummy; + else + vpo = &vp->cfg.io[n]; + + m = nv50_sreg4_map(map, m, lin, &fp->cfg.io[i], vpo); + } + + if (nv50->rasterizer->pipe.point_size_per_vertex) { + map[m / 4] |= vp->cfg.psiz << ((m % 4) * 8); + reg[3] = (m++ << 4) | 1; + } + + /* now fill the stateobj */ + so = so_new(64, 0); + + n = (m + 3) / 4; + so_method(so, tesla, NV50TCL_VP_RESULT_MAP_SIZE, 1); + so_data (so, m); + so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), n); + so_datap (so, map, n); + + so_method(so, tesla, NV50TCL_MAP_SEMANTIC_0, 4); + so_datap (so, reg, 4); + + so_method(so, tesla, NV50TCL_FP_INTERPOLANT_CTRL, 1); + so_data (so, reg[4]); + + so_method(so, tesla, 0x1540, 4); + so_datap (so, lin, 4); + + if (nv50->rasterizer->pipe.point_sprite) { + nv50_pntc_replace(nv50, pcrd, (reg[4] >> 8) & 0xff); + + so_method(so, tesla, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8); + so_datap (so, pcrd, 8); + } + + so_ref(so, &nv50->state.programs); + so_ref(NULL, &so); +} + void nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p) { @@ -2476,7 +2874,6 @@ nv50_program_destroy(struct nv50_context *nv50, struct nv50_program *p) nouveau_bo_ref(NULL, &p->bo); nouveau_resource_free(&p->data[0]); - nouveau_resource_free(&p->data[1]); p->translated = 0; } diff --git a/src/gallium/drivers/nv50/nv50_program.h b/src/gallium/drivers/nv50/nv50_program.h index 096e0476aa..d78dee083f 100644 --- a/src/gallium/drivers/nv50/nv50_program.h +++ b/src/gallium/drivers/nv50/nv50_program.h @@ -15,6 +15,15 @@ struct nv50_program_exec { } param; }; +struct nv50_sreg4 { + uint8_t hw; + uint8_t id_vp; + uint8_t id_fp; + + uint8_t mask; + boolean linear; +}; + struct nv50_program { struct pipe_shader_state pipe; struct tgsi_shader_info info; @@ -24,8 +33,8 @@ struct nv50_program { struct nv50_program_exec *exec_head; struct nv50_program_exec *exec_tail; unsigned exec_size; - struct nouveau_resource *data[2]; - unsigned data_start[2]; + struct nouveau_resource *data[1]; + unsigned data_start[1]; struct nouveau_bo *bo; @@ -36,14 +45,20 @@ struct nv50_program { struct { unsigned high_temp; unsigned high_result; - struct { - unsigned attr[2]; - } vp; - struct { - unsigned regs[4]; - unsigned map[5]; - unsigned high_map; - } fp; + + uint32_t attr[2]; + uint32_t regs[4]; + + /* for VPs, io_nr doesn't count 'private' results (PSIZ etc.) */ + unsigned io_nr; + struct nv50_sreg4 io[PIPE_MAX_SHADER_OUTPUTS]; + + /* FP colour inputs, VP/GP back colour outputs */ + struct nv50_sreg4 two_side[2]; + + /* VP only */ + uint8_t clpd, clpd_nr; + uint8_t psiz; } cfg; }; diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c index c7f80a2203..3b08e1b89f 100644 --- a/src/gallium/drivers/nv50/nv50_screen.c +++ b/src/gallium/drivers/nv50/nv50_screen.c @@ -87,12 +87,10 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param) return 1; case PIPE_CAP_GLSL: return 0; - case PIPE_CAP_S3TC: - return 1; case PIPE_CAP_ANISOTROPIC_FILTER: return 1; case PIPE_CAP_POINT_SPRITE: - return 0; + return 1; case PIPE_CAP_MAX_RENDER_TARGETS: return 8; case PIPE_CAP_OCCLUSION_QUERY: diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c index 4283808ed9..81fa3e34c5 100644 --- a/src/gallium/drivers/nv50/nv50_state.c +++ b/src/gallium/drivers/nv50/nv50_state.c @@ -276,6 +276,9 @@ nv50_rasterizer_state_create(struct pipe_context *pipe, so_method(so, tesla, 0x1684, 1); so_data (so, cso->flatshade_first ? 