From b5cef6eb69f75a0bf3a80438b28c2c6c7c34f5d6 Mon Sep 17 00:00:00 2001
From: nmlgc <nmlgc@nmlgc.net>
Date: Sun, 12 Jan 2020 20:34:42 +0100
Subject: [PATCH] [Maintenance] Define VRAM access macros for arbitrary bit
 counts

Templates would have been nicer, but as soon as you add just one
non-immediate parameter, Turbo C++ generates a useless store to a new
local variable, ruining the generated code.

Part of P0069, funded by [Anonymous] and Yanga.
---
 ReC98.h                  | 25 +++++++++++++++++++++++--
 th01/grps2xsc.c          |  5 +----
 th01/hardware/egccopyr.c |  4 ++--
 th02/op_02_1.c           | 12 ++----------
 th02/op_05.c             |  4 +---
 5 files changed, 29 insertions(+), 21 deletions(-)

diff --git a/ReC98.h b/ReC98.h
index 4578d984..16b493d5 100644
--- a/ReC98.h
+++ b/ReC98.h
@@ -103,12 +103,33 @@ extern planar8_t *VRAM_PLANE_G;
 extern planar8_t *VRAM_PLANE_R;
 extern planar8_t *VRAM_PLANE_E;
 
+#define VRAM_OFFSET(x, y) ((x) >> 3) + (y << 6) + (y << 4)
+
+#define VRAM_CHUNK(plane, offset, bit_count) \
+	*(planar##bit_count##_t *)(VRAM_PLANE_##plane + offset)
+
+#define VRAM_SNAP(dst, plane, offset, bit_count) \
+	dst = VRAM_CHUNK(plane, offset, bit_count);
+
+#define VRAM_SNAP_4(dst, offset, bit_count) \
+	VRAM_SNAP(dst.B, B, offset, bit_count); \
+	VRAM_SNAP(dst.R, R, offset, bit_count); \
+	VRAM_SNAP(dst.G, G, offset, bit_count); \
+	VRAM_SNAP(dst.E, E, offset, bit_count);
+
+#define VRAM_PUT(plane, offset, src, bit_count) \
+	VRAM_CHUNK(plane, offset, bit_count) = src;
+
+#define VRAM_PUT_4(offset, src, bit_count) \
+	VRAM_PUT(B, offset, src.B, bit_count); \
+	VRAM_PUT(R, offset, src.R, bit_count); \
+	VRAM_PUT(G, offset, src.G, bit_count); \
+	VRAM_PUT(E, offset, src.E, bit_count);
+
 #define PLANE_DWORD_BLIT(dst, src) \
 	for(p = 0; p < PLANE_SIZE; p += (int)sizeof(planar32_t)) { \
 		*(planar32_t*)((dst) + p) = *(planar32_t*)((src) + p); \
 	}
 
-#define VRAM_OFFSET(x, y) ((x) >> 3) + (y << 6) + (y << 4)
-
 void pascal vram_planes_set(void);
 // -----------------
diff --git a/th01/grps2xsc.c b/th01/grps2xsc.c
index a2b344b5..f8977c7a 100644
--- a/th01/grps2xsc.c
+++ b/th01/grps2xsc.c
@@ -32,10 +32,7 @@ void graph_slow_2xscale_rect_1_to_0(int x0, int y0, int x1, int y1, int w1, int
 		for(col16 = 0; col16 < w1 / 16; col16++) {
 			int scale_p;
 			graph_accesspage_func(1);
-			px16.B = *(planar16_t*)(VRAM_PLANE_B + p1);
-			px16.R = *(planar16_t*)(VRAM_PLANE_R + p1);
-			px16.G = *(planar16_t*)(VRAM_PLANE_G + p1);
-			px16.E = *(planar16_t*)(VRAM_PLANE_E + p1);
+			VRAM_SNAP_4(px16, p1, 16);
 			px16_nonzero = px16.B | px16.R | px16.G | px16.E;
 			for(scale_p = 0; scale_p < ROW_SIZE * 2; scale_p += ROW_SIZE) {
 				planar32_t dst32;
diff --git a/th01/hardware/egccopyr.c b/th01/hardware/egccopyr.c
index 84bc7922..129dad8c 100644
--- a/th01/hardware/egccopyr.c
+++ b/th01/hardware/egccopyr.c
@@ -33,8 +33,8 @@ void egc_copy_rect_1_to_0(int x, int y, int w, int h)
 	egc_start_copy();
 	for(row = 0; row < h; row++) {
 		for(col = x_floor, p = row_p; col < x_end; p += 2, col += 16) {
-			graph_accesspage_func(1); bits = *(planar16_t*)(VRAM_PLANE_B + p);
-			graph_accesspage_func(0); *(planar16_t*)(VRAM_PLANE_B + p) = bits;
+			graph_accesspage_func(1); VRAM_SNAP(bits, B, p, 16);
+			graph_accesspage_func(0); VRAM_PUT(B, p, bits, 16);
 		}
 		row_p += 640 / 8;
 	}
diff --git a/th02/op_02_1.c b/th02/op_02_1.c
index c0a91d6b..a5116690 100644
--- a/th02/op_02_1.c
+++ b/th02/op_02_1.c
@@ -33,16 +33,8 @@ void pascal graph_copy_rect_1_to_0(int x, int y, int w, int h)
 		for(col = 0, p = row_p; col < (w >> 4); col++, p += 2) {
 			if(row_p >= 0) {
 				vram_planar_16_pixels_t p16;
-				graph_accesspage(1);
-				p16.B = *(planar16_t*)(VRAM_PLANE_B + p);
-				p16.R = *(planar16_t*)(VRAM_PLANE_R + p);
-				p16.G = *(planar16_t*)(VRAM_PLANE_G + p);
-				p16.E = *(planar16_t*)(VRAM_PLANE_E + p);
-				graph_accesspage(0);
-				*(planar16_t*)(VRAM_PLANE_B + p) = p16.B;
-				*(planar16_t*)(VRAM_PLANE_R + p) = p16.R;
-				*(planar16_t*)(VRAM_PLANE_G + p) = p16.G;
-				*(planar16_t*)(VRAM_PLANE_E + p) = p16.E;
+				graph_accesspage(1);	VRAM_SNAP_4(p16, p, 16);
+				graph_accesspage(0);	VRAM_PUT_4(p, p16, 16);
 			}
 		}
 		row_p += ROW_SIZE;
diff --git a/th02/op_05.c b/th02/op_05.c
index f14a147e..2814fe0c 100644
--- a/th02/op_05.c
+++ b/th02/op_05.c
@@ -69,9 +69,7 @@ void darken_pic_at(int x, int y)
 	grcg_setcolor(GC_RMW, 0);
 	for(row = 0; row < 144; row++, row_p += 640 / 8)  {
 		for(col = 0; col < 192 / 8; col += 2) {
-			*(planar16_t*)(VRAM_PLANE_B + row_p + col) =
-				row & 1 ? 0xAAAA : 0x5555
-			;
+			VRAM_PUT(B, row_p + col, (row & 1 ? 0xAAAA : 0x5555), 16);
 		}
 	}
 	grcg_off();