0 : 1); + so_method(so, tesla, NV50TCL_VERTEX_TWO_SIDE_ENABLE, 1); + so_data (so, cso->light_twoside); + so_method(so, tesla, NV50TCL_LINE_WIDTH, 1); so_data (so, fui(cso->line_width)); so_method(so, tesla, NV50TCL_LINE_SMOOTH_ENABLE, 1); @@ -294,6 +297,9 @@ nv50_rasterizer_state_create(struct pipe_context *pipe, so_method(so, tesla, NV50TCL_POINT_SIZE, 1); so_data (so, fui(cso->point_size)); + so_method(so, tesla, NV50TCL_POINT_SPRITE_ENABLE, 1); + so_data (so, cso->point_sprite); + so_method(so, tesla, NV50TCL_POLYGON_MODE_FRONT, 3); if (cso->front_winding == PIPE_WINDING_CCW) { so_data(so, nvgl_polygon_mode(cso->fill_ccw)); diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c index d294356f75..5a3559ed18 100644 --- a/src/gallium/drivers/nv50/nv50_state_validate.c +++ b/src/gallium/drivers/nv50/nv50_state_validate.c @@ -189,6 +189,8 @@ nv50_state_emit(struct nv50_context *nv50) so_emit(chan, nv50->state.vertprog); if (nv50->state.dirty & NV50_NEW_FRAGPROG) so_emit(chan, nv50->state.fragprog); + if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG)) + so_emit(chan, nv50->state.programs); if (nv50->state.dirty & NV50_NEW_RASTERIZER) so_emit(chan, nv50->state.rast); if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR) @@ -210,6 +212,12 @@ nv50_state_emit(struct nv50_context *nv50) so_emit(chan, nv50->state.vtxattr); } nv50->state.dirty = 0; +} + +void +nv50_state_flush_notify(struct nouveau_channel *chan) +{ + struct nv50_context *nv50 = chan->user_private; so_emit_reloc_markers(chan, nv50->state.fb); so_emit_reloc_markers(chan, nv50->state.vertprog); @@ -240,6 +248,9 @@ nv50_state_validate(struct nv50_context *nv50) if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB)) nv50_fragprog_validate(nv50); + if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG)) + nv50_linkage_validate(nv50); + if (nv50->dirty & NV50_NEW_RASTERIZER) so_ref(nv50->rasterizer->so, &nv50->state.rast); diff --git a/src/gallium/drivers/nv50/nv50_transfer.c b/src/gallium/drivers/nv50/nv50_transfer.c index e9c3562194..bb7731855c 100644 --- a/src/gallium/drivers/nv50/nv50_transfer.c +++ b/src/gallium/drivers/nv50/nv50_transfer.c @@ -89,14 +89,14 @@ nv50_transfer_rect_m2mf(struct pipe_screen *pscreen, if (src_bo->tile_flags) { BEGIN_RING(chan, m2mf, NV50_MEMORY_TO_MEMORY_FORMAT_TILING_POSITION_IN, 1); - OUT_RING (chan, (sy << 16) | sx); + OUT_RING (chan, (sy << 16) | (sx * cpp)); } else { src_offset += (line_count * src_pitch); } if (dst_bo->tile_flags) { BEGIN_RING(chan, m2mf, NV50_MEMORY_TO_MEMORY_FORMAT_TILING_POSITION_OUT, 1); - OUT_RING (chan, (dy << 16) | dx); + OUT_RING (chan, (dy << 16) | (dx * cpp)); } else { dst_offset += (line_count * dst_pitch); } diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c index 593178c50b..3b5b1bbd37 100644 --- a/src/gallium/drivers/r300/r300_screen.c +++ b/src/gallium/drivers/r300/r300_screen.c @@ -93,8 +93,6 @@ static int r300_get_param(struct pipe_screen* pscreen, int param) } else { return 0; } - case PIPE_CAP_S3TC: - return 1; case PIPE_CAP_ANISOTROPIC_FILTER: return 1; case PIPE_CAP_POINT_SPRITE: diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c index 6178c4ac7e..ce77018415 100644 --- a/src/gallium/drivers/softpipe/sp_screen.c +++ b/src/gallium/drivers/softpipe/sp_screen.c @@ -65,8 +65,6 @@ softpipe_get_param(struct pipe_screen *screen, int param) return 1; case PIPE_CAP_GLSL: return 1; - case PIPE_CAP_S3TC: - return 0; case PIPE_CAP_ANISOTROPIC_FILTER: return 0; case PIPE_CAP_POINT_SPRITE: @@ -141,6 +139,7 @@ softpipe_is_format_supported( struct pipe_screen *screen, case PIPE_FORMAT_DXT1_RGBA: case PIPE_FORMAT_DXT3_RGBA: case PIPE_FORMAT_DXT5_RGBA: + case PIPE_FORMAT_Z32_FLOAT: return FALSE; default: return TRUE; |